feat(agent): add optional plan-before-execute turn for complex codergen sessions by Copilot · Pull Request #137 · 2389-research/tracker

Copilot · 2026-04-21T16:57:30Z

Review new PR comment and identify actionable review threads
Investigate latest GitHub Actions runs/logs for any CI/build failures
Reproduce baseline locally (go build ./..., go test ./... -short)
Apply minimal fixes for requested review feedback in planning turn and config precedence
Run targeted tests for touched files
Run full validation (go build ./..., go test ./... -short) and parallel validation
Reply to PR comment with commit hash

Agent-Logs-Url: https://github.com/2389-research/tracker/sessions/13e98b22-007e-4fb3-9f06-d88c3546f4f5 Co-authored-by: clintecker <26985+clintecker@users.noreply.github.com>

Copilot

Pull request overview

Adds an opt-in “plan before execute” phase to the agent session loop to reduce thrash on multi-step codergen tasks by injecting a single planning-only LLM call before normal tool-using turns.

Changes:

Introduces SessionConfig.PlanBeforeExecute (default false) and wires a one-time planning call into Session.Run.
Adds codergen config mapping from graph/node attrs (plan_before_execute, plus node alias plan) and tests for the mapping.
Updates tests and changelog to cover the new planning behavior and defaults.

Reviewed changes

Copilot reviewed 8 out of 8 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
`agent/config.go`	Adds `PlanBeforeExecute` to `SessionConfig`.
`agent/config_test.go`	Verifies `PlanBeforeExecute` default is `false`.
`agent/session.go`	Runs the optional planning turn before the main turn loop.
`agent/session_run.go`	Implements planning prompt constants, planning-call path without tools, and context injection.
`agent/session_test.go`	Adds unit tests ensuring planning call is injected/omitted correctly and tools are suppressed during planning.
`pipeline/handlers/codergen.go`	Maps pipeline attrs into `SessionConfig.PlanBeforeExecute`.
`pipeline/handlers/codergen_test.go`	Adds tests for node attr + alias mapping.
`CHANGELOG.md`	Documents the new opt-in plan-before-execute phase.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Copilot · 2026-04-21T18:05:52Z

+	planBeforeExecutePrompt = "Before you start executing this task, outline your plan: what files to examine, what changes to make, and in what order. Do not call tools yet."
+	executeAfterPlanPrompt  = "Now execute the task using the plan above. Start making progress with tool calls."


The planning prompts don’t account for sessions using structured output (ResponseFormat/strict json_schema). Since doPlanningCall reuses buildResponseFormat(), the planning response may be forced into a JSON schema that was designed for the final execution output, causing provider-side schema failures or invalid planning output. Consider either (a) omitting ResponseFormat for the planning call, or (b) adapting planBeforeExecutePrompt to explicitly instruct the model how to express the plan within the configured response format/schema (e.g., a dedicated plan field).

Suggested change

planBeforeExecutePrompt = "Before you start executing this task, outline your plan: what files to examine, what changes to make, and in what order. Do not call tools yet."

executeAfterPlanPrompt = "Now execute the task using the plan above. Start making progress with tool calls."

planBeforeExecutePrompt = "Before you start executing this task, outline your plan: what files to examine, what changes to make, and in what order. Do not call tools yet. If a structured response format or JSON schema is active for this session, express the plan using that required format instead of free-form text: put the plan in the most appropriate existing text/string field(s), keep it concise, and do not invent new fields that are not allowed by the schema."

executeAfterPlanPrompt = "Now execute the task using the plan above. Start making progress with tool calls. If the prior plan was expressed through a structured response format or JSON schema, continue based on that plan rather than repeating it."

Copilot · 2026-04-21T18:05:52Z

+	result.Usage = result.Usage.Add(resp.Usage)
+	if resp.Usage.EstimatedCost == 0 {
+		result.Usage.EstimatedCost += llm.EstimateCost(s.config.Model, resp.Usage)
+	}
+	s.messages = append(s.messages, resp.Message)


maybeRunPlanningTurn appends the planning response directly into the conversation without handling the case where the model returns tool calls (despite the prompt). If resp.ToolCalls() is non-empty / finish reason indicates tool calls, those calls will never be executed (planning runs outside the normal tool-execution loop), leaving the conversation in an inconsistent state for the first execution turn. Add an explicit guard here (error out, or strip tool-call parts and re-prompt for a plain plan) before appending resp.Message.

Copilot · 2026-04-21T18:05:53Z

+// and node-level overrides. Node-level "plan" is accepted as a shorthand alias.
+func (h *CodergenHandler) applyPlanningConfig(config *agent.SessionConfig, node *pipeline.Node) {
+	if v, ok := h.graphAttrs["plan_before_execute"]; ok {
+		config.PlanBeforeExecute = v == "true"
+	}
+	if v, ok := node.Attrs["plan_before_execute"]; ok {
+		config.PlanBeforeExecute = v == "true"
+	}
+	if v, ok := node.Attrs["plan"]; ok {


applyPlanningConfig lets the shorthand alias plan override an explicit plan_before_execute value on the same node because it’s applied last. That makes configuration ambiguous (e.g., plan_before_execute="false" + plan="true" ends up enabled). Prefer treating plan as an alias only when plan_before_execute is not present, or define and enforce a clear precedence order (typically the explicit key wins).

Suggested change

// and node-level overrides. Node-level "plan" is accepted as a shorthand alias.

func (h *CodergenHandler) applyPlanningConfig(config *agent.SessionConfig, node *pipeline.Node) {

if v, ok := h.graphAttrs["plan_before_execute"]; ok {

config.PlanBeforeExecute = v == "true"

}

if v, ok := node.Attrs["plan_before_execute"]; ok {

config.PlanBeforeExecute = v == "true"

}

if v, ok := node.Attrs["plan"]; ok {

// and node-level overrides. Node-level "plan" is accepted as a shorthand alias

// only when the explicit "plan_before_execute" attribute is not present.

func (h *CodergenHandler) applyPlanningConfig(config *agent.SessionConfig, node *pipeline.Node) {

if v, ok := h.graphAttrs["plan_before_execute"]; ok {

config.PlanBeforeExecute = v == "true"

}

if v, ok := node.Attrs["plan_before_execute"]; ok {

config.PlanBeforeExecute = v == "true"

} else if v, ok := node.Attrs["plan"]; ok {

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 7d83ae8daa

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

chatgpt-codex-connector · 2026-04-21T18:06:24Z

+		return err
+	}
+
+	result.Usage = result.Usage.Add(resp.Usage)


Exclude planning tokens from empty-response accounting

Adding resp.Usage from the planning-only call into SessionResult.Usage causes downstream empty-response handling to misclassify failed execution turns when planning is enabled. buildEmptyResponseOutcome in pipeline/handlers/codergen.go treats Usage.OutputTokens == 0 as the signal to retry swallowed provider responses; after this change, a normal planning reply (non-zero output tokens) can mask a subsequent execution turn that returned zero tokens/no tool calls, so the node is marked as normal/turn-limit failure instead of retryable. This is most visible with low max_turns (for example 1), where the execution retry prompt cannot run and the pipeline now loses its fallback retry path.

Useful? React with 👍 / 👎.

clintecker · 2026-04-21T18:12:04Z

@copilot address all the feedback

Agent-Logs-Url: https://github.com/2389-research/tracker/sessions/3b682829-1719-4abb-bca3-487cb08d8ac4 Co-authored-by: clintecker <26985+clintecker@users.noreply.github.com>

Copilot · 2026-04-21T18:16:15Z

@copilot address all the feedback

Addressed all requested feedback in commit eafeae3:

planning call no longer enforces execution response format/schema
planning turn now errors if tool calls are returned
plan alias now applies only when plan_before_execute is absent
added targeted tests for each case

Validation is green (go build ./..., go test ./... -short, and parallel validation).

Copilot

Pull request overview

Copilot reviewed 8 out of 8 changed files in this pull request and generated no new comments.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Initial plan

5307061

Copilot AI assigned Copilot and clintecker Apr 21, 2026

Copilot started work on behalf of clintecker April 21, 2026 16:57 View session

Copilot AI linked an issue Apr 21, 2026 that may be closed by this pull request

feat(agent): plan-before-execute phase for complex tasks #97

Closed

4 tasks

Copilot AI and others added 3 commits April 21, 2026 17:03

feat(agent): add optional plan-before-execute turn

b307bf3

Agent-Logs-Url: https://github.com/2389-research/tracker/sessions/13e98b22-007e-4fb3-9f06-d88c3546f4f5 Co-authored-by: clintecker <26985+clintecker@users.noreply.github.com>

test(agent): guard against empty planning message capture

9ca40d4

Agent-Logs-Url: https://github.com/2389-research/tracker/sessions/13e98b22-007e-4fb3-9f06-d88c3546f4f5 Co-authored-by: clintecker <26985+clintecker@users.noreply.github.com>

fix(agent): carry response format through planning turn

7d83ae8

Agent-Logs-Url: https://github.com/2389-research/tracker/sessions/13e98b22-007e-4fb3-9f06-d88c3546f4f5 Co-authored-by: clintecker <26985+clintecker@users.noreply.github.com>

Copilot AI changed the title ~~[WIP] Add plan-before-execute phase for complex tasks~~ feat(agent): add optional plan-before-execute turn for complex codergen sessions Apr 21, 2026

Copilot AI requested a review from clintecker April 21, 2026 17:08

Copilot finished work on behalf of clintecker April 21, 2026 17:08

clintecker requested a review from Copilot April 21, 2026 18:01

clintecker marked this pull request as ready for review April 21, 2026 18:01

Copilot started reviewing on behalf of clintecker April 21, 2026 18:01 View session

Copilot AI reviewed Apr 21, 2026

View reviewed changes

chatgpt-codex-connector Bot reviewed Apr 21, 2026

View reviewed changes

Copilot started work on behalf of clintecker April 21, 2026 18:12 View session

fix(agent): address planning-turn review feedback

eafeae3

Agent-Logs-Url: https://github.com/2389-research/tracker/sessions/3b682829-1719-4abb-bca3-487cb08d8ac4 Co-authored-by: clintecker <26985+clintecker@users.noreply.github.com>

clintecker requested a review from Copilot April 21, 2026 18:17

Copilot finished work on behalf of clintecker April 21, 2026 18:17

Copilot started reviewing on behalf of clintecker April 21, 2026 18:17 View session

Copilot AI reviewed Apr 21, 2026

View reviewed changes

clintecker merged commit 934aa88 into main Apr 21, 2026
5 checks passed

clintecker mentioned this pull request Apr 21, 2026

release: v0.21.0 #146

Merged

3 tasks

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

feat(agent): add optional plan-before-execute turn for complex codergen sessions#137

feat(agent): add optional plan-before-execute turn for complex codergen sessions#137
clintecker merged 5 commits intomainfrom
copilot/feat-plan-before-execute-phase

Copilot AI commented Apr 21, 2026 •

edited

Loading

Uh oh!

Copilot AI left a comment

Uh oh!

Copilot AI Apr 21, 2026

Uh oh!

Copilot AI Apr 21, 2026

Uh oh!

Copilot AI Apr 21, 2026

Uh oh!

chatgpt-codex-connector Bot left a comment

Uh oh!

chatgpt-codex-connector Bot Apr 21, 2026

Uh oh!

clintecker commented Apr 21, 2026

Uh oh!

Copilot AI commented Apr 21, 2026

Uh oh!

Copilot AI left a comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

		planBeforeExecutePrompt = "Before you start executing this task, outline your plan: what files to examine, what changes to make, and in what order. Do not call tools yet."
		executeAfterPlanPrompt = "Now execute the task using the plan above. Start making progress with tool calls."

Conversation

Copilot AI commented Apr 21, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Copilot AI Apr 21, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 21, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 21, 2026

Choose a reason for hiding this comment

Uh oh!

chatgpt-codex-connector Bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

chatgpt-codex-connector Bot Apr 21, 2026

Choose a reason for hiding this comment

Uh oh!

clintecker commented Apr 21, 2026

Uh oh!

Copilot AI commented Apr 21, 2026

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Copilot AI commented Apr 21, 2026 •

edited

Loading