Merged
Conversation
Co-authored-by: Codex <noreply@openai.com>
Co-authored-by: Codex <noreply@openai.com>
Co-authored-by: Codex <noreply@openai.com>
Co-authored-by: Codex <noreply@openai.com>
|
Caution Review failedPull request was closed or merged during review 📝 WalkthroughWalkthroughThe changes refactor authentication token storage to embed a shared Changes
Estimated code review effort🎯 3 (Moderate) | ⏱️ ~20 minutes Possibly related PRs
Poem
🚥 Pre-merge checks | ✅ 1 | ❌ 2❌ Failed checks (1 warning, 1 inconclusive)
✅ Passed checks (1 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. ✨ Finishing Touches
🧪 Generate unit tests (beta)
📝 Coding Plan
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
Summary of ChangesHello, I'm Gemini Code Assist1! I'm currently reviewing this pull request and will post my feedback shortly. In the meantime, here's a summary to help you and other reviewers quickly get up to speed! This pull request focuses on enhancing the stability, compatibility, and maintainability of the LLM proxy. It introduces a significant refactoring of authentication token storage to use a common base, simplifies the Kiro executor by removing redundant authentication logic, and improves the handling of "thinking" configurations for iFlow models. Additionally, it refines OpenAI endpoint compatibility and restores critical functionalities within the authentication manager, ensuring more robust and consistent operation across different LLM providers and features. Highlights
Changelog
Activity
Using Gemini Code AssistThe full guide for Gemini Code Assist can be found on our documentation page, here are some quick tips. Invoking Gemini You can request assistance from Gemini at any point by creating a comment using either
Customization To customize Gemini Code Assist for GitHub experience, repository maintainers can create a configuration file and/or provide a custom code review style guide (such as PEP-8 for Python) by creating and adding files to a Limitations & Feedback Gemini Code Assist may make mistakes. Please leave feedback on any instances where its feedback is incorrect or counter productive. You can react with 👍 and 👎 on @gemini-code-assist comments. If you're interested in giving your feedback about your experience with Gemini Code Assist for GitHub and other Google products, sign up here. You can also get AI-powered code generation, chat, as well as code reviews directly in the IDE at no cost with the Gemini Code Assist IDE Extension. Footnotes
|
coderabbitai
Bot
added
HELIOS-CODEX
![gemini-code-assist[bot]](https://avatars.githubusercontent.com/in/956858?s=60&v=4){kind=small}
There was a problem hiding this comment.
Code Review
This pull request introduces a series of stability fixes and refactorings across the authentication and execution paths. Key changes include refactoring various token storage implementations to use a common base struct, which improves maintainability. Additionally, there are fixes to stabilize the 'thinking' feature for the iFlow provider and to improve OpenAI endpoint compatibility. A significant part of this PR appears to be the restoration of the auth manager and executor compatibility, which should enhance overall system stability. My review includes one suggestion for removing redundant code in the Qwen token handling.
Comment on lines
55
to
57
| if _, err := cleanTokenFilePath(authFilePath, "qwen token"); err != nil { | ||
| return err | ||
| } |
There was a problem hiding this comment.
This path validation check is redundant. The base.BaseTokenStorage.Save method, which is called on line 60, already performs the same path validation. Removing this block and the local cleanTokenFilePath function will help avoid code duplication and rely on the centralized validation in the base class.
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 16
Caution
Some comments are outside the diff and can’t be posted inline due to platform limitations.
⚠️ Outside diff range comments (3)
pkg/llmproxy/auth/claude/utls_transport.go (1)
67-82:⚠️ Potential issue | 🔴 CriticalRace condition: multiple goroutines can create duplicate connections.
When
Broadcast()wakes multiple waiting goroutines, they sequentially acquire the lock. The first waiter finds no connection, falls through, and creates a new pending entry. However, subsequent waiters also fall through (the original pending was deleted before Broadcast) without checking if another waiter has already started creating a connection—they create their own pending entry and connection.This can lead to resource leaks (duplicate connections) and wasted resources.
🐛 Proposed fix: Use a loop to re-check after waking
func (t *utlsRoundTripper) getOrCreateConnection(host, addr string) (*http2.ClientConn, error) { t.mu.Lock() - // Check if connection exists and is usable - if h2Conn, ok := t.connections[host]; ok && h2Conn.CanTakeNewRequest() { - t.mu.Unlock() - return h2Conn, nil - } - - // Check if another goroutine is already creating a connection - if cond, ok := t.pending[host]; ok { - // Wait for the other goroutine to finish - cond.Wait() - // Check if connection is now available + for { + // Check if connection exists and is usable if h2Conn, ok := t.connections[host]; ok && h2Conn.CanTakeNewRequest() { t.mu.Unlock() return h2Conn, nil } - // Connection still not available, we'll create one + + // Check if another goroutine is already creating a connection + if cond, ok := t.pending[host]; ok { + // Wait for the other goroutine to finish, then re-check + cond.Wait() + continue + } + + // No one is creating a connection, we'll do it + break } // Mark this host as pending🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@pkg/llmproxy/auth/claude/utls_transport.go` around lines 67 - 82, The current wait logic can let multiple waiters proceed to create duplicate connections; change the single Wait() to a loop that re-checks both t.connections[host] and t.pending[host] after each wake: acquire the lock, and while there is an existing pending entry for host, call cond.Wait(); after waking, if a usable connection exists return it; if no pending entry exists anymore then this goroutine should create the new sync.Cond, assign t.pending[host]=cond and proceed as the sole creator; ensure all checks reference t.pending, t.connections, cond.Wait(), and CanTakeNewRequest() so only one goroutine becomes the creator.sdk/api/handlers/openai/openai_responses_handlers.go (1)
91-98:⚠️ Potential issue | 🟠 MajorThe Minimax non-stream fallback never takes effect here.
This bridge still honors
streamunconditionally.shouldForceNonStreamingChatBridgewas added for this compatibility path, but this branch never applies it, so Minimax requests will keep going through the streaming chat bridge instead of the intended non-stream fallback.🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@sdk/api/handlers/openai/openai_responses_handlers.go` around lines 91 - 98, The branch that converts Responses->Chat currently respects the incoming "stream" flag unconditionally (in the block using resolveEndpointOverride and responsesconverter.ConvertOpenAIResponsesRequestToOpenAIChatCompletions), so the Minimax non-stream fallback never triggers; update this block to consult shouldForceNonStreamingChatBridge(modelName, rawJSON) after creating chatJSON and override the stream value to false when that function returns true before calling h.handleStreamingResponseViaChat or h.handleNonStreamingResponseViaChat so Minimax requests are routed to the non-streaming handler; reference resolveEndpointOverride, openAIResponsesEndpoint, OpenaiResponse, openAIChatEndpoint, responsesconverter.ConvertOpenAIResponsesRequestToOpenAIChatCompletions, shouldForceNonStreamingChatBridge, h.handleStreamingResponseViaChat and h.handleNonStreamingResponseViaChat.sdk/api/handlers/openai/openai_handlers.go (1)
117-130: 🛠️ Refactor suggestion | 🟠 MajorExtract this compat branch out of
ChatCompletions.This routing path pushes
ChatCompletionsfurther past the 40-line limit and makes the endpoint/format dispatch harder to follow. Pull the override-and-conversion flow into a helper before adding more cases.As per coding guidelines, "Maximum function length: 40 lines".
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@sdk/api/handlers/openai/openai_handlers.go` around lines 117 - 130, Extract the compat branch inside ChatCompletions into a new helper method (e.g., handleResponsesCompat or resolveAndHandleResponsesOverride) that accepts the same inputs used there (modelName, rawJSON, stream flag, and the context/caller receiver h and c) and returns a boolean indicating “handled” (true means caller should return). Move the resolveEndpointOverride check, originalChat assignment, shouldTreatAsResponsesFormat check, codexconverter.ConvertOpenAIRequestToCodex call, re-evaluation of stream via gjson.GetBytes(rawJSON, "stream").Bool(), and the conditional calls to h.handleStreamingResponseViaResponses / h.handleNonStreamingResponseViaResponses into that helper. Replace the original inlined block with a single if helper(...) { return } so ChatCompletions stays under 40 lines and routing is clearer.
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Inline comments:
In `@pkg/llmproxy/auth/claude/utls_transport.go`:
- Line 13: Current code imports logrus as log (import alias: log
"github.com/sirupsen/logrus"); follow-up refactor should replace that import
with zerolog (e.g., github.com/rs/zerolog) and update all usages of the log
alias (calls like log.Info, log.Errorf, log.Debug, etc.) to zerolog equivalents
(logger.Info().Msg, logger.Error().Err(err).Msg, etc.), initializing a zerolog
logger instance where needed and ensuring structured fields are migrated to
With/Dict patterns; search for the import line and all references to the log
identifier in utls_transport.go (and related files) to perform the replacement
consistently.
In `@pkg/llmproxy/thinking/provider/iflow/apply_test.go`:
- Around line 54-93: Update the
TestApplyMiniMaxStripsReasoningVariantsAndLegacyFields test to include a
regression case where the JSON body contains a top-level "reasoning.effort" key
(e.g. "reasoning.effort":"medium") alongside the existing fields; call
a.Apply(body, cfg, model) as before and add an assertion that
res.Get("reasoning.effort").Exists() is false (or equivalent check) to ensure
Apply removes the literal reasoning.effort key; keep the rest of the assertions
unchanged so unrelated fields (e.g. "foo") are still preserved.
In `@pkg/llmproxy/thinking/provider/iflow/apply.go`:
- Around line 146-148: The cleanup in applyMiniMax misses the literal dotted key
"reasoning.effort" so add an sjson.DeleteBytes call to remove the escaped path
"reasoning\\.effort" alongside the existing deletes; locate the applyMiniMax
function in pkg/llmproxy/thinking/provider/iflow/apply.go and insert a deletion
using sjson.DeleteBytes(result, "reasoning\\.effort") (follow existing error
handling pattern) so that keys accepted by extractIFlowConfig (which reads
"reasoning\\.effort") are not leaked upstream.
In `@pkg/llmproxy/thinking/strip.go`:
- Around line 43-50: The iflow stripping list in pkg/llmproxy/thinking/strip.go
currently removes nested "reasoning" but not a literal top-level key
"reasoning.effort"; update the paths slice (the variable assigned to paths in
strip.go) to include the escaped literal key "reasoning\.effort" so top-level
dot-keys are stripped as well, and consider moving this paths list into a shared
constant used by the iflow cleanup code in provider/iflow/apply.go to avoid
drift.
In `@sdk/api/handlers/openai/endpoint_compat_test.go`:
- Around line 62-81: Register a second provider for the same modelID that only
supports "/responses" (e.g., call registry.GetGlobalRegistry().RegisterClient
with a different client name and provider id like "bridge" and ModelInfo
SupportedEndpoints []string{"/responses"}), add a t.Cleanup to UnregisterClient
that second provider, and then update the assertions around
resolveEndpointOverride(modelID, "/responses", OpenaiResponse) to assert that ok
is true and overrideEndpoint == "/responses" (i.e., the requested /responses
remains unmodified) while keeping the original iflow registration in place so
the mixed-provider case is exercised.
- Around line 45-50: The cleanup is unregistering a different client ID than the
one registered, leaking the registry entry; update the UnregisterClient call so
it uses the exact same client ID string used in RegisterClient
("endpoint-compat-iflow-direction")—i.e., locate the register call via
registry.GetGlobalRegistry().RegisterClient(...) and make the corresponding
registry.GetGlobalRegistry().UnregisterClient(...) use that identical ID.
In `@sdk/api/handlers/openai/endpoint_compat.go`:
- Around line 19-26: The loop treating
resolveEndpointOverrideForInfo(reg.GetModelInfo(...), requestedEndpoint) the
same for both "no override" and "provider already supports endpoint" causes
native-support cases to be overwritten by later providers; update the logic so
that when iterating providers from reg.GetModelProviders(modelName) you stop and
return immediately if the provider already supports the requested endpoint
(e.g., check modelInfo.SupportsEndpoint(requestedEndpoint) on
reg.GetModelInfo(modelName, provider) or change resolveEndpointOverrideForInfo
to return an explicit tri-state like (override, ok, supported)); only continue
scanning when there is no native support and an override is not present, and
only accept an override when resolveEndpointOverrideForInfo indicates an actual
override (override + ok).
In `@sdk/cliproxy/auth/cooldown_compat.go`:
- Around line 3-5: Add a Go doc comment for the exported function
SetQuotaCooldownDisabled describing its purpose and usage; place a brief
sentence or two immediately above the SetQuotaCooldownDisabled function stating
that it enables/disables quota cooldown behavior and when callers should use it,
and reference the underlying quotaCooldownDisabled store variable in the comment
for clarity.
In `@sdk/cliproxy/auth/manager_compat.go`:
- Around line 16-24: Create shared error message constants (e.g.,
ErrMsgAuthRequired, ErrMsgAuthIDRequired, ErrMsgProviderRequired) in the auth
package (or a nearby shared file) and replace the inline string literals in
manager_compat.go (locations around the auth validation branches and the other
occurrences you noted at lines ~54-60, ~439, ~476) with those constants; update
the Error{Message: ...} initializers to use the new constants (or fmt.Sprintf
with template constants if interpolation is needed) so all
register/update/select paths reference the same template values.
- Around line 39-46: The code updates the in-memory map m.auths (via
m.mu.Lock()/m.auths[next.ID] = next.Clone()) before calling m.store.Save(ctx,
next.Clone()), which can leave runtime state ahead of durable state on save
failure or races; change the flow in the methods touching m.auths and
m.store.Save (the two mirrored blocks around m.auths/next.Clone and m.store.Save
guarded by shouldSkipPersist) so you attempt the persistent Save first (call
m.store.Save with the clone), and only if Save succeeds acquire m.mu, assign
m.auths[next.ID] = next.Clone() and then unlock; alternatively hold m.mu for the
duration of Save to serialize races—ensure you use next.Clone for both Save and
the in-memory assignment and return the original error if Save fails.
- Around line 148-163: StartAutoRefresh ignores the interval and does nothing
besides waiting for ctx.Done(); change it to use the provided interval parameter
(remove the blank identifier) and spawn a goroutine that ticks at that interval
to run the Manager's refresh logic until canceled. Specifically: accept the
interval argument in StartAutoRefresh, call StopAutoRefresh() as you already do,
create refreshCtx and cancel, store cancel in m.refreshCancel, then start a
goroutine that creates a time.Ticker from the interval (handle zero/negative by
defaulting or returning) and loops with select { case <-ticker.C: invoke the
manager's refresh routine (e.g., m.Refresh() or the method that renews auths);
case <-refreshCtx.Done(): ticker.Stop(); return }. Retain the existing names
Manager, StartAutoRefresh, StopAutoRefresh, m.refreshCancel, refreshCtx, cancel
to locate the change.
- Around line 452-487: The selection mixes auths from all providers into
candidates but picks an executor only for the first provider with an executor
and always calls selector.Pick with providers[0], causing provider mismatch;
instead, for each provider iterate building provider-scoped candidates from
m.auths (using strings.EqualFold(auth.Provider, provider) and pinnedAuthID
filtering), get the corresponding exec from m.executors, and if exec != nil and
that provider has candidates call selector.Pick(ctx, provider, model, opts,
providerCandidates); if selector.Pick returns an auth, call
updateAggregatedAvailability and selectedAuthCallback(auth.ID) and immediately
return that auth and the matching exec; only if no provider yields a pick return
the auth_not_found error—this ensures executor and auth come from the same
provider (referencing candidates, executor, m.executors, providers,
selector.Pick, m.auths, pinnedAuthID, updateAggregatedAvailability,
selectedAuthCallback).
- Around line 428-434: The ExecuteStream path currently returns
executor.ExecuteStream errors without updating quota/backoff state; modify
Manager.ExecuteStream so after selecting auth/executor (selectAuthAndExecutor)
and applying applyAPIKeyModelAlias, you call executor.ExecuteStream and if it
returns an error immediately call m.recordExecutionResult(...) with the same
ctx, auth, req.Model (or the aliased model) and an appropriate failure result to
update quota/backoff state before returning the error; keep the existing success
return path unchanged so only startup failures are recorded.
- Around line 543-574: The switch on
strings.ToLower(strings.TrimSpace(auth.Provider)) duplicates the same
API-key/model lookup for cfg.GeminiKey, cfg.ClaudeKey and cfg.CodexKey; extract
a provider registry/helper and replace the three branches with a single generic
lookup. Implement a small provider registry (map[string]func(*Config)[]KeyType
or an interface) that maps "gemini"/"claude"/"codex" to functions returning the
corresponding cfg.<GeminiKey|ClaudeKey|CodexKey> slice, then in the current
function call that helper once: iterate the returned keys, compare
strings.TrimSpace(key.APIKey) and BaseURL, build models []modelAliasEntry and
call resolveModelAliasFromConfigModels(requestedModel, models). Update names
referenced here (auth.Provider, cfg.GeminiKey, cfg.ClaudeKey, cfg.CodexKey,
resolveModelAliasFromConfigModels, modelAliasEntry) and ensure the refactor
keeps the logic identical and reduces the function length below 40 lines.
- Around line 186-193: The manager currently calls CloseExecutionSession while
holding m.mu which can deadlock; modify Set/replace logic so you capture the
replaced executor (m.executors[key]) while holding m.mu, then release the lock
and call CloseExecutionSession(CloseAllExecutionSessionsID) on it outside the
lock; specifically, in the function that assigns m.executors[key] (the block
that checks for replaced, ok := m.executors[key] and type-asserts to
ExecutionSessionCloser), store the replaced value in a local variable while
locked, perform m.executors[key] = executor, unlock (defer removal of defer if
needed), and only after the mutex is released, if the local variable implements
ExecutionSessionCloser call
closer.CloseExecutionSession(CloseAllExecutionSessionsID).
In `@sdk/cliproxy/auth/oauth_model_alias.go`:
- Around line 60-65: The method rebuildAPIKeyModelAliasFromRuntimeConfig
currently only clears m.apiKeyModelAlias but its name implies it reads runtime
config and repopulates entries; either rename it to clearAPIKeyModelAliasCache
(leave body as m.apiKeyModelAlias.Store(apiKeyModelAliasTable(nil))) or
implement the intended rebuild: fetch the runtime config from Manager (e.g.
m.runtimeConfig or a getter like m.GetRuntimeConfig()), build the alias table
via apiKeyModelAliasTable(cfg) and then call m.apiKeyModelAlias.Store(...) to
replace the cache; update callers to match the new name or behavior and keep
references to Manager, rebuildAPIKeyModelAliasFromRuntimeConfig,
apiKeyModelAlias, and apiKeyModelAliasTable to locate the change.
---
Outside diff comments:
In `@pkg/llmproxy/auth/claude/utls_transport.go`:
- Around line 67-82: The current wait logic can let multiple waiters proceed to
create duplicate connections; change the single Wait() to a loop that re-checks
both t.connections[host] and t.pending[host] after each wake: acquire the lock,
and while there is an existing pending entry for host, call cond.Wait(); after
waking, if a usable connection exists return it; if no pending entry exists
anymore then this goroutine should create the new sync.Cond, assign
t.pending[host]=cond and proceed as the sole creator; ensure all checks
reference t.pending, t.connections, cond.Wait(), and CanTakeNewRequest() so only
one goroutine becomes the creator.
In `@sdk/api/handlers/openai/openai_handlers.go`:
- Around line 117-130: Extract the compat branch inside ChatCompletions into a
new helper method (e.g., handleResponsesCompat or
resolveAndHandleResponsesOverride) that accepts the same inputs used there
(modelName, rawJSON, stream flag, and the context/caller receiver h and c) and
returns a boolean indicating “handled” (true means caller should return). Move
the resolveEndpointOverride check, originalChat assignment,
shouldTreatAsResponsesFormat check, codexconverter.ConvertOpenAIRequestToCodex
call, re-evaluation of stream via gjson.GetBytes(rawJSON, "stream").Bool(), and
the conditional calls to h.handleStreamingResponseViaResponses /
h.handleNonStreamingResponseViaResponses into that helper. Replace the original
inlined block with a single if helper(...) { return } so ChatCompletions stays
under 40 lines and routing is clearer.
In `@sdk/api/handlers/openai/openai_responses_handlers.go`:
- Around line 91-98: The branch that converts Responses->Chat currently respects
the incoming "stream" flag unconditionally (in the block using
resolveEndpointOverride and
responsesconverter.ConvertOpenAIResponsesRequestToOpenAIChatCompletions), so the
Minimax non-stream fallback never triggers; update this block to consult
shouldForceNonStreamingChatBridge(modelName, rawJSON) after creating chatJSON
and override the stream value to false when that function returns true before
calling h.handleStreamingResponseViaChat or h.handleNonStreamingResponseViaChat
so Minimax requests are routed to the non-streaming handler; reference
resolveEndpointOverride, openAIResponsesEndpoint, OpenaiResponse,
openAIChatEndpoint,
responsesconverter.ConvertOpenAIResponsesRequestToOpenAIChatCompletions,
shouldForceNonStreamingChatBridge, h.handleStreamingResponseViaChat and
h.handleNonStreamingResponseViaChat.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: Organization UI
Review profile: ASSERTIVE
Plan: Pro
Run ID: ceff7acf-7eb6-4d80-a16e-d01eb1d31360
📒 Files selected for processing (39)
pkg/llmproxy/api/handlers/management/auth_gemini.gopkg/llmproxy/api/handlers/management/auth_github.gopkg/llmproxy/api/handlers/management/auth_kilo.gopkg/llmproxy/auth/claude/anthropic_auth.gopkg/llmproxy/auth/claude/utls_transport.gopkg/llmproxy/auth/codex/openai_auth.gopkg/llmproxy/auth/codex/openai_auth_test.gopkg/llmproxy/auth/codex/token_test.gopkg/llmproxy/auth/copilot/copilot_auth.gopkg/llmproxy/auth/copilot/copilot_extra_test.gopkg/llmproxy/auth/copilot/token_test.gopkg/llmproxy/auth/gemini/gemini_auth.gopkg/llmproxy/auth/gemini/gemini_auth_test.gopkg/llmproxy/auth/iflow/iflow_auth.gopkg/llmproxy/auth/kimi/kimi.gopkg/llmproxy/auth/kimi/token_path_test.gopkg/llmproxy/auth/qwen/qwen_auth.gopkg/llmproxy/auth/qwen/qwen_auth_test.gopkg/llmproxy/auth/qwen/qwen_token.gopkg/llmproxy/auth/qwen/qwen_token_test.gopkg/llmproxy/executor/kiro_auth.gopkg/llmproxy/executor/kiro_executor.gopkg/llmproxy/executor/kiro_streaming.gopkg/llmproxy/executor/kiro_transform.gopkg/llmproxy/thinking/apply.gopkg/llmproxy/thinking/apply_iflow_test.gopkg/llmproxy/thinking/provider/iflow/apply.gopkg/llmproxy/thinking/provider/iflow/apply_test.gopkg/llmproxy/thinking/strip.gopkg/llmproxy/thinking/strip_test.gopkg/llmproxy/usage/metrics.gosdk/api/handlers/openai/endpoint_compat.gosdk/api/handlers/openai/endpoint_compat_test.gosdk/api/handlers/openai/openai_handlers.gosdk/api/handlers/openai/openai_responses_handlers.gosdk/auth/kilo.gosdk/cliproxy/auth/cooldown_compat.gosdk/cliproxy/auth/manager_compat.gosdk/cliproxy/auth/oauth_model_alias.go
💤 Files with no reviewable changes (2)
- pkg/llmproxy/executor/kiro_executor.go
- pkg/llmproxy/executor/kiro_auth.go
📜 Review details
🧰 Additional context used
📓 Path-based instructions (1)
**/*.go
📄 CodeRabbit inference engine (AGENTS.md)
**/*.go: NEVER create a v2 file - refactor the original instead
NEVER create a new class if an existing one can be made generic
NEVER create custom implementations when an OSS library exists - search pkg.go.dev for existing libraries before writing code
Build generic building blocks (provider interface + registry) before application logic
Use chi for HTTP routing (NOT custom routers)
Use zerolog for logging (NOT fmt.Print)
Use viper for configuration (NOT manual env parsing)
Use go-playground/validator for validation (NOT manual if/else validation)
Use golang.org/x/time/rate for rate limiting (NOT custom limiters)
Use template strings for messages instead of hardcoded messages and config-driven logic instead of code-driven
Zero new lint suppressions without inline justification
All new code must pass: go fmt, go vet, golint
Maximum function length: 40 lines
No placeholder TODOs in committed code
Files:
pkg/llmproxy/auth/gemini/gemini_auth_test.gopkg/llmproxy/auth/codex/openai_auth.gopkg/llmproxy/auth/iflow/iflow_auth.gopkg/llmproxy/auth/claude/anthropic_auth.gopkg/llmproxy/auth/claude/utls_transport.gopkg/llmproxy/auth/copilot/copilot_auth.gopkg/llmproxy/thinking/provider/iflow/apply.gosdk/cliproxy/auth/cooldown_compat.gopkg/llmproxy/auth/gemini/gemini_auth.gopkg/llmproxy/auth/codex/openai_auth_test.gosdk/api/handlers/openai/openai_responses_handlers.gopkg/llmproxy/thinking/strip_test.gopkg/llmproxy/thinking/provider/iflow/apply_test.gosdk/cliproxy/auth/oauth_model_alias.gosdk/api/handlers/openai/endpoint_compat_test.gopkg/llmproxy/thinking/apply_iflow_test.gopkg/llmproxy/auth/copilot/copilot_extra_test.gopkg/llmproxy/auth/kimi/kimi.gopkg/llmproxy/executor/kiro_streaming.gosdk/auth/kilo.gosdk/api/handlers/openai/openai_handlers.gopkg/llmproxy/api/handlers/management/auth_kilo.gopkg/llmproxy/usage/metrics.gopkg/llmproxy/api/handlers/management/auth_gemini.gopkg/llmproxy/executor/kiro_transform.gosdk/api/handlers/openai/endpoint_compat.gopkg/llmproxy/auth/qwen/qwen_auth_test.gopkg/llmproxy/auth/codex/token_test.gopkg/llmproxy/api/handlers/management/auth_github.gopkg/llmproxy/thinking/strip.gopkg/llmproxy/auth/qwen/qwen_token_test.gopkg/llmproxy/auth/copilot/token_test.gopkg/llmproxy/auth/qwen/qwen_token.gopkg/llmproxy/auth/qwen/qwen_auth.gopkg/llmproxy/auth/kimi/token_path_test.gopkg/llmproxy/thinking/apply.gosdk/cliproxy/auth/manager_compat.go
🧠 Learnings (1)
📚 Learning: 2026-02-25T10:11:41.448Z
Learnt from: CR
Repo: KooshaPari/cliproxyapi-plusplus PR: 0
File: AGENTS.md:0-0
Timestamp: 2026-02-25T10:11:41.448Z
Learning: Applies to **/*.go : Use template strings for messages instead of hardcoded messages and config-driven logic instead of code-driven
Applied to files:
pkg/llmproxy/thinking/strip.go
🔇 Additional comments (38)
pkg/llmproxy/usage/metrics.go (1)
7-7: Import path migration looks correct.Line 7 now matches the module declared in
go.modand the util package path, which avoids mixed-prefix import/build issues.pkg/llmproxy/executor/kiro_transform.go (1)
13-13: Import alias rename is clean and non-breaking.Line 13 updates the alias to
cliproxyauthand usage remains consistent in this file.pkg/llmproxy/executor/kiro_streaming.go (1)
22-23: Alias updates are consistent and safe.Lines 22-23 align import aliases with current naming (
cliproxyauth,cliproxyexecutor) and all in-file references match.pkg/llmproxy/auth/claude/anthropic_auth.go (1)
16-16: Import path migration looks correct.This update is consistent with relocating config usage to
pkg/llmproxy/configand introduces no functional risk in this segment.sdk/cliproxy/auth/cooldown_compat.go (1)
7-9: LGTM!Correct use of
atomic.Bool.Load()for thread-safe read access. The unexported getter properly encapsulates access to the package-level variable and aligns with its usage inmanager_compat.go.pkg/llmproxy/auth/iflow/iflow_auth.go (1)
16-16: Import migration verified as complete and consistent across all auth modules.The import path update from
internal/configtopkg/llmproxy/configis correct and has been successfully applied across all auth providers (copilot, codex, synthesizer, qwen, kiro, kimi, claude, diff, gemini, antigravity). No remaining references tointernal/configexist in auth modules.pkg/llmproxy/auth/claude/utls_transport.go (1)
11-11: LGTM!The import path update from
internal/configtopkg/llmproxy/configis consistent with the broader refactoring across auth modules described in the PR objectives.pkg/llmproxy/auth/kimi/kimi.go (1)
18-18: LGTM!The import path change from
internal/configtopkg/llmproxy/configis consistent with the broader refactoring to expose configuration as a public package.pkg/llmproxy/auth/codex/openai_auth.go (1)
17-17: LGTM!The import path update aligns with the consistent migration from
internal/configtopkg/llmproxy/configacross the codebase.pkg/llmproxy/auth/codex/openai_auth_test.go (2)
12-12: LGTM!Import correctly added to support the embedded
BaseTokenStoragestructure in test initializations.
300-300: LGTM!Test initialization correctly updated to use the embedded
BaseTokenStoragestructure, maintaining compatibility with field promotion for assertions onstorage.AccessToken.pkg/llmproxy/api/handlers/management/auth_github.go (2)
10-10: LGTM!Import for
basepackage correctly added to support the embeddedBaseTokenStoragepattern.
55-58: LGTM!Token storage initialization correctly migrated to use the embedded
BaseTokenStoragestructure withAccessTokenandTypefields properly nested.pkg/llmproxy/auth/copilot/copilot_extra_test.go (3)
13-13: LGTM!Import correctly added to support
BaseTokenStorageusage in test initializations.
179-181: LGTM!Test initialization correctly updated to construct
CopilotTokenStoragewith embeddedBaseTokenStorage.
187-189: LGTM!Consistent with the valid token test case, correctly using embedded
BaseTokenStoragefor the expired token scenario.pkg/llmproxy/auth/copilot/copilot_auth.go (1)
13-13: LGTM!Import path correctly updated to
pkg/llmproxy/config, consistent with the codebase-wide migration. TheSDKConfigfield access viacfg.SDKConfigremains valid per the config struct definition.pkg/llmproxy/auth/kimi/token_path_test.go (3)
6-7: LGTM!Import correctly added to support
BaseTokenStorageusage.
11-11: LGTM!Test initialization correctly updated to use embedded
BaseTokenStoragestructure.
18-18: No action needed. The test correctly validates the error message. The underlyingSaveTokenToFiledelegates toBaseTokenStorage.Save(), which generates the error message"base token storage: invalid file path: %w". Since the test usesstrings.Contains()to check for"invalid file path", it will successfully match this error message. The implementation and test are aligned—there is no mismatch or regression.> Likely an incorrect or invalid review comment.pkg/llmproxy/auth/codex/token_test.go (2)
8-9: LGTM!Import correctly added to support
BaseTokenStorageusage in test initializations.
21-29: LGTM!Test initialization correctly updated to nest
AccessToken,RefreshToken, andBaseTokenStorage, while keepingIDTokenandAccountIDat the top level. The test assertions on lines 46 and 49 will work correctly due to Go's embedded struct field promotion.pkg/llmproxy/auth/copilot/token_test.go (1)
9-9: Base token embedding update in tests looks correct.The fixture now matches the migrated token storage shape and should keep serialization coverage aligned with production structs.
Also applies to: 22-25
pkg/llmproxy/auth/gemini/gemini_auth.go (1)
17-17: Config import path update is consistent and safe.This change cleanly aligns Gemini auth with the new
pkg/llmproxy/configpackage location.pkg/llmproxy/api/handlers/management/auth_kilo.go (1)
10-10: Kilo token storage migration is applied correctly.Using
BaseTokenStoragehere keeps the handler aligned with the shared auth storage model while preserving current behavior.Also applies to: 60-63
pkg/llmproxy/auth/qwen/qwen_auth_test.go (1)
13-13: Qwen auth test fixture update is consistent with BaseTokenStorage embedding.The changed setup keeps the traversal-path test aligned with the new storage structure.
Also applies to: 157-157
sdk/auth/kilo.go (1)
8-8: SDK Kilo storage initialization is compatible with the embedded base struct.This update correctly preserves
TypethroughBaseTokenStorageduring auth record construction.Also applies to: 101-104
pkg/llmproxy/auth/qwen/qwen_token_test.go (1)
8-8: Qwen token tests are correctly updated for BaseTokenStorage embedding.Both test cases now initialize token data using the shared base storage shape.
Also applies to: 17-20, 35-37
pkg/llmproxy/auth/gemini/gemini_auth_test.go (1)
15-15: Gemini auth tests are aligned with the storage refactor and updated error wording.The changed fixture shape and traversal-path assertion remain coherent with the new token storage contract.
Also applies to: 50-53, 82-82
pkg/llmproxy/api/handlers/management/auth_gemini.go (1)
14-14: Gemini management handler uses the new base token storage pattern correctly.This keeps handler-generated auth records compatible with the shared storage model.
Also applies to: 142-144
pkg/llmproxy/auth/qwen/qwen_auth.go (2)
16-16: LGTM!Import correctly added for the new base package containing
BaseTokenStorage.
351-363: LGTM!The refactored
CreateTokenStoragecorrectly initializes the embeddedBaseTokenStoragewith token data. This approach:
- Properly sets
AccessTokenandRefreshTokenwithin the embedded struct- Maintains the Qwen-specific fields (
LastRefresh,ResourceURL,Expire) at the outer struct level- Aligns with the pattern used across other provider modules in this PR
pkg/llmproxy/auth/qwen/qwen_token.go (3)
19-30: LGTM!The struct refactoring is well-designed:
- Anonymous embedding of
base.BaseTokenStoragepromotes fields to the JSON top level, maintaining backward compatibility with existing token files- Value embedding (not pointer) ensures the base storage is always initialized, avoiding nil pointer issues
- Qwen-specific fields remain cleanly separated at the outer struct level
38-42: LGTM!Constructor correctly initializes only the
FilePathfield, which is appropriate since other fields will be populated during authentication or when loading from file.
53-61: LGTM!The
SaveTokenToFilemethod correctly:
- Sets the
Typefield on the embeddedBaseTokenStoragebefore saving- Passes the full
QwenTokenStorage(ts) as the second parameter toSave(), ensuring all provider-specific fields are serialized alongside base fieldsThis aligns with the
Savemethod signature shown inpkg/llmproxy/auth/base/token_storage.go:52-95.pkg/llmproxy/thinking/apply.go (1)
535-565: Good iFlow compatibility handling for effort parsing.Supporting both nested
reasoning.effortand escaped literalreasoning\.effortis a solid stabilization change.pkg/llmproxy/thinking/strip_test.go (1)
9-71: Test coverage for iFlow strip behavior looks solid.These cases clearly validate field stripping and preservation of unrelated payload content.
pkg/llmproxy/thinking/apply_iflow_test.go (1)
5-64: Good targeted tests forextractIFlowConfig.The table-driven cases exercise both payload shapes and expected mode/level mapping cleanly.
| pkgconfig "github.com/kooshapari/cliproxyapi-plusplus/v6/internal/config" | ||
| pkgconfig "github.com/kooshapari/cliproxyapi-plusplus/v6/pkg/llmproxy/config" | ||
| tls "github.com/refraction-networking/utls" | ||
| log "github.com/sirupsen/logrus" |
There was a problem hiding this comment.
🧹 Nitpick | 🔵 Trivial
Consider migrating to zerolog.
The code uses logrus for logging, but the coding guidelines specify using zerolog. While this is pre-existing code, consider aligning with the guideline in a future refactor.
As per coding guidelines: "Use zerolog for logging (NOT fmt.Print)".
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed.
In `@pkg/llmproxy/auth/claude/utls_transport.go` at line 13, Current code imports
logrus as log (import alias: log "github.com/sirupsen/logrus"); follow-up
refactor should replace that import with zerolog (e.g., github.com/rs/zerolog)
and update all usages of the log alias (calls like log.Info, log.Errorf,
log.Debug, etc.) to zerolog equivalents (logger.Info().Msg,
logger.Error().Err(err).Msg, etc.), initializing a zerolog logger instance where
needed and ensuring structured fields are migrated to With/Dict patterns; search
for the import line and all references to the log identifier in
utls_transport.go (and related files) to perform the replacement consistently.
Comment on lines
+54
to
+93
| func TestApplyMiniMaxStripsReasoningVariantsAndLegacyFields(t *testing.T) { | ||
| a := NewApplier() | ||
| body := []byte(`{ | ||
| "reasoning_split":true, | ||
| "reasoning_effort":"high", | ||
| "reasoning":{"effort":"medium","summary":{"text":"legacy"}}, | ||
| "variant":"low", | ||
| "foo":"bar" | ||
| }`) | ||
| cfg := thinking.ThinkingConfig{Mode: thinking.ModeLevel, Level: thinking.LevelLow} | ||
| model := ®istry.ModelInfo{ | ||
| ID: "minimax-m2.5", | ||
| Thinking: ®istry.ThinkingSupport{Levels: []string{"low", "medium", "high"}}, | ||
| } | ||
|
|
||
| out, err := a.Apply(body, cfg, model) | ||
| if err != nil { | ||
| t.Fatalf("unexpected error: %v", err) | ||
| } | ||
|
|
||
| res := gjson.ParseBytes(out) | ||
| if !res.Get("reasoning_split").Bool() { | ||
| t.Fatalf("expected reasoning_split=true") | ||
| } | ||
| if res.Get("reasoning_effort").Exists() { | ||
| t.Fatalf("expected reasoning_effort to be removed") | ||
| } | ||
| if res.Get("reasoning").Exists() { | ||
| t.Fatalf("expected reasoning object to be removed") | ||
| } | ||
| if res.Get("reasoning.effort").Exists() { | ||
| t.Fatalf("expected reasoning.effort to be removed") | ||
| } | ||
| if res.Get("variant").Exists() { | ||
| t.Fatalf("expected variant to be removed") | ||
| } | ||
| if res.Get("foo").String() != "bar" { | ||
| t.Fatalf("expected unrelated fields to be preserved") | ||
| } | ||
| } |
There was a problem hiding this comment.
🧹 Nitpick | 🔵 Trivial
Add a regression case for literal reasoning.effort input.
Please add one MiniMax test body containing a top-level "reasoning.effort" key and assert it is removed after Apply, to guard the compatibility cleanup path.
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed.
In `@pkg/llmproxy/thinking/provider/iflow/apply_test.go` around lines 54 - 93,
Update the TestApplyMiniMaxStripsReasoningVariantsAndLegacyFields test to
include a regression case where the JSON body contains a top-level
"reasoning.effort" key (e.g. "reasoning.effort":"medium") alongside the existing
fields; call a.Apply(body, cfg, model) as before and add an assertion that
res.Get("reasoning.effort").Exists() is false (or equivalent check) to ensure
Apply removes the literal reasoning.effort key; keep the rest of the assertions
unchanged so unrelated fields (e.g. "foo") are still preserved.
Comment on lines
+146
to
+148
| result, _ = sjson.DeleteBytes(result, "reasoning_effort") | ||
| result, _ = sjson.DeleteBytes(result, "reasoning") | ||
| result, _ = sjson.DeleteBytes(result, "variant") |
There was a problem hiding this comment.
🧩 Analysis chain
🏁 Script executed:
# First, get the file structure and locate the apply.go file
find . -type f -name "apply.go" | head -20Repository: KooshaPari/cliproxyapi-plusplus
Length of output: 503
🏁 Script executed:
# Read the apply.go file with context around lines 146-148
cat -n pkg/llmproxy/thinking/provider/iflow/apply.go | head -170 | tail -50Repository: KooshaPari/cliproxyapi-plusplus
Length of output: 1842
🏁 Script executed:
# Look for extractIFlowConfig and understand what fields it expects
rg -B 5 -A 15 "extractIFlowConfig" pkg/llmproxy/thinking/Repository: KooshaPari/cliproxyapi-plusplus
Length of output: 5435
🏁 Script executed:
# Search for "reasoning.effort" or "reasoning_effort" patterns in the codebase
rg "reasoning[._]effort" --type goRepository: KooshaPari/cliproxyapi-plusplus
Length of output: 31454
🏁 Script executed:
# Check the test for applyMiniMax to see what's expected
cat -n pkg/llmproxy/thinking/provider/iflow/apply_test.go | grep -A 30 "applyMiniMax"Repository: KooshaPari/cliproxyapi-plusplus
Length of output: 57
🏁 Script executed:
# Look for how sjson is used with dotted keys in the codebase
rg -B 2 -A 2 "sjson.*\\\\\." --type goRepository: KooshaPari/cliproxyapi-plusplus
Length of output: 57
🏁 Script executed:
# Check the full apply_iflow_test.go test cases
cat -n pkg/llmproxy/thinking/apply_iflow_test.go | grep -B 10 -A 10 "reasoning.effort"Repository: KooshaPari/cliproxyapi-plusplus
Length of output: 1105
🏁 Script executed:
# Search for actual sjson usage patterns with special characters
rg "sjson\.(Set|Delete)" pkg/llmproxy/thinking/provider/iflow/ --type go -B 1 -A 1Repository: KooshaPari/cliproxyapi-plusplus
Length of output: 1368
🏁 Script executed:
# Look for applyMiniMax tests
cat -n pkg/llmproxy/thinking/provider/iflow/apply_test.goRepository: KooshaPari/cliproxyapi-plusplus
Length of output: 3416
🏁 Script executed:
# Check how sjson handles dotted keys - search for documentation or usage examples
rg "sjson" pkg/llmproxy/thinking/provider/iflow/apply_test.go -B 2 -A 2Repository: KooshaPari/cliproxyapi-plusplus
Length of output: 57
🏁 Script executed:
# Let's check if there are any tests that verify the cleanup of reasoning.effort (literal key)
rg "reasoning\.effort" pkg/llmproxy/thinking/provider/iflow/ -B 3 -A 3Repository: KooshaPari/cliproxyapi-plusplus
Length of output: 754
🏁 Script executed:
# Check if there are any tests that specifically test literal "reasoning.effort" key removal in applyMiniMax
cat -n pkg/llmproxy/thinking/provider/iflow/apply_test.go | grep -B 20 "reasoning.effort"Repository: KooshaPari/cliproxyapi-plusplus
Length of output: 3114
🏁 Script executed:
# Look at the full test to see if literal keys are tested
cat -n pkg/llmproxy/thinking/provider/iflow/apply_test.go | sed -n '54,93p'Repository: KooshaPari/cliproxyapi-plusplus
Length of output: 1546
🏁 Script executed:
# Search for any sjson escape examples or documentation in the codebase
rg "sjson.*Delete.*\\\\\\\\" --type goRepository: KooshaPari/cliproxyapi-plusplus
Length of output: 57
🏁 Script executed:
# Check the tidwall/sjson documentation by searching for test patterns
rg -B 5 -A 5 "literal.*key\|key.*literal" pkg/llmproxy/thinking/provider/iflow/Repository: KooshaPari/cliproxyapi-plusplus
Length of output: 57
🏁 Script executed:
# Run the actual test to see if it passes
cd . && go test ./pkg/llmproxy/thinking/provider/iflow -v -run TestApplyMiniMaxStripsReasoningVariantsAndLegacyFields 2>&1 | head -100Repository: KooshaPari/cliproxyapi-plusplus
Length of output: 597
🏁 Script executed:
# Check if the test input can actually contain a literal "reasoning.effort" key
python3 << 'EOF'
import json
import subprocess
# Test 1: nested structure (what's in the test)
nested = {
"reasoning_split": True,
"reasoning_effort": "high",
"reasoning": {"effort": "medium", "summary": {"text": "legacy"}},
"variant": "low",
"foo": "bar"
}
# Test 2: literal dotted key
literal = {
"reasoning.effort": "high", # literal key with a dot
"foo": "bar"
}
print("Test 1 - Nested structure:")
print(json.dumps(nested))
print("\nTest 2 - Literal dotted key:")
print(json.dumps(literal))
EOFRepository: KooshaPari/cliproxyapi-plusplus
Length of output: 326
🏁 Script executed:
# Now let's test if applyMiniMax would clean up a literal "reasoning.effort" key
# by creating a simple test
python3 << 'EOF'
import json
# When JSON is parsed, "reasoning.effort" as a key is literally that string
# gjson.Get("reasoning.effort") looks for nested path reasoning -> effort
# But a literal key "reasoning.effort" is at the top level
body_with_literal_key = {
"reasoning.effort": "high",
"foo": "bar"
}
print("JSON with literal dotted key:")
print(json.dumps(body_with_literal_key))
print("\nAccessing via gjson path 'reasoning.effort' would NOT find this")
print("because it looks for nested path, not the literal key")
print("\nAccessing via literal path 'reasoning\\.effort' (escaped) would find it")
EOFRepository: KooshaPari/cliproxyapi-plusplus
Length of output: 338
🏁 Script executed:
# Look for how sjson handles literal keys with dots
rg -B 3 -A 3 "sjson.*Delete.*reasoning" pkg/llmproxy/thinking/ --type goRepository: KooshaPari/cliproxyapi-plusplus
Length of output: 693
🏁 Script executed:
# Check if there's documentation or examples of cleaning literal dotted keys in tests
cat -n pkg/llmproxy/thinking/strip_test.go | head -50Repository: KooshaPari/cliproxyapi-plusplus
Length of output: 1770
🏁 Script executed:
# Let's verify what the extractIFlowConfig actually extracts from a literal key
cat -n pkg/llmproxy/thinking/apply.go | grep -A 30 "func extractIFlowConfig"Repository: KooshaPari/cliproxyapi-plusplus
Length of output: 1409
🏁 Script executed:
# Now create a test to verify the behavior with a literal "reasoning.effort" key
python3 << 'EOF'
import json
import subprocess
# Create test input with literal "reasoning.effort" key
test_json = {
"reasoning.effort": "high",
"reasoning_split": False,
"foo": "bar"
}
print("Test JSON with literal 'reasoning.effort' key:")
print(json.dumps(test_json))
# This would be sent as input to applyMiniMax
# The question is: does the cleanup remove it?
EOFRepository: KooshaPari/cliproxyapi-plusplus
Length of output: 190
🏁 Script executed:
# Check the actual behavior by looking at strip.go which seems to handle similar cleanup
cat -n pkg/llmproxy/thinking/strip.go | head -80Repository: KooshaPari/cliproxyapi-plusplus
Length of output: 2353
🏁 Script executed:
# Perfect! I can see the key finding: in extractIFlowConfig (line 541),
# it specifically looks for escaped literal key "reasoning\\.effort"
# Let's verify if applyMiniMax actually handles this
# Create a test to see if the literal key would be cleaned up
python3 << 'EOF'
# The critical finding:
# extractIFlowConfig DOES handle both:
# 1. Nested: gjson.GetBytes(body, "reasoning.effort")
# 2. Literal: gjson.GetBytes(body, "reasoning\\.effort")
# But applyMiniMax only cleans:
# - "reasoning_effort" (underscore)
# - "reasoning" (nested object)
# - "variant"
# It does NOT clean the literal "reasoning.effort" key!
# This means if extractIFlowConfig accepts it, but applyMiniMax doesn't remove it,
# it will leak upstream.
print("Confirmed: The literal 'reasoning.effort' key is supported by extractIFlowConfig")
print("but NOT cleaned up by applyMiniMax - this is indeed a bug!")
EOFRepository: KooshaPari/cliproxyapi-plusplus
Length of output: 214
🏁 Script executed:
# Now let's verify the correct escaping for sjson
# Check how sjson is used elsewhere with dots
rg "sjson.*\\\\\\\\" --type go | head -20Repository: KooshaPari/cliproxyapi-plusplus
Length of output: 57
🏁 Script executed:
# Let's see if there are any other examples of sjson with escaped keys
rg "sjson.Delete.*\\\\" pkg/llmproxy/ --type goRepository: KooshaPari/cliproxyapi-plusplus
Length of output: 57
🏁 Script executed:
# Check what the web says about sjson path escaping
cat > /tmp/test_sjson.go << 'EOF'
package main
import (
"fmt"
"github.com/tidwall/sjson"
"github.com/tidwall/gjson"
)
func main() {
// Test: does escaping work for sjson.DeleteBytes?
body := []byte(`{"reasoning.effort":"high","foo":"bar"}`)
fmt.Println("Original:", string(body))
// Try to delete with escaped path
result, _ := sjson.DeleteBytes(body, "reasoning\\.effort")
fmt.Println("After delete:", string(result))
// Verify it's gone
if gjson.GetBytes(result, "reasoning\\.effort").Exists() {
fmt.Println("ERROR: literal key still exists!")
} else {
fmt.Println("SUCCESS: literal key was removed")
}
}
EOF
cd /tmp && go run test_sjson.go 2>&1Repository: KooshaPari/cliproxyapi-plusplus
Length of output: 470
🏁 Script executed:
# Let's run the actual test in the repo context to see if there's a test case for literal keys
cd pkg/llmproxy/thinking/provider/iflow && go test -v -run TestApplyMiniMaxStripsReasoningVariantsAndLegacyFields 2>&1 | tail -20Repository: KooshaPari/cliproxyapi-plusplus
Length of output: 312
🏁 Script executed:
# Check if the test input actually includes a literal "reasoning.effort" key
cat -n pkg/llmproxy/thinking/provider/iflow/apply_test.go | sed -n '54,62p'Repository: KooshaPari/cliproxyapi-plusplus
Length of output: 407
🏁 Script executed:
# Let's create a focused test in the actual repo to verify the behavior
cat > /tmp/test_literal_key.sh << 'EOF'
#!/bin/bash
cd /repo
cat > test_literal_key.go << 'GOEOF'
package main
import (
"fmt"
"github.com/tidwall/sjson"
"github.com/tidwall/gjson"
)
func main() {
// Test case: literal "reasoning.effort" key
body := []byte(`{"reasoning.effort":"high","foo":"bar"}`)
fmt.Println("Original:", string(body))
// Current applyMiniMax cleanup (no escape)
result1, _ := sjson.DeleteBytes(body, "reasoning.effort")
fmt.Println("After unescaped delete:", string(result1))
fmt.Println("Literal key still exists?", gjson.GetBytes(result1, "reasoning\\.effort").Exists())
// Proposed fix with escape
result2, _ := sjson.DeleteBytes(body, "reasoning\\.effort")
fmt.Println("After escaped delete:", string(result2))
fmt.Println("Literal key still exists?", gjson.GetBytes(result2, "reasoning\\.effort").Exists())
}
GOEOF
go run test_literal_key.go 2>&1
EOF
bash /tmp/test_literal_key.shRepository: KooshaPari/cliproxyapi-plusplus
Length of output: 375
Add cleanup for literal reasoning.effort key in applyMiniMax.
The current cleanup removes nested reasoning and reasoning_effort (underscore), but misses the literal dotted key "reasoning.effort". Since extractIFlowConfig explicitly accepts this format (via escaped path lookup "reasoning\\.effort"), it can be present in input. When not removed, it leaks upstream in the transformed request.
Use the escaped path "reasoning\\.effort" with sjson.DeleteBytes to remove the literal key:
Proposed fix
result, _ := sjson.SetBytes(body, "reasoning_split", reasoningSplit)
result, _ = sjson.DeleteBytes(result, "reasoning_effort")
+result, _ = sjson.DeleteBytes(result, "reasoning\\.effort")
result, _ = sjson.DeleteBytes(result, "reasoning")
result, _ = sjson.DeleteBytes(result, "variant")🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed.
In `@pkg/llmproxy/thinking/provider/iflow/apply.go` around lines 146 - 148, The
cleanup in applyMiniMax misses the literal dotted key "reasoning.effort" so add
an sjson.DeleteBytes call to remove the escaped path "reasoning\\.effort"
alongside the existing deletes; locate the applyMiniMax function in
pkg/llmproxy/thinking/provider/iflow/apply.go and insert a deletion using
sjson.DeleteBytes(result, "reasoning\\.effort") (follow existing error handling
pattern) so that keys accepted by extractIFlowConfig (which reads
"reasoning\\.effort") are not leaked upstream.
Comment on lines
43
to
50
| paths = []string{ | ||
| "chat_template_kwargs.enable_thinking", | ||
| "chat_template_kwargs.clear_thinking", | ||
| "reasoning", | ||
| "reasoning_split", | ||
| "reasoning_effort", | ||
| "variant", | ||
| } |
There was a problem hiding this comment.
❓ Verification inconclusive
Script executed:
find . -type f -name "*.go" | grep -E "(strip|apply)" | head -20Repository: KooshaPari/cliproxyapi-plusplus
Repository: KooshaPari/cliproxyapi-plusplus
Exit code: 0
stdout:
./pkg/llmproxy/thinking/apply.go
./pkg/llmproxy/thinking/apply_codex_variant_test.go
./pkg/llmproxy/thinking/apply_iflow_test.go
./pkg/llmproxy/thinking/apply_logging_test.go
./pkg/llmproxy/thinking/provider/antigravity/apply.go
./pkg/llmproxy/thinking/provider/antigravity/apply_test.go
./pkg/llmproxy/thinking/provider/claude/apply.go
./pkg/llmproxy/thinking/provider/claude/apply_test.go
./pkg/llmproxy/thinking/provider/codex/apply.go
./pkg/llmproxy/thinking/provider/gemini/apply.go
./pkg/llmproxy/thinking/provider/gemini/apply_test.go
./pkg/llmproxy/thinking/provider/geminicli/apply.go
./pkg/llmproxy/thinking/provider/geminicli/apply_test.go
./pkg/llmproxy/thinking/provider/iflow/apply.go
./pkg/llmproxy/thinking/provider/iflow/apply_test.go
./pkg/llmproxy/thinking/provider/kimi/apply.go
./pkg/llmproxy/thinking/provider/openai/apply.go
./pkg/llmproxy/thinking/strip.go
./pkg/llmproxy/thinking/strip_test.go
Script executed:
cd . && git ls-files | grep -E "llmproxy.*thinking" | head -20Repository: KooshaPari/cliproxyapi-plusplus
Repository: KooshaPari/cliproxyapi-plusplus
Exit code: 0
stdout:
pkg/llmproxy/executor/thinking_providers.go
pkg/llmproxy/thinking/apply.go
pkg/llmproxy/thinking/apply_codex_variant_test.go
pkg/llmproxy/thinking/apply_iflow_test.go
pkg/llmproxy/thinking/apply_logging_test.go
pkg/llmproxy/thinking/convert.go
pkg/llmproxy/thinking/convert_test.go
pkg/llmproxy/thinking/errors.go
pkg/llmproxy/thinking/log_redaction.go
pkg/llmproxy/thinking/log_redaction_test.go
pkg/llmproxy/thinking/provider/antigravity/apply.go
pkg/llmproxy/thinking/provider/antigravity/apply_test.go
pkg/llmproxy/thinking/provider/claude/apply.go
pkg/llmproxy/thinking/provider/claude/apply_test.go
pkg/llmproxy/thinking/provider/codex/apply.go
pkg/llmproxy/thinking/provider/gemini/apply.go
pkg/llmproxy/thinking/provider/gemini/apply_test.go
pkg/llmproxy/thinking/provider/geminicli/apply.go
pkg/llmproxy/thinking/provider/geminicli/apply_test.go
pkg/llmproxy/thinking/provider/iflow/apply.go
Script executed:
cat -n pkg/llmproxy/thinking/strip.goRepository: KooshaPari/cliproxyapi-plusplus
Repository: KooshaPari/cliproxyapi-plusplus
Exit code: 0
stdout:
1 // Package thinking provides unified thinking configuration processing.
2 package thinking
3
4 import (
5 "github.com/tidwall/gjson"
6 "github.com/tidwall/sjson"
7 )
8
9 // StripThinkingConfig removes thinking configuration fields from request body.
10 //
11 // This function is used when a model doesn't support thinking but the request
12 // contains thinking configuration. The configuration is silently removed to
13 // prevent upstream API errors.
14 //
15 // Parameters:
16 // - body: Original request body JSON
17 // - provider: Provider name (determines which fields to strip)
18 //
19 // Returns:
20 // - Modified request body JSON with thinking configuration removed
21 // - Original body is returned unchanged if:
22 // - body is empty or invalid JSON
23 // - provider is unknown
24 // - no thinking configuration found
25 func StripThinkingConfig(body []byte, provider string) []byte {
26 if len(body) == 0 || !gjson.ValidBytes(body) {
27 return body
28 }
29
30 var paths []string
31 switch provider {
32 case "claude":
33 paths = []string{"thinking"}
34 case "gemini":
35 paths = []string{"generationConfig.thinkingConfig"}
36 case "gemini-cli", "antigravity":
37 paths = []string{"request.generationConfig.thinkingConfig"}
38 case "openai":
39 paths = []string{"reasoning_effort"}
40 case "codex":
41 paths = []string{"reasoning.effort"}
42 case "iflow":
43 paths = []string{
44 "chat_template_kwargs.enable_thinking",
45 "chat_template_kwargs.clear_thinking",
46 "reasoning",
47 "reasoning_split",
48 "reasoning_effort",
49 "variant",
50 }
51 default:
52 return body
53 }
54
55 result := body
56 for _, path := range paths {
57 result, _ = sjson.DeleteBytes(result, path)
58 }
59 if provider == "iflow" {
60 if kwargs := gjson.GetBytes(result, "chat_template_kwargs"); kwargs.Exists() && len(kwargs.Map()) == 0 {
61 result, _ = sjson.DeleteBytes(result, "chat_template_kwargs")
62 }
63 }
64 return result
65 }
Script executed:
cat -n pkg/llmproxy/thinking/provider/iflow/apply.goRepository: KooshaPari/cliproxyapi-plusplus
Repository: KooshaPari/cliproxyapi-plusplus
Exit code: 0
stdout:
1 // Package iflow implements thinking configuration for iFlow models.
2 //
3 // iFlow models use boolean toggle semantics:
4 // - Models using chat_template_kwargs.enable_thinking (boolean toggle)
5 // - MiniMax models: reasoning_split (boolean)
6 //
7 // Level values are converted to boolean: none=false, all others=true
8 // See: _bmad-output/planning-artifacts/architecture.md#Epic-9
9 package iflow
10
11 import (
12 "strings"
13
14 "github.com/kooshapari/cliproxyapi-plusplus/v6/pkg/llmproxy/registry"
15 "github.com/kooshapari/cliproxyapi-plusplus/v6/pkg/llmproxy/thinking"
16 "github.com/tidwall/gjson"
17 "github.com/tidwall/sjson"
18 )
19
20 // Applier implements thinking.ProviderApplier for iFlow models.
21 //
22 // iFlow-specific behavior:
23 // - enable_thinking toggle models: enable_thinking boolean
24 // - GLM models: enable_thinking boolean + clear_thinking=false
25 // - MiniMax models: reasoning_split boolean
26 // - Level to boolean: none=false, others=true
27 // - No quantized support (only on/off)
28 type Applier struct{}
29
30 var _ thinking.ProviderApplier = (*Applier)(nil)
31
32 // NewApplier creates a new iFlow thinking applier.
33 func NewApplier() *Applier {
34 return &Applier{}
35 }
36
37 func init() {
38 thinking.RegisterProvider("iflow", NewApplier())
39 }
40
41 // Apply applies thinking configuration to iFlow request body.
42 //
43 // Expected output format (GLM):
44 //
45 // {
46 // "chat_template_kwargs": {
47 // "enable_thinking": true,
48 // "clear_thinking": false
49 // }
50 // }
51 //
52 // Expected output format (MiniMax):
53 //
54 // {
55 // "reasoning_split": true
56 // }
57 func (a *Applier) Apply(body []byte, config thinking.ThinkingConfig, modelInfo *registry.ModelInfo) ([]byte, error) {
58 if thinking.IsUserDefinedModel(modelInfo) {
59 return body, nil
60 }
61 if modelInfo.Thinking == nil {
62 return body, nil
63 }
64
65 if isEnableThinkingModel(modelInfo.ID) {
66 return applyEnableThinking(body, config, isGLMModel(modelInfo.ID)), nil
67 }
68
69 if isMiniMaxModel(modelInfo.ID) {
70 return applyMiniMax(body, config), nil
71 }
72
73 return body, nil
74 }
75
76 // configToBoolean converts ThinkingConfig to boolean for iFlow models.
77 //
78 // Conversion rules:
79 // - ModeNone: false
80 // - ModeAuto: true
81 // - ModeBudget + Budget=0: false
82 // - ModeBudget + Budget>0: true
83 // - ModeLevel + Level="none": false
84 // - ModeLevel + any other level: true
85 // - Default (unknown mode): true
86 func configToBoolean(config thinking.ThinkingConfig) bool {
87 switch config.Mode {
88 case thinking.ModeNone:
89 return false
90 case thinking.ModeAuto:
91 return true
92 case thinking.ModeBudget:
93 return config.Budget > 0
94 case thinking.ModeLevel:
95 return config.Level != thinking.LevelNone
96 default:
97 return true
98 }
99 }
100
101 // applyEnableThinking applies thinking configuration for models that use
102 // chat_template_kwargs.enable_thinking format.
103 //
104 // Output format when enabled:
105 //
106 // {"chat_template_kwargs": {"enable_thinking": true, "clear_thinking": false}}
107 //
108 // Output format when disabled:
109 //
110 // {"chat_template_kwargs": {"enable_thinking": false}}
111 //
112 // Note: clear_thinking is only set for GLM models when thinking is enabled.
113 func applyEnableThinking(body []byte, config thinking.ThinkingConfig, setClearThinking bool) []byte {
114 enableThinking := configToBoolean(config)
115
116 if len(body) == 0 || !gjson.ValidBytes(body) {
117 body = []byte(`{}`)
118 }
119
120 result, _ := sjson.SetBytes(body, "chat_template_kwargs.enable_thinking", enableThinking)
121
122 // clear_thinking is a GLM-only knob, strip it for other models.
123 result, _ = sjson.DeleteBytes(result, "chat_template_kwargs.clear_thinking")
124
125 // clear_thinking only needed when thinking is enabled
126 if enableThinking && setClearThinking {
127 result, _ = sjson.SetBytes(result, "chat_template_kwargs.clear_thinking", false)
128 }
129
130 return result
131 }
132
133 // applyMiniMax applies thinking configuration for MiniMax models.
134 //
135 // Output format:
136 //
137 // {"reasoning_split": true/false}
138 func applyMiniMax(body []byte, config thinking.ThinkingConfig) []byte {
139 reasoningSplit := configToBoolean(config)
140
141 if len(body) == 0 || !gjson.ValidBytes(body) {
142 body = []byte(`{}`)
143 }
144
145 result, _ := sjson.SetBytes(body, "reasoning_split", reasoningSplit)
146 result, _ = sjson.DeleteBytes(result, "reasoning_effort")
147 result, _ = sjson.DeleteBytes(result, "reasoning")
148 result, _ = sjson.DeleteBytes(result, "variant")
149
150 return result
151 }
152
153 // isEnableThinkingModel determines if the model uses chat_template_kwargs.enable_thinking format.
154 func isEnableThinkingModel(modelID string) bool {
155 if isGLMModel(modelID) {
156 return true
157 }
158 id := strings.ToLower(modelID)
159 switch id {
160 case "qwen3-max-preview", "deepseek-v3.2", "deepseek-v3.1":
161 return true
162 default:
163 return false
164 }
165 }
166
167 // isGLMModel determines if the model is a GLM series model.
168 func isGLMModel(modelID string) bool {
169 return strings.HasPrefix(strings.ToLower(modelID), "glm")
170 }
171
172 // isMiniMaxModel determines if the model is a MiniMax series model.
173 // MiniMax models use reasoning_split format.
174 func isMiniMaxModel(modelID string) bool {
175 return strings.HasPrefix(strings.ToLower(modelID), "minimax")
176 }
Script executed:
cat -n pkg/llmproxy/thinking/provider/iflow/apply_test.goRepository: KooshaPari/cliproxyapi-plusplus
Repository: KooshaPari/cliproxyapi-plusplus
Exit code: 0
stdout:
1 package iflow
2
3 import (
4 "testing"
5
6 "github.com/kooshapari/cliproxyapi-plusplus/v6/pkg/llmproxy/registry"
7 "github.com/kooshapari/cliproxyapi-plusplus/v6/pkg/llmproxy/thinking"
8 "github.com/tidwall/gjson"
9 )
10
11 func TestApplyMiniMaxStripsReasoningEffort(t *testing.T) {
12 a := NewApplier()
13 body := []byte(`{"reasoning_effort":"high","foo":"bar"}`)
14 cfg := thinking.ThinkingConfig{Mode: thinking.ModeLevel, Level: thinking.LevelHigh}
15 model := ®istry.ModelInfo{
16 ID: "minimax-m2.5",
17 Thinking: ®istry.ThinkingSupport{Levels: []string{"low", "medium", "high"}},
18 }
19
20 out, err := a.Apply(body, cfg, model)
21 if err != nil {
22 t.Fatalf("unexpected error: %v", err)
23 }
24
25 res := gjson.ParseBytes(out)
26 if !res.Get("reasoning_split").Bool() {
27 t.Fatalf("expected reasoning_split=true for non-none config")
28 }
29 if res.Get("reasoning_effort").Exists() {
30 t.Fatalf("expected reasoning_effort to be removed")
31 }
32 }
33
34 func TestApplyMiniMaxSetsReasoningSplitFalseForModeNone(t *testing.T) {
35 a := NewApplier()
36 body := []byte(`{"reasoning_effort":"high","foo":"bar"}`)
37 cfg := thinking.ThinkingConfig{Mode: thinking.ModeNone}
38 model := ®istry.ModelInfo{
39 ID: "minimax-m2",
40 Thinking: ®istry.ThinkingSupport{Levels: []string{"none", "low", "high"}},
41 }
42
43 out, err := a.Apply(body, cfg, model)
44 if err != nil {
45 t.Fatalf("unexpected error: %v", err)
46 }
47
48 res := gjson.ParseBytes(out)
49 if res.Get("reasoning_split").Bool() {
50 t.Fatalf("expected reasoning_split=false for ModeNone")
51 }
52 }
53
54 func TestApplyMiniMaxStripsReasoningVariantsAndLegacyFields(t *testing.T) {
55 a := NewApplier()
56 body := []byte(`{
57 "reasoning_split":true,
58 "reasoning_effort":"high",
59 "reasoning":{"effort":"medium","summary":{"text":"legacy"}},
60 "variant":"low",
61 "foo":"bar"
62 }`)
63 cfg := thinking.ThinkingConfig{Mode: thinking.ModeLevel, Level: thinking.LevelLow}
64 model := ®istry.ModelInfo{
65 ID: "minimax-m2.5",
66 Thinking: ®istry.ThinkingSupport{Levels: []string{"low", "medium", "high"}},
67 }
68
69 out, err := a.Apply(body, cfg, model)
70 if err != nil {
71 t.Fatalf("unexpected error: %v", err)
72 }
73
74 res := gjson.ParseBytes(out)
75 if !res.Get("reasoning_split").Bool() {
76 t.Fatalf("expected reasoning_split=true")
77 }
78 if res.Get("reasoning_effort").Exists() {
79 t.Fatalf("expected reasoning_effort to be removed")
80 }
81 if res.Get("reasoning").Exists() {
82 t.Fatalf("expected reasoning object to be removed")
83 }
84 if res.Get("reasoning.effort").Exists() {
85 t.Fatalf("expected reasoning.effort to be removed")
86 }
87 if res.Get("variant").Exists() {
88 t.Fatalf("expected variant to be removed")
89 }
90 if res.Get("foo").String() != "bar" {
91 t.Fatalf("expected unrelated fields to be preserved")
92 }
93 }
Script executed:
rg "reasoning\.effort" --type go -A 2 -B 2Repository: KooshaPari/cliproxyapi-plusplus
Repository: KooshaPari/cliproxyapi-plusplus
Exit code: 0
stdout:
test/thinking_conversion_test.go- model: "level-model",
test/thinking_conversion_test.go- inputJSON: `{"model":"level-model","messages":[{"role":"user","content":"hi"}]}`,
test/thinking_conversion_test.go: expectField: "reasoning.effort",
test/thinking_conversion_test.go- expectValue: "medium",
test/thinking_conversion_test.go- expectErr: false,
--
test/thinking_conversion_test.go- model: "level-model(medium)",
test/thinking_conversion_test.go- inputJSON: `{"model":"level-model(medium)","messages":[{"role":"user","content":"hi"}]}`,
test/thinking_conversion_test.go: expectField: "reasoning.effort",
test/thinking_conversion_test.go- expectValue: "medium",
test/thinking_conversion_test.go- expectErr: false,
--
test/thinking_conversion_test.go- model: "level-model(none)",
test/thinking_conversion_test.go- inputJSON: `{"model":"level-model(none)","messages":[{"role":"user","content":"hi"}]}`,
test/thinking_conversion_test.go: expectField: "reasoning.effort",
test/thinking_conversion_test.go- expectValue: "minimal",
test/thinking_conversion_test.go- expectErr: false,
--
test/thinking_conversion_test.go- model: "level-model(auto)",
test/thinking_conversion_test.go- inputJSON: `{"model":"level-model(auto)","messages":[{"role":"user","content":"hi"}]}`,
test/thinking_conversion_test.go: expectField: "reasoning.effort",
test/thinking_conversion_test.go- expectValue: "medium",
test/thinking_conversion_test.go- expectErr: false,
--
test/thinking_conversion_test.go- model: "level-model",
test/thinking_conversion_test.go- inputJSON: `{"model":"level-model","contents":[{"role":"user","parts":[{"text":"hi"}]}]}`,
test/thinking_conversion_test.go: expectField: "reasoning.effort",
test/thinking_conversion_test.go- expectValue: "medium",
test/thinking_conversion_test.go- expectErr: false,
--
test/thinking_conversion_test.go- model: "level-model(8192)",
test/thinking_conversion_test.go- inputJSON: `{"model":"level-model(8192)","contents":[{"role":"user","parts":[{"text":"hi"}]}]}`,
test/thinking_conversion_test.go: expectField: "reasoning.effort",
test/thinking_conversion_test.go- expectValue: "medium",
test/thinking_conversion_test.go- expectErr: false,
--
test/thinking_conversion_test.go- model: "level-model(64000)",
test/thinking_conversion_test.go- inputJSON: `{"model":"level-model(64000)","contents":[{"role":"user","parts":[{"text":"hi"}]}]}`,
test/thinking_conversion_test.go: expectField: "reasoning.effort",
test/thinking_conversion_test.go- expectValue: "high",
test/thinking_conversion_test.go- expectErr: false,
--
test/thinking_conversion_test.go- model: "level-model(0)",
test/thinking_conversion_test.go- inputJSON: `{"model":"level-model(0)","contents":[{"role":"user","parts":[{"text":"hi"}]}]}`,
test/thinking_conversion_test.go: expectField: "reasoning.effort",
test/thinking_conversion_test.go- expectValue: "minimal",
test/thinking_conversion_test.go- expectErr: false,
--
test/thinking_conversion_test.go- model: "level-model(-1)",
test/thinking_conversion_test.go- inputJSON: `{"model":"level-model(-1)","contents":[{"role":"user","parts":[{"text":"hi"}]}]}`,
test/thinking_conversion_test.go: expectField: "reasoning.effort",
test/thinking_conversion_test.go- expectValue: "medium",
test/thinking_conversion_test.go- expectErr: false,
--
test/thinking_conversion_test.go- model: "user-defined-model",
test/thinking_conversion_test.go- inputJSON: `{"model":"user-defined-model","messages":[{"role":"user","content":"hi"}]}`,
test/thinking_conversion_test.go: expectField: "reasoning.effort",
test/thinking_conversion_test.go- expectValue: "medium",
test/thinking_conversion_test.go- expectErr: false,
--
test/thinking_conversion_test.go- model: "user-defined-model(8192)",
test/thinking_conversion_test.go- inputJSON: `{"model":"user-defined-model(8192)","messages":[{"role":"user","content":"hi"}]}`,
test/thinking_conversion_test.go: expectField: "reasoning.effort",
test/thinking_conversion_test.go- expectValue: "medium",
test/thinking_conversion_test.go- expectErr: false,
--
test/thinking_conversion_test.go- model: "user-defined-model(64000)",
test/thinking_conversion_test.go- inputJSON: `{"model":"user-defined-model(64000)","messages":[{"role":"user","content":"hi"}]}`,
test/thinking_conversion_test.go: expectField: "reasoning.effort",
test/thinking_conversion_test.go- expectValue: "xhigh",
test/thinking_conversion_test.go- expectErr: false,
--
test/thinking_conversion_test.go- model: "user-defined-model(0)",
test/thinking_conversion_test.go- inputJSON: `{"model":"user-defined-model(0)","messages":[{"role":"user","content":"hi"}]}`,
test/thinking_conversion_test.go: expectField: "reasoning.effort",
test/thinking_conversion_test.go- expectValue: "none",
test/thinking_conversion_test.go- expectErr: false,
--
test/thinking_conversion_test.go- model: "user-defined-model(-1)",
test/thinking_conversion_test.go- inputJSON: `{"model":"user-defined-model(-1)","messages":[{"role":"user","content":"hi"}]}`,
test/thinking_conversion_test.go: expectField: "reasoning.effort",
test/thinking_conversion_test.go- expectValue: "auto",
test/thinking_conversion_test.go- expectErr: false,
--
test/thinking_conversion_test.go- expectErr: true,
test/thinking_conversion_test.go- },
test/thinking_conversion_test.go: // Case 82: OpenAI-Response to Codex, level high → passthrough reasoning.effort
test/thinking_conversion_test.go- {
test/thinking_conversion_test.go- name: "82",
--
test/thinking_conversion_test.go- model: "level-model(high)",
test/thinking_conversion_test.go- inputJSON: `{"model":"level-model(high)","input":[{"role":"user","content":"hi"}]}`,
test/thinking_conversion_test.go: expectField: "reasoning.effort",
test/thinking_conversion_test.go- expectValue: "high",
test/thinking_conversion_test.go- expectErr: false,
--
test/thinking_conversion_test.go- model: "gpt-5-codex(high)",
test/thinking_conversion_test.go- inputJSON: `{"model":"gpt-5-codex(high)","input":[{"role":"user","content":"hi"}]}`,
test/thinking_conversion_test.go: expectField: "reasoning.effort",
test/thinking_conversion_test.go- expectValue: "high",
test/thinking_conversion_test.go- expectErr: false,
--
test/thinking_conversion_test.go- model: "gpt-5.2-codex(xhigh)",
test/thinking_conversion_test.go- inputJSON: `{"model":"gpt-5.2-codex(xhigh)","input":[{"role":"user","content":"hi"}]}`,
test/thinking_conversion_test.go: expectField: "reasoning.effort",
test/thinking_conversion_test.go- expectValue: "xhigh",
test/thinking_conversion_test.go- expectErr: false,
--
test/thinking_conversion_test.go- model: "gpt-5.2-codex(none)",
test/thinking_conversion_test.go- inputJSON: `{"model":"gpt-5.2-codex(none)","input":[{"role":"user","content":"hi"}]}`,
test/thinking_conversion_test.go: expectField: "reasoning.effort",
test/thinking_conversion_test.go- expectValue: "none",
test/thinking_conversion_test.go- expectErr: false,
--
test/thinking_conversion_test.go- model: "gpt-5-codex(none)",
test/thinking_conversion_test.go- inputJSON: `{"model":"gpt-5-codex(none)","input":[{"role":"user","content":"hi"}]}`,
test/thinking_conversion_test.go: expectField: "reasoning.effort",
test/thinking_conversion_test.go- expectValue: "low",
test/thinking_conversion_test.go- expectErr: false,
--
test/thinking_conversion_test.go- model: "level-model",
test/thinking_conversion_test.go- inputJSON: `{"model":"level-model","messages":[{"role":"user","content":"hi"}]}`,
test/thinking_conversion_test.go: expectField: "reasoning.effort",
test/thinking_conversion_test.go- expectValue: "medium",
test/thinking_conversion_test.go- expectErr: false,
--
test/thinking_conversion_test.go- model: "level-model",
test/thinking_conversion_test.go- inputJSON: `{"model":"level-model","messages":[{"role":"user","content":"hi"}],"reasoning_effort":"medium"}`,
test/thinking_conversion_test.go: expectField: "reasoning.effort",
test/thinking_conversion_test.go- expectValue: "medium",
test/thinking_conversion_test.go- expectErr: false,
--
test/thinking_conversion_test.go- model: "level-model",
test/thinking_conversion_test.go- inputJSON: `{"model":"level-model","messages":[{"role":"user","content":"hi"}],"reasoning_effort":"none"}`,
test/thinking_conversion_test.go: expectField: "reasoning.effort",
test/thinking_conversion_test.go- expectValue: "minimal",
test/thinking_conversion_test.go- expectErr: false,
--
test/thinking_conversion_test.go- model: "level-model",
test/thinking_conversion_test.go- inputJSON: `{"model":"level-model","messages":[{"role":"user","content":"hi"}],"reasoning_effort":"auto"}`,
test/thinking_conversion_test.go: expectField: "reasoning.effort",
test/thinking_conversion_test.go- expectValue: "medium",
test/thinking_conversion_test.go- expectErr: false,
--
test/thinking_conversion_test.go- model: "level-model",
test/thinking_conversion_test.go- inputJSON: `{"model":"level-model","contents":[{"role":"user","parts":[{"text":"hi"}]}]}`,
test/thinking_conversion_test.go: expectField: "reasoning.effort",
test/thinking_conversion_test.go- expectValue: "medium",
test/thinking_conversion_test.go- expectErr: false,
--
test/thinking_conversion_test.go- model: "level-model",
test/thinking_conversion_test.go- inputJSON: `{"model":"level-model","contents":[{"role":"user","parts":[{"text":"hi"}]}],"generationConfig":{"thinkingConfig":{"thinkingBudget":8192}}}`,
test/thinking_conversion_test.go: expectField: "reasoning.effort",
test/thinking_conversion_test.go- expectValue: "medium",
test/thinking_conversion_test.go- expectErr: false,
--
test/thinking_conversion_test.go- model: "level-model",
test/thinking_conversion_test.go- inputJSON: `{"model":"level-model","contents":[{"role":"user","parts":[{"text":"hi"}]}],"generationConfig":{"thinkingConfig":{"thinkingBudget":64000}}}`,
test/thinking_conversion_test.go: expectField: "reasoning.effort",
test/thinking_conversion_test.go- expectValue: "high",
test/thinking_conversion_test.go- expectErr: false,
--
test/thinking_conversion_test.go- model: "level-model",
test/thinking_conversion_test.go- inputJSON: `{"model":"level-model","contents":[{"role":"user","parts":[{"text":"hi"}]}],"generationConfig":{"thinkingConfig":{"thinkingBudget":0}}}`,
test/thinking_conversion_test.go: expectField: "reasoning.effort",
test/thinking_conversion_test.go- expectValue: "minimal",
test/thinking_conversion_test.go- expectErr: false,
--
test/thinking_conversion_test.go- model: "level-model",
test/thinking_conversion_test.go- inputJSON: `{"model":"level-model","contents":[{"role":"user","parts":[{"text":"hi"}]}],"generationConfig":{"thinkingConfig":{"thinkingBudget":-1}}}`,
test/thinking_conversion_test.go: expectField: "reasoning.effort",
test/thinking_conversion_test.go- expectValue: "medium",
test/thinking_conversion_test.go- expectErr: false,
--
test/thinking_conversion_test.go- model: "user-defined-model",
test/thinking_conversion_test.go- inputJSON: `{"model":"user-defined-model","messages":[{"role":"user","content":"hi"}]}`,
test/thinking_conversion_test.go: expectField: "reasoning.effort",
test/thinking_conversion_test.go- expectValue: "medium",
test/thinking_conversion_test.go- expectErr: false,
--
test/thinking_conversion_test.go- model: "user-defined-model",
test/thinking_conversion_test.go- inputJSON: `{"model":"user-defined-model","messages":[{"role":"user","content":"hi"}],"thinking":{"type":"enabled","budget_tokens":8192}}`,
test/thinking_conversion_test.go: expectField: "reasoning.effort",
test/thinking_conversion_test.go- expectValue: "medium",
test/thinking_conversion_test.go- expectErr: false,
--
test/thinking_conversion_test.go- model: "user-defined-model",
test/thinking_conversion_test.go- inputJSON: `{"model":"user-defined-model","messages":[{"role":"user","content":"hi"}],"thinking":{"type":"enabled","budget_tokens":64000}}`,
test/thinking_conversion_test.go: expectField: "reasoning.effort",
test/thinking_conversion_test.go- expectValue: "xhigh",
test/thinking_conversion_test.go- expectErr: false,
--
test/thinking_conversion_test.go- model: "user-defined-model",
test/thinking_conversion_test.go- inputJSON: `{"model":"user-defined-model","messages":[{"role":"user","content":"hi"}],"thinking":{"type":"enabled","budget_tokens":0}}`,
test/thinking_conversion_test.go: expectField: "reasoning.effort",
test/thinking_conversion_test.go- expectValue: "none",
test/thinking_conversion_test.go- expectErr: false,
--
test/thinking_conversion_test.go- model: "user-defined-model",
test/thinking_conversion_test.go- inputJSON: `{"model":"user-defined-model","messages":[{"role":"user","content":"hi"}],"thinking":{"type":"enabled","budget_tokens":-1}}`,
test/thinking_conversion_test.go: expectField: "reasoning.effort",
test/thinking_conversion_test.go- expectValue: "auto",
test/thinking_conversion_test.go- expectErr: false,
--
test/thinking_conversion_test.go- expectErr: false,
test/thinking_conversion_test.go- },
test/thinking_conversion_test.go: // Case 78: OpenAI-Response reasoning.effort=medium to Gemini → 8192
test/thinking_conversion_test.go- {
test/thinking_conversion_test.go- name: "78",
--
test/thinking_conversion_test.go- expectErr: false,
test/thinking_conversion_test.go- },
test/thinking_conversion_test.go: // Case 79: OpenAI-Response reasoning.effort=medium to Claude → 8192
test/thinking_conversion_test.go- {
test/thinking_conversion_test.go- name: "79",
--
test/thinking_conversion_test.go- expectErr: true,
test/thinking_conversion_test.go- },
test/thinking_conversion_test.go: // Case 82: OpenAI-Response to Codex, reasoning.effort=high → passthrough
test/thinking_conversion_test.go- {
test/thinking_conversion_test.go- name: "82",
--
test/thinking_conversion_test.go- model: "level-model",
test/thinking_conversion_test.go- inputJSON: `{"model":"level-model","input":[{"role":"user","content":"hi"}],"reasoning":{"effort":"high"}}`,
test/thinking_conversion_test.go: expectField: "reasoning.effort",
test/thinking_conversion_test.go- expectValue: "high",
test/thinking_conversion_test.go- expectErr: false,
test/thinking_conversion_test.go- },
test/thinking_conversion_test.go: // Case 83: OpenAI-Response to Codex, reasoning.effort=xhigh → out of range error
test/thinking_conversion_test.go- {
test/thinking_conversion_test.go- name: "83",
--
test/thinking_conversion_test.go- // GitHub Copilot tests: /responses endpoint (codex format) - with body params
test/thinking_conversion_test.go-
test/thinking_conversion_test.go: // Case 118: OpenAI-Response to gpt-5-codex, reasoning.effort=high → high
test/thinking_conversion_test.go- {
test/thinking_conversion_test.go- name: "118",
--
test/thinking_conversion_test.go- model: "gpt-5-codex",
test/thinking_conversion_test.go- inputJSON: `{"model":"gpt-5-codex","input":[{"role":"user","content":"hi"}],"reasoning":{"effort":"high"}}`,
test/thinking_conversion_test.go: expectField: "reasoning.effort",
test/thinking_conversion_test.go- expectValue: "high",
test/thinking_conversion_test.go- expectErr: false,
test/thinking_conversion_test.go- },
test/thinking_conversion_test.go: // Case 119: OpenAI-Response to gpt-5.2-codex, reasoning.effort=xhigh → xhigh
test/thinking_conversion_test.go- {
test/thinking_conversion_test.go- name: "119",
--
test/thinking_conversion_test.go- model: "gpt-5.2-codex",
test/thinking_conversion_test.go- inputJSON: `{"model":"gpt-5.2-codex","input":[{"role":"user","content":"hi"}],"reasoning":{"effort":"xhigh"}}`,
test/thinking_conversion_test.go: expectField: "reasoning.effort",
test/thinking_conversion_test.go- expectValue: "xhigh",
test/thinking_conversion_test.go- expectErr: false,
test/thinking_conversion_test.go- },
test/thinking_conversion_test.go: // Case 120: OpenAI-Response to gpt-5.2-codex, reasoning.effort=none → none
test/thinking_conversion_test.go- {
test/thinking_conversion_test.go- name: "120",
--
test/thinking_conversion_test.go- model: "gpt-5.2-codex",
test/thinking_conversion_test.go- inputJSON: `{"model":"gpt-5.2-codex","input":[{"role":"user","content":"hi"}],"reasoning":{"effort":"none"}}`,
test/thinking_conversion_test.go: expectField: "reasoning.effort",
test/thinking_conversion_test.go- expectValue: "none",
test/thinking_conversion_test.go- expectErr: false,
test/thinking_conversion_test.go- },
test/thinking_conversion_test.go: // Case 121: OpenAI-Response to gpt-5-codex, reasoning.effort=none → clamped to low (ZeroAllowed=false)
test/thinking_conversion_test.go- {
test/thinking_conversion_test.go- name: "121",
--
test/thinking_conversion_test.go- model: "gpt-5-codex",
test/thinking_conversion_test.go- inputJSON: `{"model":"gpt-5-codex","input":[{"role":"user","content":"hi"}],"reasoning":{"effort":"none"}}`,
test/thinking_conversion_test.go: expectField: "reasoning.effort",
test/thinking_conversion_test.go- expectValue: "low",
test/thinking_conversion_test.go- expectErr: false,
--
test/thinking_conversion_test.go- model: "level-model",
test/thinking_conversion_test.go- inputJSON: `{"model":"level-model","messages":[{"role":"user","content":"hi"}],"thinking":{"type":"adaptive"}}`,
test/thinking_conversion_test.go: expectField: "reasoning.effort",
test/thinking_conversion_test.go- expectValue: "high",
test/thinking_conversion_test.go- expectErr: false,
--
test/thinking_conversion_test.go- hasThinking = gjson.GetBytes(body, "reasoning_effort").Exists()
test/thinking_conversion_test.go- case "codex":
test/thinking_conversion_test.go: hasThinking = gjson.GetBytes(body, "reasoning.effort").Exists() || gjson.GetBytes(body, "reasoning").Exists()
test/thinking_conversion_test.go- case "iflow":
test/thinking_conversion_test.go- hasThinking = gjson.GetBytes(body, "chat_template_kwargs.enable_thinking").Exists() || gjson.GetBytes(body, "reasoning_split").Exists()
--
pkg/llmproxy/translator/openai/openai/responses/openai_openai-responses_request.go- //
pkg/llmproxy/translator/openai/openai/responses/openai_openai-responses_request.go- // Priority:
pkg/llmproxy/translator/openai/openai/responses/openai_openai-responses_request.go: // 1. reasoning.effort object field
pkg/llmproxy/translator/openai/openai/responses/openai_openai-responses_request.go: // 2. flat legacy field "reasoning.effort"
pkg/llmproxy/translator/openai/openai/responses/openai_openai-responses_request.go- // 3. variant
pkg/llmproxy/translator/openai/openai/responses/openai_openai-responses_request.go: if reasoningEffort := root.Get("reasoning.effort"); reasoningEffort.Exists() {
pkg/llmproxy/translator/openai/openai/responses/openai_openai-responses_request.go- effort := strings.ToLower(strings.TrimSpace(reasoningEffort.String()))
pkg/llmproxy/translator/openai/openai/responses/openai_openai-responses_request.go- if effort != "" {
--
pkg/llmproxy/translator/gemini/openai/responses/gemini_openai-responses_request.go- }
pkg/llmproxy/translator/gemini/openai/responses/gemini_openai-responses_request.go-
pkg/llmproxy/translator/gemini/openai/responses/gemini_openai-responses_request.go: // Apply thinking configuration: convert OpenAI Responses API reasoning.effort to Gemini thinkingConfig.
pkg/llmproxy/translator/gemini/openai/responses/gemini_openai-responses_request.go- // Inline translation-only mapping; capability checks happen later in ApplyThinking.
pkg/llmproxy/translator/gemini/openai/responses/gemini_openai-responses_request.go: re := root.Get("reasoning.effort")
pkg/llmproxy/translator/gemini/openai/responses/gemini_openai-responses_request.go- if re.Exists() {
pkg/llmproxy/translator/gemini/openai/responses/gemini_openai-responses_request.go- effort := strings.ToLower(strings.TrimSpace(re.String()))
--
pkg/llmproxy/translator/codex/openai/chat-completions/codex_openai_request_test.go- }
pkg/llmproxy/translator/codex/openai/chat-completions/codex_openai_request_test.go-
pkg/llmproxy/translator/codex/openai/chat-completions/codex_openai_request_test.go: if res.Get("reasoning.effort").String() != "medium" {
pkg/llmproxy/translator/codex/openai/chat-completions/codex_openai_request_test.go: t.Errorf("expected reasoning.effort medium, got %s", res.Get("reasoning.effort").String())
pkg/llmproxy/translator/codex/openai/chat-completions/codex_openai_request_test.go- }
pkg/llmproxy/translator/codex/openai/chat-completions/codex_openai_request_test.go-
--
pkg/llmproxy/translator/codex/openai/chat-completions/codex_openai_request_test.go- res2 := gjson.ParseBytes(got2)
pkg/llmproxy/translator/codex/openai/chat-completions/codex_openai_request_test.go-
pkg/llmproxy/translator/codex/openai/chat-completions/codex_openai_request_test.go: if res2.Get("reasoning.effort").String() != "high" {
pkg/llmproxy/translator/codex/openai/chat-completions/codex_openai_request_test.go: t.Errorf("expected reasoning.effort high, got %s", res2.Get("reasoning.effort").String())
pkg/llmproxy/translator/codex/openai/chat-completions/codex_openai_request_test.go- }
pkg/llmproxy/translator/codex/openai/chat-completions/codex_openai_request_test.go-
--
pkg/llmproxy/translator/codex/openai/chat-completions/codex_openai_request_test.go- res := gjson.ParseBytes(got)
pkg/llmproxy/translator/codex/openai/chat-completions/codex_openai_request_test.go-
pkg/llmproxy/translator/codex/openai/chat-completions/codex_openai_request_test.go: if gotEffort := res.Get("reasoning.effort").String(); gotEffort != "high" {
pkg/llmproxy/translator/codex/openai/chat-completions/codex_openai_request_test.go: t.Fatalf("expected reasoning.effort to use variant fallback high, got %s", gotEffort)
pkg/llmproxy/translator/codex/openai/chat-completions/codex_openai_request_test.go- }
pkg/llmproxy/translator/codex/openai/chat-completions/codex_openai_request_test.go-}
--
pkg/llmproxy/translator/codex/openai/chat-completions/codex_openai_request_test.go- "model": "gpt-4o",
pkg/llmproxy/translator/codex/openai/chat-completions/codex_openai_request_test.go- "messages": [{"role":"user","content":"hello"}],
pkg/llmproxy/translator/codex/openai/chat-completions/codex_openai_request_test.go: "reasoning.effort": "low"
pkg/llmproxy/translator/codex/openai/chat-completions/codex_openai_request_test.go- }`)
pkg/llmproxy/translator/codex/openai/chat-completions/codex_openai_request_test.go- got := ConvertOpenAIRequestToCodex("gpt-4o", input, false)
pkg/llmproxy/translator/codex/openai/chat-completions/codex_openai_request_test.go- res := gjson.ParseBytes(got)
pkg/llmproxy/translator/codex/openai/chat-completions/codex_openai_request_test.go-
pkg/llmproxy/translator/codex/openai/chat-completions/codex_openai_request_test.go: if gotEffort := res.Get("reasoning.effort").String(); gotEffort != "low" {
pkg/llmproxy/translator/codex/openai/chat-completions/codex_openai_request_test.go: t.Fatalf("expected reasoning.effort to use legacy flat field low, got %s", gotEffort)
pkg/llmproxy/translator/codex/openai/chat-completions/codex_openai_request_test.go- }
pkg/llmproxy/translator/codex/openai/chat-completions/codex_openai_request_test.go-}
--
pkg/llmproxy/translator/codex/openai/chat-completions/codex_openai_request_test.go- res := gjson.ParseBytes(got)
pkg/llmproxy/translator/codex/openai/chat-completions/codex_openai_request_test.go-
pkg/llmproxy/translator/codex/openai/chat-completions/codex_openai_request_test.go: if gotEffort := res.Get("reasoning.effort").String(); gotEffort != "low" {
pkg/llmproxy/translator/codex/openai/chat-completions/codex_openai_request_test.go: t.Fatalf("expected reasoning.effort to prefer reasoning_effort low, got %s", gotEffort)
pkg/llmproxy/translator/codex/openai/chat-completions/codex_openai_request_test.go- }
pkg/llmproxy/translator/codex/openai/chat-completions/codex_openai_request_test.go-}
--
pkg/llmproxy/translator/codex/openai/responses/codex_openai-responses_request_test.go- outputStr := string(output)
pkg/llmproxy/translator/codex/openai/responses/codex_openai-responses_request_test.go-
pkg/llmproxy/translator/codex/openai/responses/codex_openai-responses_request_test.go: if got := gjson.Get(outputStr, "reasoning.effort").String(); got != "high" {
pkg/llmproxy/translator/codex/openai/responses/codex_openai-responses_request_test.go: t.Fatalf("expected reasoning.effort=high fallback, got %s", got)
pkg/llmproxy/translator/codex/openai/responses/codex_openai-responses_request_test.go- }
pkg/llmproxy/translator/codex/openai/responses/codex_openai-responses_request_test.go-}
--
pkg/llmproxy/translator/codex/openai/responses/codex_openai-responses_request_test.go- outputStr := string(output)
pkg/llmproxy/translator/codex/openai/responses/codex_openai-responses_request_test.go-
pkg/llmproxy/translator/codex/openai/responses/codex_openai-responses_request_test.go: if got := gjson.Get(outputStr, "reasoning.effort").String(); got != "low" {
pkg/llmproxy/translator/codex/openai/responses/codex_openai-responses_request_test.go: t.Fatalf("expected reasoning.effort to prefer explicit reasoning.effort low, got %s", got)
pkg/llmproxy/translator/codex/openai/responses/codex_openai-responses_request_test.go- }
pkg/llmproxy/translator/codex/openai/responses/codex_openai-responses_request_test.go-}
--
pkg/llmproxy/translator/codex/openai/chat-completions/codex_openai_request.go- // Map reasoning effort; support flat legacy field and variant fallback.
pkg/llmproxy/translator/codex/openai/chat-completions/codex_openai_request.go- if v := gjson.GetBytes(rawJSON, "reasoning_effort"); v.Exists() {
pkg/llmproxy/translator/codex/openai/chat-completions/codex_openai_request.go: out, _ = sjson.Set(out, "reasoning.effort", v.Value())
pkg/llmproxy/translator/codex/openai/chat-completions/codex_openai_request.go- } else if v := gjson.GetBytes(rawJSON, `reasoning\.effort`); v.Exists() {
pkg/llmproxy/translator/codex/openai/chat-completions/codex_openai_request.go: out, _ = sjson.Set(out, "reasoning.effort", v.Value())
pkg/llmproxy/translator/codex/openai/chat-completions/codex_openai_request.go- } else if v := gjson.GetBytes(rawJSON, "variant"); v.Exists() {
pkg/llmproxy/translator/codex/openai/chat-completions/codex_openai_request.go- effort := strings.ToLower(strings.TrimSpace(v.String()))
pkg/llmproxy/translator/codex/openai/chat-completions/codex_openai_request.go- if effort == "" {
pkg/llmproxy/translator/codex/openai/chat-completions/codex_openai_request.go: out, _ = sjson.Set(out, "reasoning.effort", "medium")
pkg/llmproxy/translator/codex/openai/chat-completions/codex_openai_request.go- } else {
pkg/llmproxy/translator/codex/openai/chat-completions/codex_openai_request.go: out, _ = sjson.Set(out, "reasoning.effort", effort)
pkg/llmproxy/translator/codex/openai/chat-completions/codex_openai_request.go- }
pkg/llmproxy/translator/codex/openai/chat-completions/codex_openai_request.go- } else {
pkg/llmproxy/translator/codex/openai/chat-completions/codex_openai_request.go: out, _ = sjson.Set(out, "reasoning.effort", "medium")
pkg/llmproxy/translator/codex/openai/chat-completions/codex_openai_request.go- }
pkg/llmproxy/translator/codex/openai/chat-completions/codex_openai_request.go- out, _ = sjson.Set(out, "parallel_tool_calls", true)
--
pkg/llmproxy/translator/codex/gemini/codex_gemini_request.go- out, _ = sjson.Set(out, "parallel_tool_calls", true)
pkg/llmproxy/translator/codex/gemini/codex_gemini_request.go-
pkg/llmproxy/translator/codex/gemini/codex_gemini_request.go: // Convert Gemini thinkingConfig to Codex reasoning.effort.
pkg/llmproxy/translator/codex/gemini/codex_gemini_request.go- // Note: Google official Python SDK sends snake_case fields (thinking_level/thinking_budget).
pkg/llmproxy/translator/codex/gemini/codex_gemini_request.go- effortSet := false
--
pkg/llmproxy/translator/codex/gemini/codex_gemini_request.go- effort := strings.ToLower(strings.TrimSpace(thinkingLevel.String()))
pkg/llmproxy/translator/codex/gemini/codex_gemini_request.go- if effort != "" {
pkg/llmproxy/translator/codex/gemini/codex_gemini_request.go: out, _ = sjson.Set(out, "reasoning.effort", effort)
pkg/llmproxy/translator/codex/gemini/codex_gemini_request.go- effortSet = true
pkg/llmproxy/translator/codex/gemini/codex_gemini_request.go- }
--
pkg/llmproxy/translator/codex/gemini/codex_gemini_request.go- if thinkingBudget.Exists() {
pkg/llmproxy/translator/codex/gemini/codex_gemini_request.go- if effort, ok := thinking.ConvertBudgetToLevel(int(thinkingBudget.Int())); ok {
pkg/llmproxy/translator/codex/gemini/codex_gemini_request.go: out, _ = sjson.Set(out, "reasoning.effort", effort)
pkg/llmproxy/translator/codex/gemini/codex_gemini_request.go- effortSet = true
pkg/llmproxy/translator/codex/gemini/codex_gemini_request.go- }
--
pkg/llmproxy/translator/codex/gemini/codex_gemini_request.go- if !effortSet {
pkg/llmproxy/translator/codex/gemini/codex_gemini_request.go- // No thinking config, set default effort
pkg/llmproxy/translator/codex/gemini/codex_gemini_request.go: out, _ = sjson.Set(out, "reasoning.effort", "medium")
pkg/llmproxy/translator/codex/gemini/codex_gemini_request.go- }
pkg/llmproxy/translator/codex/gemini/codex_gemini_request.go- out, _ = sjson.Set(out, "reasoning.summary", "auto")
--
pkg/llmproxy/translator/codex/openai/responses/codex_openai-responses_request.go- rawJSON, _ = sjson.SetBytes(rawJSON, "stream", true)
pkg/llmproxy/translator/codex/openai/responses/codex_openai-responses_request.go- rawJSON, _ = sjson.SetBytes(rawJSON, "store", false)
pkg/llmproxy/translator/codex/openai/responses/codex_openai-responses_request.go: // Map variant -> reasoning.effort when reasoning.effort is not explicitly provided.
pkg/llmproxy/translator/codex/openai/responses/codex_openai-responses_request.go: if !gjson.GetBytes(rawJSON, "reasoning.effort").Exists() {
pkg/llmproxy/translator/codex/openai/responses/codex_openai-responses_request.go- if variant := gjson.GetBytes(rawJSON, "variant"); variant.Exists() {
pkg/llmproxy/translator/codex/openai/responses/codex_openai-responses_request.go- effort := strings.ToLower(strings.TrimSpace(variant.String()))
pkg/llmproxy/translator/codex/openai/responses/codex_openai-responses_request.go- if effort != "" {
pkg/llmproxy/translator/codex/openai/responses/codex_openai-responses_request.go: rawJSON, _ = sjson.SetBytes(rawJSON, "reasoning.effort", effort)
pkg/llmproxy/translator/codex/openai/responses/codex_openai-responses_request.go- }
pkg/llmproxy/translator/codex/openai/responses/codex_openai-responses_request.go- }
--
pkg/llmproxy/thinking/apply_iflow_test.go- }{
pkg/llmproxy/thinking/apply_iflow_test.go- {
pkg/llmproxy/thinking/apply_iflow_test.go: name: "nested reasoning.effort maps to level",
pkg/llmproxy/thinking/apply_iflow_test.go- body: `{"reasoning":{"effort":"high"}}`,
pkg/llmproxy/thinking/apply_iflow_test.go- want: ThinkingConfig{Mode: ModeLevel, Level: LevelHigh},
pkg/llmproxy/thinking/apply_iflow_test.go- },
pkg/llmproxy/thinking/apply_iflow_test.go- {
pkg/llmproxy/thinking/apply_iflow_test.go: name: "literal reasoning.effort key maps to level",
pkg/llmproxy/thinking/apply_iflow_test.go: body: `{"reasoning.effort":"high"}`,
pkg/llmproxy/thinking/apply_iflow_test.go- want: ThinkingConfig{Mode: ModeLevel, Level: LevelHigh},
pkg/llmproxy/thinking/apply_iflow_test.go- },
pkg/llmproxy/thinking/apply_iflow_test.go- {
pkg/llmproxy/thinking/apply_iflow_test.go- name: "none maps to disabled",
pkg/llmproxy/thinking/apply_iflow_test.go: body: `{"reasoning.effort":"none"}`,
pkg/llmproxy/thinking/apply_iflow_test.go- want: ThinkingConfig{Mode: ModeNone, Budget: 0},
pkg/llmproxy/thinking/apply_iflow_test.go- },
--
pkg/llmproxy/thinking/strip.go- paths = []string{"reasoning_effort"}
pkg/llmproxy/thinking/strip.go- case "codex":
pkg/llmproxy/thinking/strip.go: paths = []string{"reasoning.effort"}
pkg/llmproxy/thinking/strip.go- case "iflow":
pkg/llmproxy/thinking/strip.go- paths = []string{
--
pkg/llmproxy/thinking/strip_test.go- t.Fatalf("expected reasoning to be removed")
pkg/llmproxy/thinking/strip_test.go- }
pkg/llmproxy/thinking/strip_test.go: if res.Get("reasoning.effort").Exists() {
pkg/llmproxy/thinking/strip_test.go: t.Fatalf("expected reasoning.effort to be removed")
pkg/llmproxy/thinking/strip_test.go- }
pkg/llmproxy/thinking/strip_test.go- if res.Get("reasoning_split").Exists() {
--
pkg/llmproxy/thinking/provider/codex/apply.go-// Package codex implements thinking configuration for Codex (OpenAI Responses API) models.
pkg/llmproxy/thinking/provider/codex/apply.go-//
pkg/llmproxy/thinking/provider/codex/apply.go:// Codex models use the reasoning.effort format with discrete levels
pkg/llmproxy/thinking/provider/codex/apply.go-// (low/medium/high). This is similar to OpenAI but uses nested field
pkg/llmproxy/thinking/provider/codex/apply.go:// "reasoning.effort" instead of "reasoning_effort".
pkg/llmproxy/thinking/provider/codex/apply.go-// See: _bmad-output/planning-artifacts/architecture.md#Epic-8
pkg/llmproxy/thinking/provider/codex/apply.go-package codex
--
pkg/llmproxy/thinking/provider/codex/apply.go-//
pkg/llmproxy/thinking/provider/codex/apply.go-// Codex-specific behavior:
pkg/llmproxy/thinking/provider/codex/apply.go:// - Output format: reasoning.effort (string: low/medium/high/xhigh)
pkg/llmproxy/thinking/provider/codex/apply.go-// - Level-only mode: no numeric budget support
pkg/llmproxy/thinking/provider/codex/apply.go-// - Some models support ZeroAllowed (gpt-5.1, gpt-5.2)
--
pkg/llmproxy/thinking/provider/codex/apply.go-
pkg/llmproxy/thinking/provider/codex/apply.go- if config.Mode == thinking.ModeLevel {
pkg/llmproxy/thinking/provider/codex/apply.go: result, _ := sjson.SetBytes(body, "reasoning.effort", string(config.Level))
pkg/llmproxy/thinking/provider/codex/apply.go- return result, nil
pkg/llmproxy/thinking/provider/codex/apply.go- }
--
pkg/llmproxy/thinking/provider/codex/apply.go- }
pkg/llmproxy/thinking/provider/codex/apply.go-
pkg/llmproxy/thinking/provider/codex/apply.go: result, _ := sjson.SetBytes(body, "reasoning.effort", effort)
pkg/llmproxy/thinking/provider/codex/apply.go- return result, nil
pkg/llmproxy/thinking/provider/codex/apply.go-}
--
pkg/llmproxy/thinking/provider/codex/apply.go- }
pkg/llmproxy/thinking/provider/codex/apply.go-
pkg/llmproxy/thinking/provider/codex/apply.go: result, _ := sjson.SetBytes(body, "reasoning.effort", effort)
pkg/llmproxy/thinking/provider/codex/apply.go- return result, nil
pkg/llmproxy/thinking/provider/codex/apply.go-}
--
pkg/llmproxy/thinking/provider/iflow/apply_test.go- t.Fatalf("expected reasoning object to be removed")
pkg/llmproxy/thinking/provider/iflow/apply_test.go- }
pkg/llmproxy/thinking/provider/iflow/apply_test.go: if res.Get("reasoning.effort").Exists() {
pkg/llmproxy/thinking/provider/iflow/apply_test.go: t.Fatalf("expected reasoning.effort to be removed")
pkg/llmproxy/thinking/provider/iflow/apply_test.go- }
pkg/llmproxy/thinking/provider/iflow/apply_test.go- if res.Get("variant").Exists() {
--
pkg/llmproxy/translator/claude/openai/responses/claude_openai-responses_request.go- root := gjson.ParseBytes(rawJSON)
pkg/llmproxy/translator/claude/openai/responses/claude_openai-responses_request.go-
pkg/llmproxy/translator/claude/openai/responses/claude_openai-responses_request.go: // Convert OpenAI Responses reasoning.effort to Claude thinking config.
pkg/llmproxy/translator/claude/openai/responses/claude_openai-responses_request.go: if v := root.Get("reasoning.effort"); v.Exists() {
pkg/llmproxy/translator/claude/openai/responses/claude_openai-responses_request.go- effort := strings.ToLower(strings.TrimSpace(v.String()))
pkg/llmproxy/translator/claude/openai/responses/claude_openai-responses_request.go- if effort != "" {
--
pkg/llmproxy/thinking/apply.go-//
pkg/llmproxy/thinking/apply.go-// Codex API format (OpenAI Responses API):
pkg/llmproxy/thinking/apply.go:// - reasoning.effort: "none", "low", "medium", "high"
pkg/llmproxy/thinking/apply.go-//
pkg/llmproxy/thinking/apply.go:// This is similar to OpenAI but uses nested field "reasoning.effort" instead of "reasoning_effort".
pkg/llmproxy/thinking/apply.go-func extractCodexConfig(body []byte) ThinkingConfig {
pkg/llmproxy/thinking/apply.go: // Check reasoning.effort (Codex / OpenAI Responses API format)
pkg/llmproxy/thinking/apply.go: if effort := gjson.GetBytes(body, "reasoning.effort"); effort.Exists() {
pkg/llmproxy/thinking/apply.go- value := effort.String()
pkg/llmproxy/thinking/apply.go- if value == "none" {
--
pkg/llmproxy/thinking/apply.go-
pkg/llmproxy/thinking/apply.go- // Compatibility fallback: some clients send Claude-style `variant`
pkg/llmproxy/thinking/apply.go: // instead of OpenAI/Codex `reasoning.effort`.
pkg/llmproxy/thinking/apply.go- if variant := gjson.GetBytes(body, "variant"); variant.Exists() {
pkg/llmproxy/thinking/apply.go- switch strings.ToLower(strings.TrimSpace(variant.String())) {
--
pkg/llmproxy/thinking/apply.go- }
pkg/llmproxy/thinking/apply.go-
pkg/llmproxy/thinking/apply.go: if effort := gjson.GetBytes(body, "reasoning.effort"); effort.Exists() {
pkg/llmproxy/thinking/apply.go- if config, ok := parseIFlowReasoningEffort(strings.TrimSpace(effort.String())); ok {
pkg/llmproxy/thinking/apply.go- return config
--
pkg/llmproxy/translator/codex/claude/codex_claude_request.go- template, _ = sjson.Set(template, "parallel_tool_calls", true)
pkg/llmproxy/translator/codex/claude/codex_claude_request.go-
pkg/llmproxy/translator/codex/claude/codex_claude_request.go: // Convert thinking.budget_tokens to reasoning.effort.
pkg/llmproxy/translator/codex/claude/codex_claude_request.go- reasoningEffort := "medium"
pkg/llmproxy/translator/codex/claude/codex_claude_request.go- if thinkingConfig := rootResult.Get("thinking"); thinkingConfig.Exists() && thinkingConfig.IsObject() {
--
pkg/llmproxy/translator/codex/claude/codex_claude_request.go- }
pkg/llmproxy/translator/codex/claude/codex_claude_request.go- }
pkg/llmproxy/translator/codex/claude/codex_claude_request.go: template, _ = sjson.Set(template, "reasoning.effort", reasoningEffort)
pkg/llmproxy/translator/codex/claude/codex_claude_request.go- template, _ = sjson.Set(template, "reasoning.summary", "auto")
pkg/llmproxy/translator/codex/claude/codex_claude_request.go- template, _ = sjson.Set(template, "stream", true)
--
pkg/llmproxy/executor/codex_executor_cpb0106_test.go- t.Fatalf("expected upstream stream=true, got %v", out.Bool())
pkg/llmproxy/executor/codex_executor_cpb0106_test.go- }
pkg/llmproxy/executor/codex_executor_cpb0106_test.go: if got := gjson.GetBytes(upstreamBody, "reasoning.effort").String(); got != "high" {
pkg/llmproxy/executor/codex_executor_cpb0106_test.go: t.Fatalf("expected reasoning.effort=high, got %q", got)
pkg/llmproxy/executor/codex_executor_cpb0106_test.go- }
pkg/llmproxy/executor/codex_executor_cpb0106_test.go-}
--
pkg/llmproxy/executor/codex_executor_cpb0106_test.go- t.Fatalf("expected upstream stream=false in ExecuteStream path, got %v", got)
pkg/llmproxy/executor/codex_executor_cpb0106_test.go- }
pkg/llmproxy/executor/codex_executor_cpb0106_test.go: if got := gjson.GetBytes(upstreamBody, "reasoning.effort").String(); got != "high" {
pkg/llmproxy/executor/codex_executor_cpb0106_test.go: t.Fatalf("expected reasoning.effort=high, got %q", got)
pkg/llmproxy/executor/codex_executor_cpb0106_test.go- }
pkg/llmproxy/executor/codex_executor_cpb0106_test.go-}
--
pkg/llmproxy/translator/openai/openai/responses/openai_openai-responses_request_test.go- "model":"gpt-4.1",
pkg/llmproxy/translator/openai/openai/responses/openai_openai-responses_request_test.go- "input":[{"type":"message","role":"user","content":[{"type":"input_text","text":"ping"}]}],
pkg/llmproxy/translator/openai/openai/responses/openai_openai-responses_request_test.go: "reasoning.effort":"low"
pkg/llmproxy/translator/openai/openai/responses/openai_openai-responses_request_test.go- }`)
pkg/llmproxy/translator/openai/openai/responses/openai_openai-responses_request_test.go-
Script executed:
cat -n pkg/llmproxy/thinking/apply_iflow_test.go | head -50Repository: KooshaPari/cliproxyapi-plusplus
Repository: KooshaPari/cliproxyapi-plusplus
Exit code: 0
stdout:
1 package thinking
2
3 import "testing"
4
5 func TestExtractIFlowConfig_ReasoningEffort(t *testing.T) {
6 tests := []struct {
7 name string
8 body string
9 want ThinkingConfig
10 }{
11 {
12 name: "nested reasoning.effort maps to level",
13 body: `{"reasoning":{"effort":"high"}}`,
14 want: ThinkingConfig{Mode: ModeLevel, Level: LevelHigh},
15 },
16 {
17 name: "literal reasoning.effort key maps to level",
18 body: `{"reasoning.effort":"high"}`,
19 want: ThinkingConfig{Mode: ModeLevel, Level: LevelHigh},
20 },
21 {
22 name: "none maps to disabled",
23 body: `{"reasoning.effort":"none"}`,
24 want: ThinkingConfig{Mode: ModeNone, Budget: 0},
25 },
26 }
27
28 for _, tt := range tests {
29 t.Run(tt.name, func(t *testing.T) {
30 got := extractIFlowConfig([]byte(tt.body))
31 if got != tt.want {
32 t.Fatalf("got=%+v want=%+v", got, tt.want)
33 }
34 })
35 }
36 }
37
38 func TestExtractIFlowConfig_ReasoningObjectEffort(t *testing.T) {
39 tests := []struct {
40 name string
41 body string
42 want ThinkingConfig
43 }{
44 {
45 name: "reasoning object effort maps to level",
46 body: `{"reasoning":{"effort":"medium"}}`,
47 want: ThinkingConfig{Mode: ModeLevel, Level: LevelMedium},
48 },
49 {
50 name: "empty effort falls back to empty config",
Script executed:
rg "reasoning\.effort" pkg/llmproxy/thinking/ --type go -B 3 -A 3Repository: KooshaPari/cliproxyapi-plusplus
Repository: KooshaPari/cliproxyapi-plusplus
Exit code: 0
stdout:
pkg/llmproxy/thinking/apply.go-// extractCodexConfig extracts thinking configuration from Codex format request body.
pkg/llmproxy/thinking/apply.go-//
pkg/llmproxy/thinking/apply.go-// Codex API format (OpenAI Responses API):
pkg/llmproxy/thinking/apply.go:// - reasoning.effort: "none", "low", "medium", "high"
pkg/llmproxy/thinking/apply.go-//
pkg/llmproxy/thinking/apply.go:// This is similar to OpenAI but uses nested field "reasoning.effort" instead of "reasoning_effort".
pkg/llmproxy/thinking/apply.go-func extractCodexConfig(body []byte) ThinkingConfig {
pkg/llmproxy/thinking/apply.go: // Check reasoning.effort (Codex / OpenAI Responses API format)
pkg/llmproxy/thinking/apply.go: if effort := gjson.GetBytes(body, "reasoning.effort"); effort.Exists() {
pkg/llmproxy/thinking/apply.go- value := effort.String()
pkg/llmproxy/thinking/apply.go- if value == "none" {
pkg/llmproxy/thinking/apply.go- return ThinkingConfig{Mode: ModeNone, Budget: 0}
--
pkg/llmproxy/thinking/apply.go- }
pkg/llmproxy/thinking/apply.go-
pkg/llmproxy/thinking/apply.go- // Compatibility fallback: some clients send Claude-style `variant`
pkg/llmproxy/thinking/apply.go: // instead of OpenAI/Codex `reasoning.effort`.
pkg/llmproxy/thinking/apply.go- if variant := gjson.GetBytes(body, "variant"); variant.Exists() {
pkg/llmproxy/thinking/apply.go- switch strings.ToLower(strings.TrimSpace(variant.String())) {
pkg/llmproxy/thinking/apply.go- case "none":
--
pkg/llmproxy/thinking/apply.go- return ThinkingConfig{Mode: ModeNone, Budget: 0}
pkg/llmproxy/thinking/apply.go- }
pkg/llmproxy/thinking/apply.go-
pkg/llmproxy/thinking/apply.go: if effort := gjson.GetBytes(body, "reasoning.effort"); effort.Exists() {
pkg/llmproxy/thinking/apply.go- if config, ok := parseIFlowReasoningEffort(strings.TrimSpace(effort.String())); ok {
pkg/llmproxy/thinking/apply.go- return config
pkg/llmproxy/thinking/apply.go- }
--
pkg/llmproxy/thinking/strip_test.go- if res.Get("reasoning").Exists() {
pkg/llmproxy/thinking/strip_test.go- t.Fatalf("expected reasoning to be removed")
pkg/llmproxy/thinking/strip_test.go- }
pkg/llmproxy/thinking/strip_test.go: if res.Get("reasoning.effort").Exists() {
pkg/llmproxy/thinking/strip_test.go: t.Fatalf("expected reasoning.effort to be removed")
pkg/llmproxy/thinking/strip_test.go- }
pkg/llmproxy/thinking/strip_test.go- if res.Get("reasoning_split").Exists() {
pkg/llmproxy/thinking/strip_test.go- t.Fatalf("expected reasoning_split to be removed")
--
pkg/llmproxy/thinking/strip.go- case "openai":
pkg/llmproxy/thinking/strip.go- paths = []string{"reasoning_effort"}
pkg/llmproxy/thinking/strip.go- case "codex":
pkg/llmproxy/thinking/strip.go: paths = []string{"reasoning.effort"}
pkg/llmproxy/thinking/strip.go- case "iflow":
pkg/llmproxy/thinking/strip.go- paths = []string{
pkg/llmproxy/thinking/strip.go- "chat_template_kwargs.enable_thinking",
--
pkg/llmproxy/thinking/apply_iflow_test.go- want ThinkingConfig
pkg/llmproxy/thinking/apply_iflow_test.go- }{
pkg/llmproxy/thinking/apply_iflow_test.go- {
pkg/llmproxy/thinking/apply_iflow_test.go: name: "nested reasoning.effort maps to level",
pkg/llmproxy/thinking/apply_iflow_test.go- body: `{"reasoning":{"effort":"high"}}`,
pkg/llmproxy/thinking/apply_iflow_test.go- want: ThinkingConfig{Mode: ModeLevel, Level: LevelHigh},
pkg/llmproxy/thinking/apply_iflow_test.go- },
pkg/llmproxy/thinking/apply_iflow_test.go- {
pkg/llmproxy/thinking/apply_iflow_test.go: name: "literal reasoning.effort key maps to level",
pkg/llmproxy/thinking/apply_iflow_test.go: body: `{"reasoning.effort":"high"}`,
pkg/llmproxy/thinking/apply_iflow_test.go- want: ThinkingConfig{Mode: ModeLevel, Level: LevelHigh},
pkg/llmproxy/thinking/apply_iflow_test.go- },
pkg/llmproxy/thinking/apply_iflow_test.go- {
pkg/llmproxy/thinking/apply_iflow_test.go- name: "none maps to disabled",
pkg/llmproxy/thinking/apply_iflow_test.go: body: `{"reasoning.effort":"none"}`,
pkg/llmproxy/thinking/apply_iflow_test.go- want: ThinkingConfig{Mode: ModeNone, Budget: 0},
pkg/llmproxy/thinking/apply_iflow_test.go- },
pkg/llmproxy/thinking/apply_iflow_test.go- }
--
pkg/llmproxy/thinking/provider/codex/apply.go-// Package codex implements thinking configuration for Codex (OpenAI Responses API) models.
pkg/llmproxy/thinking/provider/codex/apply.go-//
pkg/llmproxy/thinking/provider/codex/apply.go:// Codex models use the reasoning.effort format with discrete levels
pkg/llmproxy/thinking/provider/codex/apply.go-// (low/medium/high). This is similar to OpenAI but uses nested field
pkg/llmproxy/thinking/provider/codex/apply.go:// "reasoning.effort" instead of "reasoning_effort".
pkg/llmproxy/thinking/provider/codex/apply.go-// See: _bmad-output/planning-artifacts/architecture.md#Epic-8
pkg/llmproxy/thinking/provider/codex/apply.go-package codex
pkg/llmproxy/thinking/provider/codex/apply.go-
--
pkg/llmproxy/thinking/provider/codex/apply.go-// Applier implements thinking.ProviderApplier for Codex models.
pkg/llmproxy/thinking/provider/codex/apply.go-//
pkg/llmproxy/thinking/provider/codex/apply.go-// Codex-specific behavior:
pkg/llmproxy/thinking/provider/codex/apply.go:// - Output format: reasoning.effort (string: low/medium/high/xhigh)
pkg/llmproxy/thinking/provider/codex/apply.go-// - Level-only mode: no numeric budget support
pkg/llmproxy/thinking/provider/codex/apply.go-// - Some models support ZeroAllowed (gpt-5.1, gpt-5.2)
pkg/llmproxy/thinking/provider/codex/apply.go-type Applier struct{}
--
pkg/llmproxy/thinking/provider/codex/apply.go- }
pkg/llmproxy/thinking/provider/codex/apply.go-
pkg/llmproxy/thinking/provider/codex/apply.go- if config.Mode == thinking.ModeLevel {
pkg/llmproxy/thinking/provider/codex/apply.go: result, _ := sjson.SetBytes(body, "reasoning.effort", string(config.Level))
pkg/llmproxy/thinking/provider/codex/apply.go- return result, nil
pkg/llmproxy/thinking/provider/codex/apply.go- }
pkg/llmproxy/thinking/provider/codex/apply.go-
--
pkg/llmproxy/thinking/provider/codex/apply.go- return body, nil
pkg/llmproxy/thinking/provider/codex/apply.go- }
pkg/llmproxy/thinking/provider/codex/apply.go-
pkg/llmproxy/thinking/provider/codex/apply.go: result, _ := sjson.SetBytes(body, "reasoning.effort", effort)
pkg/llmproxy/thinking/provider/codex/apply.go- return result, nil
pkg/llmproxy/thinking/provider/codex/apply.go-}
pkg/llmproxy/thinking/provider/codex/apply.go-
--
pkg/llmproxy/thinking/provider/codex/apply.go- return body, nil
pkg/llmproxy/thinking/provider/codex/apply.go- }
pkg/llmproxy/thinking/provider/codex/apply.go-
pkg/llmproxy/thinking/provider/codex/apply.go: result, _ := sjson.SetBytes(body, "reasoning.effort", effort)
pkg/llmproxy/thinking/provider/codex/apply.go- return result, nil
pkg/llmproxy/thinking/provider/codex/apply.go-}
pkg/llmproxy/thinking/provider/codex/apply.go-
--
pkg/llmproxy/thinking/provider/iflow/apply_test.go- if res.Get("reasoning").Exists() {
pkg/llmproxy/thinking/provider/iflow/apply_test.go- t.Fatalf("expected reasoning object to be removed")
pkg/llmproxy/thinking/provider/iflow/apply_test.go- }
pkg/llmproxy/thinking/provider/iflow/apply_test.go: if res.Get("reasoning.effort").Exists() {
pkg/llmproxy/thinking/provider/iflow/apply_test.go: t.Fatalf("expected reasoning.effort to be removed")
pkg/llmproxy/thinking/provider/iflow/apply_test.go- }
pkg/llmproxy/thinking/provider/iflow/apply_test.go- if res.Get("variant").Exists() {
pkg/llmproxy/thinking/provider/iflow/apply_test.go- t.Fatalf("expected variant to be removed")
Script executed:
cat > /tmp/test_sjson.go << 'EOF'
package main
import (
"fmt"
"github.com/tidwall/gjson"
"github.com/tidwall/sjson"
)
func main() {
// Test 1: nested path deletion
input1 := []byte(`{"reasoning":{"effort":"high"},"foo":"bar"}`)
result1, _ := sjson.DeleteBytes(input1, "reasoning.effort")
fmt.Println("Nested delete result:", string(result1))
fmt.Println("After nested delete, reasoning exists:", gjson.GetBytes(result1, "reasoning").Exists())
fmt.Println("After nested delete, reasoning.effort exists:", gjson.GetBytes(result1, "reasoning.effort").Exists())
// Test 2: literal dot-key without escaping
input2 := []byte(`{"reasoning.effort":"high","foo":"bar"}`)
result2, _ := sjson.DeleteBytes(input2, "reasoning.effort")
fmt.Println("\nLiteral key (no escape) delete result:", string(result2))
fmt.Println("After literal delete, reasoning.effort exists:", gjson.GetBytes(result2, `reasoning\.effort`).Exists())
// Test 3: literal dot-key WITH escaping
input3 := []byte(`{"reasoning.effort":"high","foo":"bar"}`)
result3, _ := sjson.DeleteBytes(input3, `reasoning\.effort`)
fmt.Println("\nLiteral key (with escape) delete result:", string(result3))
fmt.Println("After escaped delete, reasoning.effort exists:", gjson.GetBytes(result3, `reasoning\.effort`).Exists())
}
EOF
cd /tmp && go mod init test 2>/dev/null; go get github.com/tidwall/sjson 2>/dev/null; go run test_sjson.goRepository: KooshaPari/cliproxyapi-plusplus
Repository: KooshaPari/cliproxyapi-plusplus
Exit code: 1
stderr:
go: warning: ignoring go.mod in system temp root /tmp
test_sjson.go:5:2: no required module provides package github.com/tidwall/gjson: go.mod file not found in current directory or any parent directory; see 'go help modules'
test_sjson.go:6:2: no required module provides package github.com/tidwall/sjson: go.mod file not found in current directory or any parent directory; see 'go help modules'
Script executed:
cat -n pkg/llmproxy/thinking/strip_test.goRepository: KooshaPari/cliproxyapi-plusplus
Repository: KooshaPari/cliproxyapi-plusplus
Exit code: 0
stdout:
1 package thinking
2
3 import (
4 "testing"
5
6 "github.com/tidwall/gjson"
7 )
8
9 func TestStripThinkingConfigIflow(t *testing.T) {
10 body := []byte(`{
11 "model":"minimax-m2.5",
12 "reasoning":{"effort":"high"},
13 "reasoning_effort":"high",
14 "reasoning_split":true,
15 "variant":"medium",
16 "chat_template_kwargs":{"enable_thinking":true,"clear_thinking":false},
17 "messages":[{"role":"user","content":"hi"}]
18 }`)
19
20 out := StripThinkingConfig(body, "iflow")
21 res := gjson.ParseBytes(out)
22
23 if res.Get("reasoning").Exists() {
24 t.Fatalf("expected reasoning to be removed")
25 }
26 if res.Get("reasoning.effort").Exists() {
27 t.Fatalf("expected reasoning.effort to be removed")
28 }
29 if res.Get("reasoning_split").Exists() {
30 t.Fatalf("expected reasoning_split to be removed")
31 }
32 if res.Get("reasoning_effort").Exists() {
33 t.Fatalf("expected reasoning_effort to be removed")
34 }
35 if res.Get("variant").Exists() {
36 t.Fatalf("expected variant to be removed")
37 }
38 if res.Get("chat_template_kwargs").Exists() {
39 t.Fatalf("expected chat_template_kwargs to be removed")
40 }
41 if res.Get("messages.0.content").String() != "hi" {
42 t.Fatalf("expected unrelated messages to remain")
43 }
44 }
45
46 func TestStripThinkingConfigIflowPreservesUnrelatedChatTemplateKwargs(t *testing.T) {
47 body := []byte(`{
48 "chat_template_kwargs":{
49 "enable_thinking":true,
50 "clear_thinking":false,
51 "system_prompt":"keep-me"
52 },
53 "messages":[{"role":"user","content":"hi"}]
54 }`)
55
56 out := StripThinkingConfig(body, "iflow")
57 res := gjson.ParseBytes(out)
58
59 if res.Get("chat_template_kwargs.enable_thinking").Exists() {
60 t.Fatalf("expected enable_thinking to be removed")
61 }
62 if res.Get("chat_template_kwargs.clear_thinking").Exists() {
63 t.Fatalf("expected clear_thinking to be removed")
64 }
65 if res.Get("chat_template_kwargs.system_prompt").String() != "keep-me" {
66 t.Fatalf("expected unrelated chat_template_kwargs fields to remain")
67 }
68 if res.Get("messages.0.content").String() != "hi" {
69 t.Fatalf("expected unrelated messages to remain")
70 }
71 }
Script executed:
rg "reasoning\.effort" pkg/llmproxy/thinking/strip_test.goRepository: KooshaPari/cliproxyapi-plusplus
Repository: KooshaPari/cliproxyapi-plusplus
Exit code: 0
stdout:
if res.Get("reasoning.effort").Exists() {
t.Fatalf("expected reasoning.effort to be removed")
Script executed:
python3 << 'PYEOF'
# Let's verify the sjson behavior with literal dot-keys
# by checking how iflow extraction tests work
import json
# From apply_iflow_test.go line 17-18:
test_body = '{"reasoning.effort":"high"}'
data = json.loads(test_body)
print("Parsed JSON with literal dot-key:")
print(json.dumps(data, indent=2))
print("\nKey in dict:", "reasoning.effort" in data)
print("Key names:", list(data.keys()))
PYEOFRepository: KooshaPari/cliproxyapi-plusplus
Repository: KooshaPari/cliproxyapi-plusplus
Exit code: 0
stdout:
Parsed JSON with literal dot-key:
{
"reasoning.effort": "high"
}
Key in dict: True
Key names: ['reasoning.effort']
Script executed:
rg -A 5 "TestStripThinkingConfigIflow\(" pkg/llmproxy/thinking/strip_test.go | head -20Repository: KooshaPari/cliproxyapi-plusplus
Repository: KooshaPari/cliproxyapi-plusplus
Exit code: 0
stdout:
func TestStripThinkingConfigIflow(t *testing.T) {
body := []byte(`{
"model":"minimax-m2.5",
"reasoning":{"effort":"high"},
"reasoning_effort":"high",
"reasoning_split":true,
Script executed:
# Check if any test input to strip has a literal reasoning.effort key
rg '"reasoning\.effort"' pkg/llmproxy/thinking/strip_test.goRepository: KooshaPari/cliproxyapi-plusplus
Repository: KooshaPari/cliproxyapi-plusplus
Exit code: 0
stdout:
if res.Get("reasoning.effort").Exists() {
Script executed:
sed -n '40,65p' pkg/llmproxy/thinking/strip.goRepository: KooshaPari/cliproxyapi-plusplus
Repository: KooshaPari/cliproxyapi-plusplus
Exit code: 0
stdout:
case "codex":
paths = []string{"reasoning.effort"}
case "iflow":
paths = []string{
"chat_template_kwargs.enable_thinking",
"chat_template_kwargs.clear_thinking",
"reasoning",
"reasoning_split",
"reasoning_effort",
"variant",
}
default:
return body
}
result := body
for _, path := range paths {
result, _ = sjson.DeleteBytes(result, path)
}
if provider == "iflow" {
if kwargs := gjson.GetBytes(result, "chat_template_kwargs"); kwargs.Exists() && len(kwargs.Map()) == 0 {
result, _ = sjson.DeleteBytes(result, "chat_template_kwargs")
}
}
return result
}
Add escaped literal key path "reasoning\.effort" to iflow stripping.
The iflow case removes "reasoning" (nested) but not a literal top-level "reasoning.effort" key. Since iflow inputs can contain this literal dot-key (as shown in apply_iflow_test.go), it must be stripped to prevent unexpected payload shapes. Add the escaped path to the paths list.
Proposed fix
case "iflow":
paths = []string{
"chat_template_kwargs.enable_thinking",
"chat_template_kwargs.clear_thinking",
"reasoning",
"reasoning_split",
"reasoning_effort",
+ "reasoning\.effort",
"variant",
}Consider centralizing iflow cleanup paths shared with pkg/llmproxy/thinking/provider/iflow/apply.go to prevent future drift.
📝 Committable suggestion
‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.
Suggested change
| paths = []string{ | |
| "chat_template_kwargs.enable_thinking", | |
| "chat_template_kwargs.clear_thinking", | |
| "reasoning", | |
| "reasoning_split", | |
| "reasoning_effort", | |
| "variant", | |
| } | |
| paths = []string{ | |
| "chat_template_kwargs.enable_thinking", | |
| "chat_template_kwargs.clear_thinking", | |
| "reasoning", | |
| "reasoning_split", | |
| "reasoning_effort", | |
| "reasoning\.effort", | |
| "variant", | |
| } |
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed.
In `@pkg/llmproxy/thinking/strip.go` around lines 43 - 50, The iflow stripping
list in pkg/llmproxy/thinking/strip.go currently removes nested "reasoning" but
not a literal top-level key "reasoning.effort"; update the paths slice (the
variable assigned to paths in strip.go) to include the escaped literal key
"reasoning\.effort" so top-level dot-keys are stripped as well, and consider
moving this paths list into a shared constant used by the iflow cleanup code in
provider/iflow/apply.go to avoid drift.
Comment on lines
+45
to
+50
| registry.GetGlobalRegistry().RegisterClient("endpoint-compat-iflow-direction", "iflow", []*registry.ModelInfo{{ | ||
| ID: "minimax-m2.5-directionality-test", | ||
| SupportedEndpoints: []string{"/chat/completions"}, | ||
| }}) | ||
| t.Cleanup(func() { | ||
| registry.GetGlobalRegistry().UnregisterClient("endpoint-compat-iflow-directionality") |
There was a problem hiding this comment.
Use the same client ID in cleanup.
Line 45 registers "endpoint-compat-iflow-direction", but Line 50 unregisters "endpoint-compat-iflow-directionality". That leaves the model in the global registry and makes later tests order-dependent.
🧹 Fix the leaked registry entry
t.Cleanup(func() {
- registry.GetGlobalRegistry().UnregisterClient("endpoint-compat-iflow-directionality")
+ registry.GetGlobalRegistry().UnregisterClient("endpoint-compat-iflow-direction")
})🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed.
In `@sdk/api/handlers/openai/endpoint_compat_test.go` around lines 45 - 50, The
cleanup is unregistering a different client ID than the one registered, leaking
the registry entry; update the UnregisterClient call so it uses the exact same
client ID string used in RegisterClient
("endpoint-compat-iflow-direction")—i.e., locate the register call via
registry.GetGlobalRegistry().RegisterClient(...) and make the corresponding
registry.GetGlobalRegistry().UnregisterClient(...) use that identical ID.
Comment on lines
+186
to
+193
| m.mu.Lock() | ||
| defer m.mu.Unlock() | ||
| if replaced, ok := m.executors[key]; ok { | ||
| if closer, okCloser := replaced.(ExecutionSessionCloser); okCloser { | ||
| closer.CloseExecutionSession(CloseAllExecutionSessionsID) | ||
| } | ||
| } | ||
| m.executors[key] = executor |
There was a problem hiding this comment.
Move CloseExecutionSession outside m.mu.
CloseExecutionSession is external code. Calling it while holding the manager write lock can deadlock if the executor re-enters the manager, and it blocks every auth/executor read until the close returns.
Possible fix
m.mu.Lock()
- defer m.mu.Unlock()
- if replaced, ok := m.executors[key]; ok {
- if closer, okCloser := replaced.(ExecutionSessionCloser); okCloser {
- closer.CloseExecutionSession(CloseAllExecutionSessionsID)
- }
- }
- m.executors[key] = executor
+ replaced := m.executors[key]
+ m.executors[key] = executor
+ m.mu.Unlock()
+ if closer, ok := replaced.(ExecutionSessionCloser); ok {
+ closer.CloseExecutionSession(CloseAllExecutionSessionsID)
+ }📝 Committable suggestion
‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.
Suggested change
| m.mu.Lock() | |
| defer m.mu.Unlock() | |
| if replaced, ok := m.executors[key]; ok { | |
| if closer, okCloser := replaced.(ExecutionSessionCloser); okCloser { | |
| closer.CloseExecutionSession(CloseAllExecutionSessionsID) | |
| } | |
| } | |
| m.executors[key] = executor | |
| m.mu.Lock() | |
| replaced := m.executors[key] | |
| m.executors[key] = executor | |
| m.mu.Unlock() | |
| if closer, ok := replaced.(ExecutionSessionCloser); ok { | |
| closer.CloseExecutionSession(CloseAllExecutionSessionsID) | |
| } |
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed.
In `@sdk/cliproxy/auth/manager_compat.go` around lines 186 - 193, The manager
currently calls CloseExecutionSession while holding m.mu which can deadlock;
modify Set/replace logic so you capture the replaced executor (m.executors[key])
while holding m.mu, then release the lock and call
CloseExecutionSession(CloseAllExecutionSessionsID) on it outside the lock;
specifically, in the function that assigns m.executors[key] (the block that
checks for replaced, ok := m.executors[key] and type-asserts to
ExecutionSessionCloser), store the replaced value in a local variable while
locked, perform m.executors[key] = executor, unlock (defer removal of defer if
needed), and only after the mutex is released, if the local variable implements
ExecutionSessionCloser call
closer.CloseExecutionSession(CloseAllExecutionSessionsID).
Comment on lines
+428
to
+434
| func (m *Manager) ExecuteStream(ctx context.Context, providers []string, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (*cliproxyexecutor.StreamResult, error) { | ||
| auth, executor, err := m.selectAuthAndExecutor(ctx, providers, req.Model, opts) | ||
| if err != nil { | ||
| return nil, err | ||
| } | ||
| req.Model = m.applyAPIKeyModelAlias(auth, req.Model) | ||
| return executor.ExecuteStream(ctx, auth, req, opts) |
There was a problem hiding this comment.
Record immediate stream startup failures.
Unlike Execute and ExecuteCount, this path returns ExecuteStream errors without calling recordExecutionResult, so quota/backoff state is never updated when stream creation fails.
Possible fix
req.Model = m.applyAPIKeyModelAlias(auth, req.Model)
- return executor.ExecuteStream(ctx, auth, req, opts)
+ stream, execErr := executor.ExecuteStream(ctx, auth, req, opts)
+ if execErr != nil {
+ m.recordExecutionResult(ctx, auth, req.Model, execErr, nil)
+ return nil, execErr
+ }
+ return stream, nil📝 Committable suggestion
‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.
Suggested change
| func (m *Manager) ExecuteStream(ctx context.Context, providers []string, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (*cliproxyexecutor.StreamResult, error) { | |
| auth, executor, err := m.selectAuthAndExecutor(ctx, providers, req.Model, opts) | |
| if err != nil { | |
| return nil, err | |
| } | |
| req.Model = m.applyAPIKeyModelAlias(auth, req.Model) | |
| return executor.ExecuteStream(ctx, auth, req, opts) | |
| func (m *Manager) ExecuteStream(ctx context.Context, providers []string, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (*cliproxyexecutor.StreamResult, error) { | |
| auth, executor, err := m.selectAuthAndExecutor(ctx, providers, req.Model, opts) | |
| if err != nil { | |
| return nil, err | |
| } | |
| req.Model = m.applyAPIKeyModelAlias(auth, req.Model) | |
| stream, execErr := executor.ExecuteStream(ctx, auth, req, opts) | |
| if execErr != nil { | |
| m.recordExecutionResult(ctx, auth, req.Model, execErr, nil) | |
| return nil, execErr | |
| } | |
| return stream, nil | |
| } |
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed.
In `@sdk/cliproxy/auth/manager_compat.go` around lines 428 - 434, The
ExecuteStream path currently returns executor.ExecuteStream errors without
updating quota/backoff state; modify Manager.ExecuteStream so after selecting
auth/executor (selectAuthAndExecutor) and applying applyAPIKeyModelAlias, you
call executor.ExecuteStream and if it returns an error immediately call
m.recordExecutionResult(...) with the same ctx, auth, req.Model (or the aliased
model) and an appropriate failure result to update quota/backoff state before
returning the error; keep the existing success return path unchanged so only
startup failures are recorded.
Comment on lines
+452
to
+487
| var candidates []*Auth | ||
| var executor ProviderExecutor | ||
| m.mu.RLock() | ||
| for _, provider := range providers { | ||
| exec := m.executors[strings.ToLower(strings.TrimSpace(provider))] | ||
| if exec == nil { | ||
| continue | ||
| } | ||
| for _, auth := range m.auths { | ||
| if auth == nil || !strings.EqualFold(auth.Provider, provider) { | ||
| continue | ||
| } | ||
| if pinnedAuthID != "" && auth.ID != pinnedAuthID { | ||
| continue | ||
| } | ||
| candidates = append(candidates, auth.Clone()) | ||
| } | ||
| if executor == nil { | ||
| executor = exec | ||
| } | ||
| } | ||
| selector := m.selector | ||
| m.mu.RUnlock() | ||
| if len(candidates) == 0 || executor == nil { | ||
| return nil, nil, &Error{Code: "auth_not_found", Message: "no auth candidates", HTTPStatus: http.StatusServiceUnavailable, Retryable: true} | ||
| } | ||
| auth, err := selector.Pick(ctx, providers[0], model, opts, candidates) | ||
| if err != nil { | ||
| return nil, nil, err | ||
| } | ||
| updateAggregatedAvailability(auth, now) | ||
| if selectedAuthCallback != nil && auth != nil { | ||
| selectedAuthCallback(auth.ID) | ||
| } | ||
| return auth, executor, nil | ||
| } |
There was a problem hiding this comment.
Keep auth selection and executor selection in the same provider scope.
candidates mixes auths from every requested provider, but executor is pinned to the first provider with an executor and Line 478 always passes providers[0] to the selector. With providers = ["gemini", "claude"], the selector can return a claude auth while Execute* still calls the gemini executor.
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed.
In `@sdk/cliproxy/auth/manager_compat.go` around lines 452 - 487, The selection
mixes auths from all providers into candidates but picks an executor only for
the first provider with an executor and always calls selector.Pick with
providers[0], causing provider mismatch; instead, for each provider iterate
building provider-scoped candidates from m.auths (using
strings.EqualFold(auth.Provider, provider) and pinnedAuthID filtering), get the
corresponding exec from m.executors, and if exec != nil and that provider has
candidates call selector.Pick(ctx, provider, model, opts, providerCandidates);
if selector.Pick returns an auth, call updateAggregatedAvailability and
selectedAuthCallback(auth.ID) and immediately return that auth and the matching
exec; only if no provider yields a pick return the auth_not_found error—this
ensures executor and auth come from the same provider (referencing candidates,
executor, m.executors, providers, selector.Pick, m.auths, pinnedAuthID,
updateAggregatedAvailability, selectedAuthCallback).
Comment on lines
+543
to
+574
| switch strings.ToLower(strings.TrimSpace(auth.Provider)) { | ||
| case "gemini": | ||
| for _, key := range cfg.GeminiKey { | ||
| if strings.TrimSpace(key.APIKey) == apiKey && (baseURL == "" || strings.TrimSpace(key.BaseURL) == baseURL) { | ||
| models := make([]modelAliasEntry, 0, len(key.Models)) | ||
| for _, model := range key.Models { | ||
| models = append(models, model) | ||
| } | ||
| return resolveModelAliasFromConfigModels(requestedModel, models) | ||
| } | ||
| } | ||
| case "claude": | ||
| for _, key := range cfg.ClaudeKey { | ||
| if strings.TrimSpace(key.APIKey) == apiKey && (baseURL == "" || strings.TrimSpace(key.BaseURL) == baseURL) { | ||
| models := make([]modelAliasEntry, 0, len(key.Models)) | ||
| for _, model := range key.Models { | ||
| models = append(models, model) | ||
| } | ||
| return resolveModelAliasFromConfigModels(requestedModel, models) | ||
| } | ||
| } | ||
| case "codex": | ||
| for _, key := range cfg.CodexKey { | ||
| if strings.TrimSpace(key.APIKey) == apiKey && (baseURL == "" || strings.TrimSpace(key.BaseURL) == baseURL) { | ||
| models := make([]modelAliasEntry, 0, len(key.Models)) | ||
| for _, model := range key.Models { | ||
| models = append(models, model) | ||
| } | ||
| return resolveModelAliasFromConfigModels(requestedModel, models) | ||
| } | ||
| } | ||
| } |
There was a problem hiding this comment.
🛠️ Refactor suggestion | 🟠 Major
Deduplicate provider-specific API-key alias lookup.
The three switch branches are the same scan with a different config slice, and this helper is already over 50 lines. Extract a provider registry/helper so new providers don't keep extending this hot path with copy/paste.
As per coding guidelines, "Build generic building blocks (provider interface + registry) before application logic" and "Maximum function length: 40 lines".
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed.
In `@sdk/cliproxy/auth/manager_compat.go` around lines 543 - 574, The switch on
strings.ToLower(strings.TrimSpace(auth.Provider)) duplicates the same
API-key/model lookup for cfg.GeminiKey, cfg.ClaudeKey and cfg.CodexKey; extract
a provider registry/helper and replace the three branches with a single generic
lookup. Implement a small provider registry (map[string]func(*Config)[]KeyType
or an interface) that maps "gemini"/"claude"/"codex" to functions returning the
corresponding cfg.<GeminiKey|ClaudeKey|CodexKey> slice, then in the current
function call that helper once: iterate the returned keys, compare
strings.TrimSpace(key.APIKey) and BaseURL, build models []modelAliasEntry and
call resolveModelAliasFromConfigModels(requestedModel, models). Update names
referenced here (auth.Provider, cfg.GeminiKey, cfg.ClaudeKey, cfg.CodexKey,
resolveModelAliasFromConfigModels, modelAliasEntry) and ensure the refactor
keeps the logic identical and reduces the function length below 40 lines.
Co-authored-by: Codex <noreply@openai.com>
Co-authored-by: Codex <noreply@openai.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary by CodeRabbit
New Features
Bug Fixes
Tests
Documentation