- Home
- Skills
- Code Quality & Review
- evidence code review
evidence code review
Review code changes — a diff, a branch, files, or pasted code — for real defects.
Free
Works with the AI Agents you already use
evidence code review
Example session with this skill installed
Review the changes in this PR diff. Pay special attention to the new withdrawal logic in wallet.py and the transaction handling.
- Read your context and instructions
- Compiled the evidence code review
I have reviewed the changes in wallet.py. The review focused on the new withdrawal logic and transaction handling.
Evidence Code Review
| File:Line | Severity | Finding | Failing Scenario | Confidence | Recommended Fix |
|---|---|---|---|---|---|
wallet.py:42 | Critical | Race condition in balance check | Two concurrent requests can pass the balance check before either subtracts the amount, leading to a negative balance. | High | Wrap the check and subtraction in a database transaction or use an atomic update. |
wallet.py:58 | Warning | Precision loss in fee calculation | Fees are calculated using floats, which can lead to rounding errors in currency distribution. | Medium | Use Decimal or store values in cents/integers. |
wallet.py:102 | Warning | Unreleased lock on exception | If notify_user() raises an exception, the wallet lock is never released, deadlocking future attempts. | High | Use a try...finally block to ensure the lock is released. |
CHECKED AND OK
wallet.py:22: The SQL interpolation uses a predefined whitelist for sorting, preventing injection.wallet.py:85: The recursive call is bounded by a maximum depth check, preventing stack overflow.
UNVERIFIED
auth_service.verify(): I cannot see the implementation of the auth check; if it fails open on timeout, the withdrawal logic is bypassed.queue.publish(): The behavior of the message queue when full is unknown; it may lead to silent failures.
Next steps
- Implement a database-level transaction for the withdrawal logic in
wallet.py:42. - Refactor fee calculations to use the
Decimallibrary to avoid float precision issues. - Wrap the notification logic in a
finallyblock to prevent lock exhaustion.
Connects securely to your tools. The creator never sees your data.
About this skill
Review code changes — a diff, a branch, files, or pasted code — for real defects. Every finding carries a line citation and a concrete failing scenario; anything unproven is labeled unverified. Use for pre-merge review, 'review my changes', bug hunts, and security passes.
What it does
- Identifies logic errors, security flaws, and concurrency issues, with a failing scenario for each finding.
- Traces invariants across inputs and state changes in the changed functions.
- Checks for off-by-one errors, resource leaks, and injection risks.
- Reports findings with line citations, severity levels, and concrete fixes.
- Lists unverified risks that depend on code outside the current context.
How it works
- Read the diff to distinguish visible code from assumed external dependencies.
- Sweep for failures across auth, money, concurrency, and external input handling.
- Map the invariants that must hold for each changed function.
- Check specific input sequences or exception paths to confirm defects.
- Output verified findings, checked items, and unverified external risks.
Frameworks & tools
Language-agnostic; works with Python, JavaScript/TypeScript, Go, Rust, Java, and C++, including code handling SQL, shell commands, financial math, or concurrent operations.
Use cases
- Pre-merge review of changes before they reach the main branch.
- Security passes focused on injection, auth bypass, and data leakage.
- Refactoring validation to check that invariants are preserved.
- Bug hunts in concurrent or state-heavy logic.
Known limitations
It reviews the code it is shown and doesn't execute anything, so runtime-only bugs are out of scope. Does not perform style, naming, or linting checks. If there is not enough context to verify cross-file dependencies, findings are moved to the unverified list.
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.
Focuses on logic and security defects only. Skips style nitpicks. Effectiveness is limited by the amount of surrounding code context provided.
How to use
- 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
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 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
Free to use
No account required to browse
Trust & safety
Security scan
Security scan passed 2 days ago
- Free with an Agensi account