Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    Review Finding Proof — PR Comment Verdicts

    1

    Find which PR comments need a fix. Check each claim against current code, its contract and a focused reproduction.

    Free

    2 usesSecurity scanned
    Review Finding Proof — PR Comment Verdicts

    Review Finding Proof — PR Comment Verdicts

    Example session with this skill installed

    Synthetic local example. Check these two comments against the supplied route and contract. Stay read-only.

    
    # router.py
    def route(user_id, query):
        if user_id is None:
            return {"status": 401, "error": "sign in required"}
        limit = int(query.get("limit", "20"))
        if limit < 1 or limit > 100:
            return {"status": 400, "error": "invalid limit"}
        return {"status": 200, "limit": limit}
    

    The contract says anonymous requests return 401 before reading query values. Signed-in requests must return 400 for any limit that is not an integer from 1 through 100, without raising.

    • R1, P2: signed-in limit=abc raises instead of returning 400.
    • R2, P1: anonymous limit=abc is parsed before the authentication guard and can raise before returning 401.

    The comments and inspected fixture both refer to revision 22917aaba598aa284022223b601c2f03c95358ee. Use the local Python probe to decide each claim.

    • Read your context and instructions
    • Compiled the review finding proof

    Review decision: HOLD for R1

    Actual synthetic local run. Both comments were checked at revision 22917aa; no production PR or web framework was exercised.

    FindingVerdictEvidenceNext action
    R1 · P2Verifiedrouter.py:4 calls int("abc"). The probe raised ValueError; CONTRACT.md:5 requires a 400 response.Handle invalid integer input before returning the route result.
    R2 · P1 as reportedRefutedrouter.py:2–3 returns 401 before line 4. The same anonymous query returned {"status":401,"error":"sign in required"}.No change is supported for this claim.

    The valid signed-in control returned status 200 with limit 20. The failure is specific to malformed signed-in input, rather than all requests to this route.

    Recommendation: resolve R1 before relying on these supplied findings for a merge decision. R2 supplies no evidenced blocker under its stated conditions. This report does not approve the rest of the PR.

    The focused probe exited 0 after recording all three observations. Repository file bytes, modification times and modes were unchanged. No code was edited, no thread was resolved and no merge was performed.

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

    About this skill

    A review comment sounds serious. Is it true in the code you are about to merge?

    Review Finding Proof helps you decide what deserves a fix before you spend time acting on the comment. Give your agent the findings, the relevant diff and the current revision. It follows the affected code path, checks guards and callers, and uses a safe focused reproduction where possible.

    A decision for each comment

    • Verified: the claimed problem is supported by the code and contract or a causal reproduction.
    • Refuted: a guard, contract or counterexample disproves the claim under its stated conditions.
    • Unverifiable: the report names the missing evidence and the next check.

    You receive exact file and line references, assessed severity and a merge recommendation for the supplied findings. When a PR moves, old comments are rechecked against the current head. A verified blocker or unresolved high-severity claim keeps the recommendation on HOLD.

    This instruction-only workflow leaves code, review threads and PR state unchanged. Start with the comments you need to decide.

    It checks supplied findings only and cannot certify the whole PR. A missing revision, contract, or safe environment can leave a finding unverifiable. A passing test alone does not clear a blocker.

    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

    2 uses

    Times used to date

    Free forever

    No account required to browse

    Trust & safety

    Security scan

    Security scan passed 25 days ago

    • Free with an Agensi account

    Permissions

    Terminal / Shell
    Read Files

    Read repository files, diffs, and supplied findings. A read-only PR connector or safe local test environment may help; no PR, repository, or external-service write permission is needed.

    Listed1 month ago
    Updated25 days ago

    What's inside

    Frequently Asked Questions

    The complete comments, their referenced revision, the current diff or code, and the relevant intended behavior. A read-only PR connector is optional; local files can be enough.

    It produces a read-only finding ledger and proposed next actions. Code changes, reviewer replies, thread resolution and merge actions belong to a separately authorized workflow.

    Yes. It records the comment revision and inspected head, then distinguishes a historical problem from the current code. Old line numbers are not treated as current evidence.

    A traceable static argument can establish a narrower conclusion. Otherwise the finding stays unverifiable with the missing evidence named; a passing unrelated test does not clear it.

    New in Code Quality & Review