hardening(p7): P0 sweep 2026-04-23 — sanitizer parity, ConfigureAwait, null-guards, ProviderName + 0.16.1 by cyberprophet · Pull Request #95 · cyberprophet/mint

cyberprophet · 2026-04-23T14:29:15Z

Summary

Fix [P0] GeminiProvider inserts user-supplied text into prompts without PromptSanitizer #92 — GeminiProvider.ExtractProductInfoAsync + AnalyzeReferenceLinkAsync now use PromptSanitizer.EscapeForPrompt / EscapeIdentifierForPrompt (parity with GptService path)
Fix [P0] ConfigureAwait(false) absent across the public library #93 — ConfigureAwait(false) applied to every library await (mechanical, 12 files); CA2007 promoted to error via .editorconfig + AnalysisLevel=latest-Recommended to prevent regression
P7-04 from [P0] Test + hardening backlog from 2026-04-23 sweep (P7 batched) #94 — GenerateTitleAsync null-guard (ArgumentException.ThrowIfNullOrWhiteSpace(conversationText)) on both GptService and GeminiProvider
P7-07 from [P0] Test + hardening backlog from 2026-04-23 sweep (P7 batched) #94 — ApiUsageEvent providerName now uses this.ProviderName instead of hardcoded "openai" at ReferenceLink + StudioMint call sites
NuGet version bump: 0.16.0 → 0.16.1

Test plan

Gemini prompt sanitizer parity tests (ExtractProductInfo + AnalyzeReferenceLink — delimiter injection, angle-bracket escaping)
GenerateTitleAsync null/empty/whitespace throws ArgumentException on both providers (6 theory cases each)
ApiUsageEvent ProviderName propagation (groq, minimax, custom)
dotnet build -c Release clean with CA2007 enforcing (0 errors)
dotnet test green: 708 total (658 pre-existing + 50 new)

Closes #92, closes #93. Absorbs P7-04 + P7-07 from #94 (remaining sub-items tracked separately).

Post-merge: publish ShareInvest.Agency.0.16.1.nupkg to nuget.org, then bump Models.csproj in P5 to <PackageReference Include="ShareInvest.Agency" Version="0.16.1" /> and redeploy P5.

🤖 Generated with Claude Code

…, null-guards, ProviderName; bump to 0.16.1 Fix #92: GeminiProvider.ExtractProductInfoAsync and AnalyzeReferenceLinkAsync now wrap all user-controlled text (d.Id, d.Text, url, html, context fields) with PromptSanitizer.EscapeIdentifierForPrompt / EscapeForPrompt, matching the existing GptService path (S-12 prompt-injection defence parity). Fix #93: ConfigureAwait(false) applied to every await expression across all 12 library source files (GptService.*, GeminiProvider, WebTools, FallbackSearchProvider). CA2007 promoted to error via .editorconfig + AnalysisLevel=latest-Recommended in Agency.csproj to prevent regression. P7-04 (#94): ArgumentException.ThrowIfNullOrWhiteSpace(conversationText) added at the top of GenerateTitleAsync on both GptService and GeminiProvider, matching the existing guard on systemPrompt. P7-07 (#94): ApiUsageEvent call sites in GptService.ReferenceLink (line 87) and GptService.StudioMint (line 120) changed from hardcoded "openai" to this.ProviderName so custom-endpoint providers (Groq, MiniMax, etc.) report correctly. Version bump: 0.16.0 → 0.16.1. Tests: 708 green (658 pre-existing + 50 new from this branch). Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

cyberprophet · 2026-04-23T14:58:23Z

@codex review

@cyberchacha @MintSurf — friendly ping, this PR has been awaiting review for ~30 min. Summary:

[P0] GeminiProvider inserts user-supplied text into prompts without PromptSanitizer #92 GeminiProvider PromptSanitizer parity
[P0] ConfigureAwait(false) absent across the public library #93 ConfigureAwait(false) across 12 files (32 await sites) + CA2007 enforce
P7-04 GenerateTitleAsync null-guard
P7-07 ApiUsageEvent ProviderName fix
Version bump 0.16.0 → 0.16.1

708 tests pass, dotnet build -c Release 0 errors.

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: f5458d1a8c

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 · 2026-04-23T15:03:11Z


        var documentBlock = string.Join("\n\n", documents.Select(d =>
-            $"--- DOCUMENT: {d.Id} ---\n{d.Text}\n--- END ---"));
+            $"--- DOCUMENT: {PromptSanitizer.EscapeIdentifierForPrompt(d.Id)} ---\n{PromptSanitizer.EscapeForPrompt(d.Text)}\n--- END ---"));


Keep Gemini document IDs canonical for provenance matching

Escaping d.Id here can change canonical IDs (for example & to &), but TryParseProductInfo later compares returned SourceDocuments to the original raw IDs via knownIds.Contains(...) without HTML-decoding. In calls where document IDs are URLs or other strings containing escapable characters, Gemini can legitimately echo the escaped ID it was shown, causing the match to fail and dropping provenance for valid extraction results. Please preserve round-trippable IDs (or decode before validation) so source attribution is not lost.

Useful? React with 👍 / 👎.

cyberprophet · 2026-04-23T15:05:47Z

Thanks @chatgpt-codex-connector. Declining the change to keep OpenAI ↔ Gemini parity:

The exact same PromptSanitizer.EscapeIdentifierForPrompt(doc.Id) call has been the OpenAI path's behavior for GptService.ProductInfo.cs:84 since Intent 043 (Agency 0.14.0, 2026-04-17). This PR brings the Gemini path into line with it, not past it.

On the round-trip concern: PromptSanitizer.EscapeIdentifierForPrompt only escapes <, >, &, ". In this codebase, doc.Id flows in from PickFile.Id (P5) which is a Guid — punctuation-free by construction, so the escape is a no-op in practice. The helper's XML docstring explicitly acknowledges this ("All other characters — including the document id's original casing and punctuation — pass through unchanged so knownIds.Contains(id) still succeeds on the parsed output").

If IDs ever did contain HTML-special characters, the fix belongs in PromptSanitizer itself (either decode on round-trip, or switch to a URL-safe delimiter) — which would benefit both providers and is out of scope for this PR. Divergence between the two provider paths would be strictly worse: either (a) Gemini opens a prompt-injection hole that OpenAI closed, or (b) the two providers accept different ID shapes, making provider switching silently brittle.

chatgpt-codex-connector · 2026-04-23T15:07:25Z

Summary

I reviewed the trigger and did not apply any code changes. The requested “decline” is consistent with current implementation: Gemini now uses PromptSanitizer.EscapeIdentifierForPrompt(d.Id) in ExtractProductInfoAsync, matching the OpenAI product-info path’s long-standing behavior. agency/Google/GeminiProvider.csL151-L166 agency/OpenAI/GptService.ProductInfo.csL79-L85
The sanitizer behavior also matches the rationale in code comments/docs: only <, >, &, and " are escaped, while other characters are preserved for ID matching semantics (knownIds.Contains(id)). agency/OpenAI/PromptSanitizer.csL65-L86
Per repo startup protocol, I checked branch/status and (correctly) skipped git pull origin main because the working branch is work, not main. AGENTS.mdL11-L14
I could not run the required P0-issue query because gh is not installed in this environment. AGENTS.mdL24-L27

No commit was created and no follow-up PR was opened, because there were no file changes to commit.

Checks

✅ git -C /workspace/mint status --short --branch
✅ sed -n '1,220p' /workspace/mint/README.md
✅ sed -n '1,260p' /workspace/mint/CHANGELOG.md
⚠️ gh issue list --repo cyberprophet/mint --state open --search '[P0]' --json number,title,body --limit 10 (failed: gh: command not found)
✅ sed -n '130,230p' /workspace/mint/agency/Google/GeminiProvider.cs
✅ sed -n '1,220p' /workspace/mint/agency/OpenAI/PromptSanitizer.cs
✅ nl -ba /workspace/mint/agency/OpenAI/GptService.ProductInfo.cs | sed -n '60,110p'
✅ git -C /workspace/mint status --short

View task →

MintSurf

Review Summary

Outcome: comment
Summary: The hardening changes look directionally correct, but the PR bumps the NuGet version to 0.16.1 without the matching CHANGELOG entry that this repo treats as the authoritative consumer-facing release record. I also could not run the optional project checks here because dotnet is not installed in this environment, and there is no *.sln/Node/Flutter manifest matching the requested check gates.

MintSurf · 2026-04-23T23:11:31Z

 		<!-- NuGet Package -->
 		<PackageId>ShareInvest.Agency</PackageId>
-		<Version>0.16.0</Version>
+		<Version>0.16.1</Version>


Missing 0.16.1 changelog entry

This PR bumps the package version to 0.16.1, but CHANGELOG.md still tops out at 0.16.0. Per the repo instructions, every release needs a changelog entry and the latest entry is the authoritative source for NuGet consumers. Shipping the version bump without that release note leaves consumers and downstream repos without an accurate record of what 0.16.1 contains.

Useful? React with 👍 / 👎.

cyberchacha

Review Summary

Outcome: approve
Summary: Clean, well-scoped hardening PR. All four findings (sanitizer parity, ConfigureAwait, null-guards, ProviderName) are correctly implemented. Build passes with 0 errors and CA2007 enforcement active. All 708 tests green (including 50 new). Mechanical changes are consistent and complete across all call sites.

cyberchacha · 2026-04-23T23:12:12Z

+
+		<!-- Analysis: CA2007 (ConfigureAwait) as error to prevent regression -->
+		<AnalysisLevel>latest-Recommended</AnalysisLevel>
+		<TreatWarningsAsErrors>false</TreatWarningsAsErrors>


TreatWarningsAsErrors=false — verify intent

AnalysisLevel=latest-Recommended enables a broad set of analyzers. With TreatWarningsAsErrors=false, only CA2007 (via .editorconfig) is promoted to error — the remaining ~120 warnings (CA1848, CA1861, CA1305, etc.) stay as warnings. If that's intentional, this is fine. If the goal is to gradually tighten, consider adding a tracking issue to promote additional rules over time rather than leaving them as non-blocking.

Useful? React with 👍 / 👎.

cyberchacha · 2026-04-23T23:12:12Z


        var documentBlock = string.Join("\n\n", documents.Select(d =>
-            $"--- DOCUMENT: {d.Id} ---\n{d.Text}\n--- END ---"));
+            $"--- DOCUMENT: {PromptSanitizer.EscapeIdentifierForPrompt(d.Id)} ---\n{PromptSanitizer.EscapeForPrompt(d.Text)}\n--- END ---"));


Good: sanitizer parity now matches GptService path

Using EscapeIdentifierForPrompt for d.Id and EscapeForPrompt for d.Text correctly mirrors the defense-in-depth pattern from the OpenAI path. The <user_input> wrapping inside --- DOCUMENT --- delimiters is a clean layered approach.

Useful? React with 👍 / 👎.

Per repo convention every NuGet version bump must land with a matching CHANGELOG section. Added a Keep-a-Changelog entry for 0.16.1 covering: - Security: GeminiProvider PromptSanitizer parity (#92) - Changed: ConfigureAwait(false) mechanical rollout + CA2007 error-level (#93) - Fixed: GenerateTitleAsync null-guard (#94 P7-04) + ApiUsageEvent.ProviderName (#94 P7-07) - Notes: P5 Models.csproj bump to 0.16.1 + redeploy required Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

cyberprophet · 2026-04-23T23:17:31Z

Addressed MintSurf P2 on f5458d1 in follow-up commit e59bf1a:

P2 — missing CHANGELOG 0.16.1 entry: added a Keep-a-Changelog section for 0.16.1 covering Security (#92 GeminiProvider sanitizer parity), Changed (#93 ConfigureAwait + CA2007 error), Fixed (#94 P7-04 null-guard + P7-07 ProviderName), and a Notes row flagging the required Models.csproj bump + P5 redeploy after nuget.org publish.

Also declined the Codex P2 (Gemini ID escape parity) inline above — divergence from the OpenAI path would strictly worsen prompt-injection defense; fix belongs in PromptSanitizer itself if anywhere, out of scope for this PR.

@MintSurf — could I get your approval once CHANGELOG looks good? main branch requires 2 approvals and I currently have 1 (cyberchacha APPROVED).

chatgpt-codex-connector Bot reviewed Apr 23, 2026

View reviewed changes

MintSurf reviewed Apr 23, 2026

View reviewed changes

cyberchacha previously approved these changes Apr 23, 2026

View reviewed changes

cyberprophet dismissed cyberchacha’s stale review via e59bf1a April 23, 2026 23:17

cyberprophet merged commit 098e56e into main Apr 23, 2026

cyberprophet deleted the hardening/p0-sweep-2026-04-23 branch April 23, 2026 23:20

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

hardening(p7): P0 sweep 2026-04-23 — sanitizer parity, ConfigureAwait, null-guards, ProviderName + 0.16.1#95

hardening(p7): P0 sweep 2026-04-23 — sanitizer parity, ConfigureAwait, null-guards, ProviderName + 0.16.1#95
cyberprophet merged 2 commits intomainfrom
hardening/p0-sweep-2026-04-23

cyberprophet commented Apr 23, 2026

Uh oh!

cyberprophet commented Apr 23, 2026

Uh oh!

chatgpt-codex-connector Bot left a comment

Uh oh!

chatgpt-codex-connector Bot Apr 23, 2026

Uh oh!

cyberprophet commented Apr 23, 2026

Uh oh!

chatgpt-codex-connector Bot commented Apr 23, 2026

Uh oh!

MintSurf left a comment

Uh oh!

MintSurf Apr 23, 2026

Uh oh!

cyberchacha left a comment

Uh oh!

cyberchacha Apr 23, 2026

Uh oh!

cyberchacha Apr 23, 2026

Uh oh!

cyberprophet commented Apr 23, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Conversation

cyberprophet commented Apr 23, 2026

Summary

Test plan

Uh oh!

cyberprophet commented Apr 23, 2026

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 23, 2026

Choose a reason for hiding this comment

Uh oh!

cyberprophet commented Apr 23, 2026

Uh oh!

chatgpt-codex-connector Bot commented Apr 23, 2026

Summary

Checks

Uh oh!

MintSurf left a comment

Choose a reason for hiding this comment

Review Summary

Uh oh!

MintSurf Apr 23, 2026

Choose a reason for hiding this comment

Uh oh!

cyberchacha left a comment

Choose a reason for hiding this comment

Review Summary

Uh oh!

cyberchacha Apr 23, 2026

Choose a reason for hiding this comment

Uh oh!

cyberchacha Apr 23, 2026

Choose a reason for hiding this comment

Uh oh!

cyberprophet commented Apr 23, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants