triage-reviews
Skill2 starsChanged 5 months agoApache-2.0View on GitHub
Fetch PR review comments, verify each against real code/docs, fix valid issues, commit and push
- Commits and pushes
3 other projects use a version of this, 1 of them edited.
Adopted by
Most recently taken up by
- digitaltracer/agent-skillstheir own version, first seen Nov 15, 2026
- ShaishavMaisuria/research-paper-lifecycle-skillstheir own version, first seen Nov 1, 2026
- stickerdaniel/linkedin-mcp-serveredited, first seen Sep 7, 2026
Other versions3 projects use a version of this.
Word for word · 0
None found.
Edited · 1
- stickerdaniel/linkedin-mcp-server★ 4.2k100% same
Wrote their own version · 2
- ShaishavMaisuria/research-paper-lifecycle-skills★ 59same name
- digitaltracer/agent-skills★ 4same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository6 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- 1-triage-issuesSkill
- 2-repro-issueSkill
- 3-verify-pr-fixSkill
- btca-localSkill
The file
We split this skill into the rules it gives an agent
8 rules. 0 are common, 5 are distinctive, and 3 are very specific to this project.
0
Common
Most of the 3 edited versions we read keep these.
5
Distinctive
This author's own choices. Worth a look.
3
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 14
Use $ARGUMENTS if provided
- Very specific: line 34
Run btca resources to see what's available, then btca ask -r <resource> -q "..." for library/framework questions
- Distinctive: line 43
Run the project's lint/test commands (check CLAUDE.md for exact commands)
- Distinctive: line 44
If lint/tests fail, fix the failures before committing
- Distinctive: line 45
If a failure cannot be fixed automatically, skip that fix and report it as Valid (unfixed) in the Phase 4 table
- Very specific: line 110
If the parent review left a separate top-level summary comment (Greptile's Greptile Summary issue-level comment, for example), leave it …
- Distinctive: line 125
Never dismiss a finding without reading the actual code first
- Distinctive: line 126
If unsure, err toward "Valid" — it's cheaper to fix than to miss a bug
“Common” is measured: kept by at least 6 in 10 of the 3 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Triage PR Review Comments
Fetch all review comments on the current PR, verify each finding against real code, fix valid issues, and push.
Phase 1: Gather Comments
- Determine the PR number:
- Use
$ARGUMENTSif provided - Otherwise:
gh pr view --json number --jq .number
- Use
- Fetch ALL comments (reviewers post in multiple places):
gh api --paginate repos/{owner}/{repo}/pulls/{pr}/reviews gh api --paginate repos/{owner}/{repo}/pulls/{pr}/comments gh api --paginate repos/{owner}/{repo}/issues/{pr}/comments - Extract unique findings — deduplicate across Copilot, Greptile, and human reviewers. Group by file and line.
Phase 2: Verify Each Finding
For EVERY finding, verify against real code before accepting or rejecting:
- Read the actual code at the referenced file:line
- Check if the issue still exists — it may already be fixed in a later commit
- Verify correctness using:
- Code analysis (read surrounding context, trace call paths)
- Run
btca resourcesto see what's available, thenbtca ask -r <resource> -q "..."for library/framework questions - Web search for API behavior, language semantics, or CVEs
- Classify each finding:
- Valid — real bug, real gap, or real improvement needed
- False positive — reviewer misread the code, outdated reference, or style preference
Phase 3: Fix & Ship
- Fix all Valid findings
- Run the project's lint/test commands (check CLAUDE.md for exact commands)
- If lint/tests fail, fix the failures before committing
- If a failure cannot be fixed automatically, skip that fix and report it as Valid (unfixed) in the Phase 4 table
git addonly changed files,git commitwith message:fix: Address PR review feedback - <one-line summary per fix>- Push:
gt submit(orgit pushif not using Graphite)
Phase 3.5: Reply on each thread, then resolve it
Pushing the fix isn't enough. Each inline review comment lives in its own thread that GitHub keeps showing as "Unresolved" until someone explicitly resolves it. The skill must close that loop for every Valid and False-positive finding so the PR view actually reflects what was triaged.
Workflow per finding:
- Reply on the thread with the verdict and evidence:
- Valid + fixed:
Fixed in <short-sha>. <one-line what changed>. - Valid + unfixed:
Valid, deferred to follow-up. Reason: <why>. - False positive:
False positive. <one-line evidence: file:line shows X, or doc link Y>.
- Valid + fixed:
- Resolve the thread only when the finding is Fixed or False positive.
Valid + unfixedthreads must stay open so the PR view continues to surface the real bug. Resolving them would let the PR look ready while the bug is still in the code. Leave the maintainer to close those threads when they file the follow-up.
GitHub's review threads can only be resolved via GraphQL (REST has no endpoint). The thread ID is a GraphQL node ID, not the REST comment ID, so fetch both together. Use --paginate so PRs with more than 100 threads are covered, the page size cap is per-request not total:
PR=<pr-number>
OWNER=<owner>
REPO=<repo>
gh api graphql --paginate -f query='
query($owner: String!, $repo: String!, $pr: Int!, $endCursor: String) {
repository(owner: $owner, name: $repo) {
pullRequest(number: $pr) {
reviewThreads(first: 100, after: $endCursor) {
pageInfo { hasNextPage endCursor }
nodes {
id
isResolved
comments(first: 1) { nodes { databaseId path line body } }
}
}
}
}
}' \
-F owner="$OWNER" -F repo="$REPO" -F pr=$PR \
--jq '.data.repository.pullRequest.reviewThreads.nodes[] | select(.isResolved == false) | {threadId: .id, commentId: .comments.nodes[0].databaseId, path: .comments.nodes[0].path, line: .comments.nodes[0].line}' \
> /tmp/triage-threads-$PR.jsonEach entry now has {threadId, commentId, path, line}. Match it against your Phase-2 finding map (by path + line or commentId). For each match:
# Post reply (REST endpoint for replies on a specific review comment)
gh api -X POST "/repos/$OWNER/$REPO/pulls/$PR/comments/$COMMENT_ID/replies" \
-f body="Fixed in $SHORT_SHA. <one-line>."
# Resolve the thread (GraphQL) — Fixed and False-positive only. SKIP for Valid+unfixed.
gh api graphql -f query="
mutation {
resolveReviewThread(input: {threadId: \"$THREAD_ID\"}) {
thread { isResolved }
}
}"If the parent review left a separate top-level summary comment (Greptile's Greptile Summary issue-level comment, for example), leave it alone, only inline review threads need resolving.
Cap: resolve only threads tied to findings you actually classified. Do not bulk-resolve unrelated threads (other reviewers, human discussion, follow-up questions that aren't from this triage round).
Phase 4: Report
Present a final summary table of ALL findings with verdicts:
| # | Source | File:Line | Finding | Verdict | Reason | Thread |
Last column: replied + resolved, replied + still open (e.g. waiting on reviewer), or n/a (no inline thread, only summary). If any thread stayed open, name it explicitly so the next pass picks it up.
Notes
- Never dismiss a finding without reading the actual code first
- If unsure, err toward "Valid" — it's cheaper to fix than to miss a bug
- For library/API questions, always use btca or web search — don't guess
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

