More screenshots

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    Process PR Comments: Review Feedback Workflow

    1

    Clear a PR's review backlog thread by thread: verify every comment against the actual code, score it, decide accept, clarify, push back, or defer, then fix, reply, and extract the rule.

    $5

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned
    Process PR Comments: Review Feedback Workflow

    Process PR Comments: Review Feedback Workflow

    Example session with this skill installed

    My PR "Add webhook retries" (#217) came back with five review threads and I can't tell which ones actually matter. Work through them: tell me what to fix, what to push back on, and what to ask about.

    • Read your context and instructions
    • Compiled the process pr comments

    TRIAGE RECORD — PR #217, produced by the skill's workflow (branch synced first; one thread anchor had shifted).

    Thread 1 — retry.ts:54, "the backoff can exceed the job timeout on the third retry." Read the code: backoff is 2^n seconds with no cap, job timeout is 30s, third retry waits 32s before the attempt even starts. Impact 2 / confidence 2 → ACCEPT. Fix: cap backoff at 20s. Reply draft: "Confirmed — uncapped backoff passes the timeout on retry three. Capping at 20s."

    Thread 2 — retry.ts:12, "this duplicates the delay helper in jobs/util.ts." Read both: the util helper sleeps a fixed duration and doesn't support the jitter this code needs; signatures and behavior differ. Impact 1 / confidence 0 → PUSH BACK, with evidence. Reply draft: "Checked jobs/util.ts — its helper is fixed-delay with no jitter support, so this isn't a duplicate. Happy to unify if we extend the helper, but that's a separate change."

    Thread 3 — webhook.ts, "payloads over 1 MB should be rejected before queueing." Valid, and the current PR doesn't touch payload validation — the queue path it names is pre-existing. → DEFER, follow-up issue filed and linked in the reply.

    Thread 4 — retry.ts:71, "should a 429 from the receiver reset the retry count?" Genuinely ambiguous: depends on whether the team treats rate-limiting as receiver-not-ready or as a failed delivery. → CLARIFY. Reply draft: "Before I implement either way — do we treat a 429 as 'not ready, keep the full retry budget' or as a failed attempt that consumes one?"

    Thread 5 — naming nit on an exported function. Repo convention requires full names on exports. Impact 0 / confidence 2 → ACCEPT, two-minute rename.

    Order of work: thread 1 fix first, rename bundled with it, replies drafted for all five, nothing posted until you approve.

    Connects securely to your tools. The creator never sees your data.

    About this skill

    A PR comes back with eleven review threads and no signal about which ones matter. Some are stale, one is a real correctness bug, two are the same convention violation in different files, and one is a valid point about code the PR never touched. Treating them as a flat list means either blind-applying everything (reviewers are wrong sometimes) or arguing everything (and burning the goodwill the review represents). This skill runs the full workflow instead: collect every unresolved thread including replies, sync the branch before judging anything, then triage each comment against the code it points at — does the issue actually exist, what category is it, is it in scope for this PR. Each thread gets scored on impact and confidence so the correctness bug in thread 9 gets handled before the style nits in threads 1–3, and each one lands in exactly one of four outcomes: accepted, clarified, pushed back with evidence, or deferred with a follow-up link. The skill ships a triage record template, reply shapes for all four verdicts, and a worked example of a six-thread PR sorted end to end. It also covers the step most workflows skip: after the fixes land, extract the principle behind the accepted comments and write it down in one place, so the same comments stop recurring. Guardrails are explicit throughout — replies are drafted, never posted without instruction; nothing is committed or pushed without the author's word; and a fix that changes behavior beyond the ask stops and surfaces as its own decision.

    • It needs the actual code: verdicts are only as good as the access to the branch and the conventions file; without them, confidence scores drop and more threads land in clarify.
    • It processes human review judgment; it does not replace CI — failing checks are noted, not diagnosed.
    • Reply drafts follow the evidence-first etiquette in the skill, but team tone varies; expect to adjust wording before sending.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before any payment.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    Fresh listing

    Recently published to Agensi

    Money-back guarantee

    Doesn't do what its page promises? You get your money back.

    Trust & safety

    Security scan

    Security scan passed 24 days ago

    Permissions

    No special permissions declared or detected

    Listed24 days ago
    Updated21 days ago

    What's inside

    Frequently Asked Questions

    No — by design. The skill drafts every reply and prepares the fixes, but posting, resolving threads, committing, and pushing all wait for your explicit go-ahead. The branch and the conversation belong to the PR author.

    That happens, and the workflow expects it. Each comment is verified against the anchored code, tests, and repo conventions before any verdict. A comment that fails on the facts gets a push-back verdict with a drafted reply that cites the evidence — the line, the test, or the convention — instead of arguing from seniority.

    Yes — threads are scored on impact and confidence and processed in that order, batched by file rather than thread order. Correctness and security threads go first; style nits get fixed silently or answered with one line pointing at the linter config.

    New in Code Quality & Review