fix: trim semantic review input for broad PRs

[HanShaoshuai-k]

Summary

Fix semantic review degradation on broad but small PRs by sending only reviewable semantic candidates to the review prompt, instead of passing verbose changed facts such as full output schemas. This keeps the request size bounded while preserving deterministic gate coverage and semantic review behavior.

Changes

• Limit the semantic input view to changed candidates that the semantic reviewer can actually judge, such as command naming, flag naming, skill command references, and boundary error hints.
• Stop sending observe-only default_output facts and verbose output field lists into semantic review.
• Narrow diagnostic context so rule-specific diagnostics only include the relevant fact surface.
• Add regression coverage for broad command/output surfaces and request-size limits.

Test Plan

• go test ./internal/qualitygate/semantic ./internal/qualitygate/cmd/semantic-review -count=1
• go test ./internal/qualitygate/... -count=1
• git diff --check
• Verified the broad shortcut repro stays below the semantic request size limit locally.
• Verified sandbox fork PRs:
  • #11: broad global clean, CI and semantic review passed, no PR comment.
  • #12: broad service clean, CI and semantic review passed, no PR comment.
  • #13: broad shortcut clean repro, CI and semantic review passed, no request-too-large degradation.
  • #14: broad change plus deterministic skill reference defect, deterministic gate and results failed with PR quality summary as expected.

Related Issues

• None

Summary by CodeRabbit

Release Notes

• Bug Fixes

  • Improved handling of broad change surfaces to avoid including excessive or irrelevant diagnostic details.
  • Strengthened enforcement of request/prompt size limits when building and sending semantic review payloads.
  • Sanitized generated semantic view content to omit verbose output fields and prevent accidental leakage.

• Tests

  • Expanded coverage for broad changed-surface summarization, stable reference behavior, and ensuring only intended candidates are retained.
  • Added assertions for payload byte limits and stricter diagnostic/field omission rules.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro

Run ID: 629111b8-6624-4f4f-bbd2-1973564651f9

📥 Commits

Reviewing files that changed from the base of the PR and between e96c3a0 and c8ee0a7.

📒 Files selected for processing (5)

• internal/qualitygate/semantic/client_test.go
• internal/qualitygate/semantic/prompt.go
• internal/qualitygate/semantic/prompt_contract_test.go
• internal/qualitygate/semantic/view.go
• internal/qualitygate/semantic/view_test.go

🚧 Files skipped from review as they are similar to previous changes (5)

• internal/qualitygate/semantic/prompt.go
• internal/qualitygate/semantic/prompt_contract_test.go
• internal/qualitygate/semantic/client_test.go
• internal/qualitygate/semantic/view.go
• internal/qualitygate/semantic/view_test.go

---

📝 Walkthrough

Walkthrough

The semantic quality gate's input view construction is refactored to select only "review candidate" changed facts using predicate-based filtering instead of including all changed facts. OutputInput serialization switches from embedding facts.OutputFact to explicit fields. Diagnostic inclusion is narrowed: ActionReject always passes; other actions require fact intersection. A system-prompt line clarifies changed_summary scope. Tests are extensively updated and added to verify candidate selection, size limits, and no verbose field leakage.

Changes

Semantic view review-candidate refactor

Layer / File(s)	Summary
OutputInput struct and builder internal/qualitygate/semantic/view.go	OutputInput drops the embedded facts.OutputFact and declares FactRef, Command, Domain, Changed, Source, IsList, HasDefaultLimit, HasDecisionField explicitly; outputInputs() populates each field directly; SkillQualityInput.ref() method is removed.
Review candidate selection logic internal/qualitygate/semantic/view.go	BuildInputView switches to addChangedReviewCandidates(), which applies per-fact predicates: commands selected on naming/alias conflicts, skills on invalid references, errors on boundary+hint+zero-count, and outputs explicitly rejected.
Diagnostic context refactoring and inclusion gating internal/qualitygate/semantic/view.go	diagnosticContext refactored to dispatch by diag.Rule using a switch statement, routing to typed helpers for commands, outputs, skills, skill quality, examples, and errors; includeDiagnosticInView always includes ActionReject but otherwise requires fact intersection with diagnostic-derived context.
System prompt changed_summary instruction internal/qualitygate/semantic/prompt.go	Adds one system-prompt line stating changed_summary may cover a broad surface and reviewers must rely only on explicitly listed facts.
View tests for review candidates, leakage, and size internal/qualitygate/semantic/view_test.go	Replaces changed-facts test with review-candidate variant; adds broad-surface summarization tests, semantic candidate retention, verbose field omission, output-candidate size bounds, label diagnostic dropping, and helper generators broadChangedFacts/broadOutputCandidateFacts.
Prompt contract and client request-size tests internal/qualitygate/semantic/prompt_contract_test.go, internal/qualitygate/semantic/client_test.go	Prompt contract test renamed to ...SelectsChangedReviewCandidatesWithStableRefs with SkillQuality/Outputs/Examples assertions updated to expect empty slices; two new client tests assert broad changed surface and broad output candidates stay within MaxRequestBytes with no verbose_output_field_ in HTTP body.

Estimated code review effort

🎯 3 (Moderate) | ⏱️ ~25 minutes

Poem

🐰 Hoppy refactor day!
I trimmed the verbose fields away,
Only candidates may stay—
The broad surface gets a summary,
Size limits keep requests from going funny.
Clean facts in, clean facts out,
That's what review gates are about!

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 0.00% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title concisely and clearly summarizes the main change: trimming semantic review input for broad PRs, which directly aligns with the primary objective of fixing degradation by filtering unnecessary verbose data.
Description check	✅ Passed	The PR description follows the template structure with all required sections: a clear Summary explaining the motivation, a detailed Changes list covering the core modifications, a comprehensive Test Plan with verified steps and sandbox testing, and Related Issues. The description is complete and well-organized.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch fix/semantic-review-input-projection

---

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.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[codecov]

Codecov Report

❌ Patch coverage is 87.23404% with 6 lines in your changes missing coverage. Please review.
✅ Project coverage is 73.72%. Comparing base (c5b5aec) to head (c8ee0a7).

Files with missing lines	Patch %	Lines
internal/qualitygate/semantic/view.go	86.95%	5 Missing and 1 partial ⚠️

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #1505   +/-   ##
=======================================
  Coverage   73.72%   73.72%           
=======================================
  Files         780      780           
  Lines       74140    74170   +30     
=======================================
+ Hits        54659    54683   +24     
- Misses      15420    15425    +5     
- Partials     4061     4062    +1     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[github-actions]

🚀 PR Preview Install Guide

🧰 CLI update

npm i -g https://pkg.pr.new/larksuite/cli/@larksuite/cli@c8ee0a71de058ae3c3c93f391ed9dc8cc7dbb13a

🧩 Skill update

npx skills add larksuite/cli#fix/semantic-review-input-projection -y -g

[github-actions]

PR Quality Summary

CI did not complete successfully. Use the failed check links below to decide whether this PR needs a code change or a rerun.

Failed checks

• e2e-live — failure — details
• results — failure — details