feat(mcp): gittensory_lint_pr_text (commit/PR-body rubric linter)

[oktofeesh1]

Fixes #549

What

Adds gittensory_lint_pr_text, a deterministic MCP tool that lints a commit message + PR body against the gittensor traceability / no-issue-rationale rubric before submitting. It returns a quality verdict (strong / adequate / weak) plus specific, public-safe fixes — catching generic/empty AI-slop text pre-submit. Metadata only; no source upload, no GitHub writes.

How

• src/signals/engine.ts — buildPrTextLint scores three rubric components (traceability, commit message, PR body), reusing the existing rubric helpers: hasClearNoIssueRationale, tokenize + STOPWORDS, and hasValidationNote. Every reason/fix is routed through sanitizePublicComment (public-safe, fail-closed).
• src/mcp/server.ts — registers the tool with an inputSchema and outputSchema.
• src/api/routes.ts — POST /v1/lint/pr-text backing the local CLI tool.
• packages/gittensory-mcp/bin/gittensory-mcp.js — exposes the tool, proxying to the API route.

Input: { commitMessages?, prBody?, linkedIssue? }. Output: { verdict, score, components[], fixes[], summary }.

Output boundaries

Public-safe verdict + fixes only; no source contents leave the process.

Tests

• test/unit/pr-text-lint.test.ts — verdict cases (strong/adequate/weak), each rubric component's weak paths (generic/empty/short commit, empty/template/thin body, missing traceability), and forbidden-term assertions on all output.
• test/unit/mcp-output-schemas.test.ts — output-schema discovery + structured-content call.
• test/integration/api.test.ts — route success + invalid-body (400).

npm run validate (typecheck + coverage) is green at the 97%+ branch threshold; build:mcp, test:mcp-pack, test:workers, and ui:openapi:check also pass.

🤖 Generated with Claude Code

[ghost]

Note

Gittensory Gate skipped

PR closed before full evaluation. No late first comment was created.

Signal	Result	Evidence	Action
Gate result	⚠️ Skipped	#634 is no longer open.	No action.

---

💰 Earn for open-source contributions like this. Gittensor lets GitHub contributors earn for the work they already do — register to start earning →.

Checked by Gittensory, a quiet PR intelligence layer for OSS maintainers.

[ghost]

Gittensory review · advisory — a maintainer merges

0 actionable · 5 nitpick(s) · 7 file(s) · two independent AI reviewers

Suggested action: ✅ Safe to merge — both reviewers found no blocking issues.

📋 Walkthrough

The change introduces a PR‑text linting tool that evaluates commit messages, PR bodies, and issue linkage, exposing a new /v1/lint/pr-text endpoint and MCP tool, with full schema validation and tests ensuring public‑safe output.

Changes

File	Summary
packages/gittensory-mcp/bin/gittensory-mcp.js	Adds input schema for the new lint tool and registers the tool with description.
src/api/routes.ts	Defines request schema, adds POST /v1/lint/pr-text route, and returns lint report.
src/mcp/server.ts	Adds lint tool registration with input and output schemas, and implements tool handler.
src/signals/engine.ts	Implements deterministic PR‑text linting logic, defines types, and exports buildPrTextLint.
test/integration/api.test.ts	Tests the new lint endpoint for success and validation errors, and checks for forbidden terms.
test/unit/mcp-output-schemas.test.ts	Adds lint tool to the list of tools with output schemas and tests its schema compliance.
test/unit/pr-text-lint.test.ts	Adds extensive unit tests for the linting function covering various scenarios and public safety.

🔍 Reviewer notes

Reviewer A · gpt-oss-120b — recommends ✅ merge
This PR adds a new MCP tool gittensory_lint_pr_text with API route, schema, server registration, implementation, and comprehensive tests. The implementation respects public‑safe sanitisation and integrates cleanly with existing tooling.

Suggestions

• Consider consolidating the duplicate lintPrTextShape definition (present in both the CLI bin and server) into a shared module to avoid future drift.
• Add a comment documenting the deterministic scoring weights (PR_TEXT_LINT_WEIGHTS) for future maintainers.
• Run the full test suite locally to confirm no other modules rely on a previous signature of hasClearNoIssueRationale (though TypeScript should accept the narrowed type).

Worth double-checking

• Potential mismatch if future changes modify lintPrTextShape in one location but not the other.
• Ensure sanitizePublicComment is imported in engine.ts; missing import would cause a compile error.

Reviewer B · mistral-small-3.1-24b-instruct — recommends ✅ merge
This PR adds a new MCP tool for linting PR text, including commit messages and PR bodies. The changes are well-structured, with thorough testing and clear documentation. The implementation respects the project's privacy and safety model.

Suggestions

• Consider adding more detailed comments in the buildPrTextLint function to explain the logic behind the scoring and verdicts.
• Ensure that the sanitizePublicComment function is robust and covers all edge cases to prevent any forbidden terms from leaking into public output.

Worth double-checking

• Verify that the new tool does not introduce any performance bottlenecks, especially with large PR bodies or commit messages.
• Double-check that the sanitizePublicComment function is applied consistently across all public-facing outputs.
• Ensure that the new tool handles edge cases, such as extremely long or empty inputs, gracefully.

---

🤖 Automated advisory review · a maintainer makes the final call.

[superagent-security]

Superagent didn't find any vulnerabilities or security issues in this PR.

[ghost]

gittensory · advisory review

Reviewed 7 changed file(s) — two independent AI reviewers.

Suggested action: ✅ Safe to merge — both reviewers found no blocking issues.

Reviewer A · gpt-oss-120b — recommends ✅ merge
This PR introduces a deterministic PR‑text linting tool, adds API endpoint, schema, CLI registration, and comprehensive unit/integration tests. The implementation respects the public‑private sanitization boundary and integrates cleanly with existing MCP tooling.

Suggestions

• Document the new PrTextLint* types in the public API docs for developers.
• Consider defining a more specific Zod schema for components in the output rather than z.unknown() for stronger validation.
• Verify that the replacement of hasClearNoIssueRationale does not leave a duplicate or stray function definition; keep a single implementation.

Worth double-checking

• Ensure sanitizePublicComment is applied to all user‑visible strings to avoid accidental forbidden term leakage.
• Confirm that the new endpoint respects the same authentication/authorization checks as other MCP tools.

Reviewer B · nemotron-3-120b-a12b — recommends ✅ merge
The PR adds a new MCP tool 'gittensory_lint_pr_text' that lints commit messages and PR bodies against a deterministic rubric (traceability, conventional commits, body substance). It includes types, helper functions, MCP/server integration, API route, and comprehensive unit/integration tests. Output is sanitized via sanitizePublicComment to prevent leaking forbidden terms. The change is well-tested, follows project conventions, and maintains the public/private boundary.

No blocking issues spotted.

[JSONbored]

Solid, well-tested, privacy-clean, conflict-free. Two revisions before it ships as an anti-slop authority:

1. It doesn't actually check Conventional Commit format (high). Despite the description, buildPrTextLint only checks length + token count + generic words — so fix(api): null check is flagged weak while non-conventional Add stuff to endpoint here ok passes. Either add a real ^(feat|fix|test|docs|refactor|build|ci|chore|revert)(\(scope\))?: … check (reuse CONTRIBUTING's type list) or drop the "Conventional-Commit" claim.
2. False positives on non-Latin text (medium). tokenize() splits on [^a-z0-9]+, so a fully-written CJK/Cyrillic body yields 0 tokens → flagged "thin." Add a Unicode-aware fallback (\p{L} runs / non-whitespace length) before declaring a body thin.

Minor: score vs verdict can rank-disagree (traceability is a hard gate) — document it or omit score when traceability fails.