agentsclimarketplace

Grava code review

Skill hoangtrungnguyen/grava/.claude/skills/grava-code-review

Use when reviewing code for a grava issue that has been marked for code review. Trigger when the user says "review grava-XXX", "code review for this issue", "review the changes", or any request to review code tied to a grava issue. The issue must have a last_commit stored in its metadata — this skill uses that commit to find changed files and conduct the review.From its SKILL.md

Install
npx -y skills add hoangtrungnguyen/grava --skill grava-code-review

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

6.0 KB, as published. Nobody here has run it

Grava Code Review

Review code changes associated with a grava issue by inspecting the files changed in the recorded commit, then post findings as a comment and label the issue as reviewed.

Announce at start: "Using the grava-code-review skill to review <issue-id>."

Prerequisites

  • A valid grava issue ID with last_commit stored in its metadata (set by the grava-complete-dev-story skill or manually via grava update <id> --last-commit <hash>)

Steps

1. Fetch the issue and extract the commit hash

grava show <issue-id> --json

Parse the JSON output to get the last_commit field. If it's missing or empty, stop and tell the user: "This issue has no commit hash recorded. Run grava update <issue-id> --last-commit <hash> first, or use the grava-complete-dev-story skill to commit and record it."

2. Get the changed files from the commit

git show --stat <commit-hash>
git diff <commit-hash>~1 <commit-hash> --name-only

This gives you the list of files that were changed in that commit. If the commit is a merge commit or the first commit, adjust accordingly:

# For merge commits
git diff <commit-hash>^1 <commit-hash> --name-only

# If ~1 fails (first commit)
git diff --root <commit-hash> --name-only

3. Read and review each changed file

For each changed file, get the diff:

git diff <commit-hash>~1 <commit-hash> -- <file-path>

Conduct the review focusing on:

  • Correctness: Does the code do what the issue description says it should?
  • Bugs: Off-by-one errors, nil/null dereferences, race conditions, resource leaks
  • Security: Injection vulnerabilities, hardcoded secrets, unsafe input handling
  • Error handling: Are errors checked and propagated properly?
  • Naming and clarity: Are names descriptive? Is the logic easy to follow?
  • Tests: Are there tests for the new code? Do they cover edge cases?
  • Style: Does the code follow the project's existing conventions?

Cross-reference the changes against the issue's description and acceptance criteria (from grava show) to verify completeness.

Classify every finding into exactly one severity bucket:

  • CRITICAL — blocks merge. Data loss, security holes, broken correctness, crashes on normal inputs, regressions in acceptance criteria.
  • HIGH — should fix before merge. Clear bugs on edge paths, missing error handling on failure modes, missing tests for new branches, API contract drift.
  • MEDIUM — fix soon. Brittle patterns, unclear naming that will cost later, weak test coverage, minor spec gaps.
  • LOW — nice to have. Style nits, comment/doc polish, micro-refactors.

4. Post findings as separate comments per severity

The issue_comments table stores message as plain TEXT with no severity column, so severity lives in the message body. Post one comment per severity level that has findings (skip empty buckets). This lets readers scan, filter, and prioritize by the leading tag without parsing a monolithic review.

Post in order from highest to lowest severity so the comment timeline reads top-down by priority:

grava comment <issue-id> -m "[CRITICAL] review <short-hash> — <N> finding(s)

- <file:line> — <finding>. <why it matters>. <suggested fix>
- <file:line> — <finding>. <why it matters>. <suggested fix>"

Use the identical template for [HIGH], [MEDIUM], [LOW]. Rules for the message body:

  • First line: [<SEVERITY>] review <short-hash> — <N> finding(s). The bracketed tag must be the very first characters so SELECT ... WHERE message LIKE '[CRITICAL]%' and text sorts work.
  • Blank line, then a bullet per finding.
  • Each bullet: <file:line> — <what> . <why> . <fix>. Keep each bullet self-contained — readers may see one comment in isolation.
  • No nested headers inside a severity comment. One severity, one flat list.

After the per-severity comments, post a final summary comment:

grava comment <issue-id> -m "[REVIEW] <short-hash> — <files> files, critical=<n> high=<n> medium=<n> low=<n>

Summary: <1–2 sentences on what changed and overall assessment>

Verdict: APPROVED | CHANGES_REQUESTED"

Verdict rule: CHANGES_REQUESTED if there is any CRITICAL or HIGH finding; otherwise APPROVED. MEDIUM and LOW alone do not block.

If there are zero findings at every severity, skip the per-severity comments and post only the [REVIEW] summary with APPROVED.

5. Update labels

Remove the code_review label (if present) and add reviewed:

grava label <issue-id> --remove code_review
grava label <issue-id> --add reviewed

If the verdict is CHANGES_REQUESTED (any CRITICAL or HIGH finding), add changes_requested instead of reviewed:

grava label <issue-id> --remove code_review
grava label <issue-id> --add changes_requested

6. Commit grava changes

grava commit -m "code review for <issue-id> (commit: <short-hash>)"

7. Print summary

--- Code Review Summary ---
Issue:      <issue-id> — <title>
Commit:     <full-hash>
Files:      <count> files reviewed
Findings:   critical=<n> high=<n> medium=<n> low=<n>
Comments:   <count> posted (one per non-empty severity + summary)
Verdict:    APPROVED | CHANGES_REQUESTED
Label:      reviewed | changes_requested

Error handling

  • Issue not found: Stop, ask user to verify the ID
  • No last_commit: Stop, tell user to record a commit hash first
  • Commit hash not found in git: The commit may have been rebased or force-pushed — tell the user and ask for the correct hash
  • Grava command fails: Show the error, suggest manual recovery

Keep looking

Skills are one crate of 608,666. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.