More screenshots

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    Diff Risk Brief — Review the Risky Files First

    1

    Reads a unified diff and ranks the files that deserve careful review first: auth, money, migrations, lockfiles, large edits and changes with no test moving with them.

    Free

    Security scanned
    Diff Risk Brief — Review the Risky Files First

    Diff Risk Brief — Review the Risky Files First

    Example session with this skill installed

    git diff main...HEAD > change.diff
    python3 scripts/diff_brief.py change.diff --fail-on high

    diff --git a/src/auth/session.ts b/src/auth/session.ts
    --- a/src/auth/session.ts
    +++ b/src/auth/session.ts
    @@ -1,6 +1,30 @@

    -const token = session.token;
    +const rotated = rotate(session.token, key);
    +if (!session.expiresAt || session.expiresAt < now) {
    
    • throw new AuthError('expired session');
      +}
    • Read your context and instructions
    • Compiled the diff risk brief

    diff-risk-brief 8 files 987 changed lines
    score 21/100 grade F findings 12
    high 5 medium 3 low 4

    REVIEW ORDER

    1. src/search/rank.ts high 457 lines 457 changed lines in one file.
    2. package-lock.json high 220 lines Dependency or dependency lockfile change.
    3. src/billing/refund.ts high 179 lines Touches money movement (179 changed lines).
    4. db/migrations/0042_add_invoices.sql high 41 lines Touches schema or migration change (41 changed lines).
    5. src/auth/session.ts high 33 lines Touches authentication or session code (33 changed lines).
    6. config/prod.ts medium 15 lines Touches configuration.
    7. src/utils/format.ts low 23 lines No matching test file changed with this source edit.
    8. README.md ok 19 lines no risk rule fired

    VERDICT: REVIEW THE RISK FILES FIRST

    2. package-lock.json

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

    About this skill

    Most reviews read a diff file by file. Reviewers do not have equal attention, and a README edit does not deserve the same reading as a refund path. This turns the patch into a review order before anyone opens the pull request.

    It reads a unified diff and ranks each changed file by evidence a reviewer can check:

    • Files touching auth, sessions, payments, billing, migrations or deployment are moved to the front
    • Large edits are split out: 150+ changed lines is a prompt, 400+ is a finding
    • New source files with no matching test in the diff are named
    • Deleted test files are flagged before the coverage quietly falls
    • Lockfiles, manifests, environment files and production config get their own check
    • Deletion-heavy files are marked, because removed branches are easy to miss
    • A source edit with no test moving with it becomes a question for the author

    The output has two blocks: the order to read, then the findings with a fix written as a review action. Every finding is something the patch shows, not a guess about the author.

    The shipped failing diff is eight files and 987 changed lines: 12 findings, five high, a refund path, a schema migration, a 457-line ranking change, a lockfile and an auth file. Score 21/100, exit 1. The clean diff is two files with a matching test and scores 100.

    Good to know: this is static analysis of the patch. It cannot know whether a change is correct, who owns the file, or whether a test already exists outside the diff. "No test in this diff" is a question, not proof. The 200/250-line review evidence comes from observational studies, not a universal cutoff. It does not clone the repository or execute code. Use it to decide where to look, not as a defect prediction.

    Cannot verify tests outside the current diff. Static path and size rules only. Excludes docs, images, and binary changes. Requires Python 3.10+.

    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

    Free forever

    No account required to browse

    Trust & safety

    Security scan

    Security scan passed 1 day ago

    • Free with an Agensi account

    Requires

    Git

    Permissions

    Terminal / Shell
    Read Files
    Write Files
    Environment Variables
    Listed1 day ago

    What's inside

    Frequently Asked Questions

    This skill analyzes your unified diffs using a deterministic scoring engine to surface high-risk changes (like auth or payments) first, ensuring you spend your peak focus on critical logic rather than trivial edits.

    It is a standalone Python 3.10+ script with zero external dependencies, making it compatible with any AI agent that can execute Python or be integrated directly into GitHub Actions and GitLab CI/CD pipelines.

    No, this skill uses static analysis and deterministic rules to score risks, providing a consistent metric that avoids the hallucinations or inconsistent logic often found in generic LLM prompts.

    The skill flags files that exceed 150 or 400 lines of changes and detects when new source files are added without corresponding changes to test directories, prompting a PR split or better coverage.

    You receive the core Python engine, documentation for CI/CD integration, and a guide for interpreting the risk categories; updates are delivered via the marketplace as we refine the scoring heuristics.

    It currently supports standard unified diff and patch files generated by Git; it does not analyze binary files, generated code, or perform deep semantic parsing of logic.

    New in Code Quality & Review