More screenshots

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIDeepSeekDeepSeekMuseMuse+22 more

    Receiving Code Review: Verify Before Implementing

    1

    Work through review feedback like an engineer: read it all first, verify each point against the code, accept what's right without theater, push back with evidence when it's wrong.

    $5

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned
    Receiving Code Review: Verify Before Implementing

    Receiving Code Review: Verify Before Implementing

    Example session with this skill installed

    My PR adds a CSV import for customer records. The reviewer left six comments and I can't tell which ones to just do. Can you triage them before I change anything?

    • Read your context and instructions
    • Compiled the receiving code review

    TRIAGE RECORD — PR: CSV customer import. Reviewer comments sorted before any code changed.

    1. "The parser loads the whole file into memory — a 2 GB upload will kill the worker." Verified against the code: readFile result is split in import.ts:38, no streaming anywhere; the worker has a 512 MB limit. Evidence 2, blast radius 1. VERDICT: correct and clear. Fix: stream-parse in chunks. Implemented first — it's the round's only blocker.

    2. "Use the csv-fast library instead of hand-rolling the parser." The observation (hand-rolled parsing is a liability) is fair. The prescription isn't checked: csv-fast isn't in the dependency set, and the hand-rolled part is only the chunk splitter — row parsing already delegates to the CSV parser the codebase standardized on. VERDICT: right observation, wrong fix. Action: keep the existing parser, delete the hand-rolled splitter in favor of its streaming API, reply states the catch was real and what changed instead.

    3. "This needs to handle duplicates." Ambiguous — duplicate rows inside one file, or re-importing a file already imported? The fixes are completely different (in-file dedupe key vs. import ledger). VERDICT: unclear. Specific question asked in the thread; no code written on a guess.

    4. "There's no file-type check — anyone can upload anything." Checked: the upload route passes through the shared acceptUploads middleware, which enforces the extension and MIME allowlist, with tests. VERDICT: wrong on the facts. Pushback reply cites the middleware and the test by name.

    5. "Row-level errors should be reported back to the user, not just logged." True and valuable — and a UX feature (error report format, storage, download) far beyond an import-mechanics PR. VERDICT: valid, out of scope. Follow-up issue filed and linked in the reply.

    6. "Nit: rename rows2 to parsedRows." Trivially correct. Fixed in the same commit as the rename batch.

    Six comments: one blocker fixed, one fix redirected, one question asked, one evidence-backed pushback, one filed follow-up, one nit. Zero implemented on faith, every thread gets its outcome reply.

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

    About this skill

    Most engineers handle code review feedback in one of two broken ways: implement everything the reviewer said and let the codebase drift toward whoever commented last, or defend every line and waste the review. This skill is the third way — a verification protocol. It starts with intake discipline: read the entire round before touching any of it, because comments interact, and separate each reviewer's observation from their prescription, since the observation is usually right even when the suggested fix is wrong. Then triage: every comment gets exactly one of five verdicts — correct and clear, right observation with the wrong fix, unclear, wrong on the facts, or valid but out of scope — recorded in a triage template so nothing evaporates. A two-axis rubric (evidence strength × blast radius) decides which comments get implemented, which get one cheap experiment first, and which get pushback with counter-evidence attached. Implementation runs in a fixed order — blockers and security first, then simple fixes, then refactors — with each fix tested individually so you know which change did what. The response style is specified just as tightly: no performative agreement, no thank-you theater, just the outcome per thread — fixed and where, pushed back and why, deferred and to which follow-up. A complete worked example runs five comments on one endpoint PR through five different verdicts. Edge cases cover unreachable reviewers, tooling comments, conflicting authorities, and security findings with exploit detail, and a pre-delivery checklist plus quality gate define when a round is actually closed rather than just quiet.

    Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Code Quality & Review.

    • It cannot verify comments that depend on access you don't have (staging data, production behavior, a private decision record) — the honest output there is the specific question to ask, which the skill will draft but cannot answer for you.
    • Verdict quality depends on the evidence you can reach; in a codebase with no tests, several rubric checks degrade to caller tracing and reading.
    • It governs how you receive feedback, not how to write it — reviewing other people's changes is a different discipline.

    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 buying a skill.

    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

    Yes, in the skill's form: pushback is evidence, not argument. "Checked — the validation runs in the middleware at auth.ts:31, here's the test" lets a senior reviewer correct their own comment in seconds. What changes with seniority is nothing; what kills pushback is making it about who's right instead of what's true.

    That's the conflicting-authorities case the skill names explicitly: stop, state both positions with the evidence for each, and let the person who owns the decision resolve it. Don't implement silently and don't refuse silently — an escalated, documented disagreement is a legitimate outcome.

    For a typical round, triage is minutes per comment — a grep, a test run, a caller trace. Implementing everything costs more: the wrong-fix and out-of-scope comments are where the rework, regressions, and bloated PRs come from. Verification is the cheap part of the round.

    New in Code Quality & Review