More screenshots
Works with the AI Agents you already use
Lazy Review: Findings Only, No Noise
Review diffs for actionable findings only: scope the change, trace data flow end to end, challenge security and compatibility, report findings.
$5
Lazy Review: Findings Only, No Noise
Example session with this skill installed
Review this PR: it adds GET /invoices/:id to our API. Here is the diff.
- Read your context and instructions
- Compiled the lazy review
Scope first: the PR is supposed to let users view their own invoices. Changed files: one route, one query helper, tests. The 40-file claim did not apply — this was a 3-file change with one file that matters. Missing from the diff: no test for the cross-user case, which is itself a finding. Trace: input is the invoice id from the URL; validation happens in the query helper, three layers deep, where half the callers skip it; state change is a read; sink is the JSON response. Error path: a missing invoice returns 404, but a forbidden one returns the invoice — the divergence. Challenge: auth/authz failed on evidence. Ran the endpoint as user A with user B's invoice id and got user B's invoice back — attached output as evidence. Injection clean (parameterized query), schema unchanged, no secrets in logs, no concurrency change. Finding, in the required shape: FINDING — unauthenticated cross-user invoice read. PATH — src/routes/invoices.ts:42, GET /invoices/:id has no ownership check. EVIDENCE — curl as user A returns user B's invoice. IMPACT — data exposure across the tenant boundary. FIX — add the ownership check matching the pattern in src/routes/orders.ts:31. Report: findings first, one high-severity finding, one question (should the 404/403 distinction stay, or return 404 for both?). No summary of what the PR does, no padding, no restatement. Approval waits on the fix or a convincing rebuttal.
Connects securely to your tools. The creator never sees your data.
About this skill
A review has one job: find problems the author cannot see. Everything else — praise, style preferences, narration of the diff — is noise that buries the findings. Lead with findings. If there are none, say so and stop. The loop is scope, trace, challenge, prove, report. Scope bounds the change first: what it is supposed to do, which files matter, and what is missing from the diff. Trace follows the data end to end — input, validation, state change, sink — watching for untrusted data, atomicity, leaks, and error paths. Challenge runs the adversarial checklist: auth, injection, schema drift, secrets, compatibility, concurrency, error handling, performance. Prove turns a suspicion into a finding with path, evidence, impact, and fix — or labels it a question. Style is never a finding; if the repo has a formatter, style is the formatter's job.
Style is never a finding — if the repo has a formatter, style is the formatter's job.
A question is not a finding: without evidence, it is labeled as a question to the author, not reported as a defect.
Large diffs with high blast radius should be split by the author; reviewing 2000 lines carefully is a human-limits issue, not a skill issue.
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 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 3 days ago
Permissions
No special permissions declared or detected