mcpField
L
light-code-review

light-code-review

github.com/light-chengTesting & quality510 installsUpdated 4d ago
Repository

About light-code-review

light-code-review is a read-only reviewer for bounded git diffs, intended for code review workflows. It checks repository standards and specification requirements in separate reports, with cited findings and no file edits or final pass/fail decision.

Key features

  1. 01Bounded git diff review
  2. 02Separate Standards and Spec checks
  3. 03Cited, uniquely identified findings
  4. 04Parallel specialist reviewers
  5. 05Read-only operation

Use cases

  1. 01Reviewing a change against repository standards
  2. 02Checking a diff against its issue or Spec
  3. 03Returning findings for a repair and re-review loop

Install

Copy the command and run it in a terminal.

$npx skills add https://github.com/light-cheng/skills --skill light-code-review

After installing, type/light-code-reviewin Claude Code to confirm it loaded.

Bundled files

References and scripts installed along with SKILL.md.

ATTRIBUTION.md
agents/
openai.yaml
references/
EXAMPLES.md
SMELL-BASELINE.md
WORKFLOW.md

Details

From the project README

Code Review

light-code-review is a read-only specialist reviewer for a bounded git diff. It reports findings on Standards and Spec axes and returns them; it never edits files, runs a repair loop, or issues the final PASS / FAIL / BLOCKED.

Core behavior

  1. Pin the fixed point. Resolve the ref with git rev-parse; capture git diff <fixed-point>...HEAD and git log <fixed-point>..HEAD --oneline. A bad ref or empty diff stops before any sub-agent runs.
  2. Locate sources. Find the originating issue/Spec and repository standards docs. A missing Spec yields "no spec available", not an invented requirement.
  3. Run both axes in parallel sub-agents. Dedicated Standards and Spec sub-agents share no context and may not re-invoke light-code-review or spawn further agents.
  4. Aggregate separately. Present ## Standards and ## Spec reports with globally unique normalized finding IDs, or Findings: [] when clean, and a one-line per-axis summary. Preserve prior IDs on recheck. Do not merge or rerank across axes.

Every finding carries a citation: standards file + rule or smell name + quoted hunk, or the Spec line. Tooling-enforced style is skipped; repo standards override the baseline smell set; smells remain labelled judgement calls.

Composition

review-loop → light-code-review → findings → Producer repair → re-review → handoff
project-review → review-loop → light-code-review → findings → Core validation → bounded repair → PASS/FAIL/BLOCKED

Detailed prompts and smell baseline are in WORKFLOW.md and SMELL-BASELINE.md; examples are in EXAMPLES.md.

Comments

Markdown supported