Works with the AI Agents you already use
Web App Hygiene Audit
A meticulous, breadth-first review of a web project that surfaces **obvious holes, misconfigurations and sloppy implementations of core components** - not exotic exploit chains.
$6
Web App Hygiene Audit
Example session with this skill installed
Audit this repo before we launch. It's a Django/DRF backend with a Next.js frontend, Postgres, Stripe checkout and S3 uploads. Two test accounts are available on staging. Focus on access control on the org APIs and the checkout flow.
- Read your context and instructions
- Compiled the web app hygiene
- Generated the data export
Summary
Depth: standard. Stack: Django 4.2 + DRF (api/) and Next.js 14 (web/), audited per pack. Coverage: 132 of 146 applicable checks answered (90%) - 14 UNVERIFIED need staging runtime access.
Top findings
| Sev | ID | Finding | Evidence |
|---|---|---|---|
| Critical | AC-04 | Org endpoints lack tenant predicate: any authenticated user can read/write any org's objects | api/views/org.py:38 |
| High | AC-02 | Checkout price accepted from client body and trusted server-side | api/serializers/checkout.py:22 |
| High | SEC-03 | Stripe webhook signature not verified; replay possible | api/views/webhooks.py:11 |
| Medium | INJ-05 | Upload filename joined into filesystem path without sanitization | api/services/files.py:57 |
| Medium | Q-BLOCKER | Concurrent coupon redemption can double-apply discount (no select_for_update) | api/models/coupon.py:81 |
Coverage ledger
- Generic checklist: 98 checks - auth 12/12, access 9/11, injection 14/14, browser/http 16/18, config/secrets 15/15, robustness 21/22 answered
- django-drf pack: 30 checks answered; nextjs pack: 22 answered; boundary stated in report appendix
Next steps
- Add org membership predicate to every org queryset (canonical fix in report F-1)
- Compute price server-side from product IDs; verify webhook signatures
- Hand the 14 UNVERIFIED items to a human with the staging instructions in the report
Full report: audit/AUDIT-REPORT.md - ledger: audit/coverage.csv
coverage.csv
CSV · data export
Example file from a real run - the skill writes it into your workspace.
Connects securely to your tools. The creator never sees your data.
About this skill
A meticulous, breadth-first review of a web project that surfaces obvious holes, misconfigurations and sloppy implementations of core components - not exotic exploit chains. It reads like a nitpicky senior appsec + backend reviewer: nothing is safe until it is seen in code, config, or an actual response.
What you get
audit/AUDIT-REPORT.md- summary, fix-first list, severity-ranked findings with exactfile:lineevidence, coverage ledger and an unverified listcoverage.csv- one row per check ID with status (PASS / FAIL / N/A / UNKNOWN) and evidence locationfindings.json+evidence/- machine-readable findings and redacted raw evidence for tracking and CI- Security findings (Critical to Info) separated from quality findings (Q-BLOCKER / Q-MAJOR / Q-MINOR)
How it works
- Inventory & trust map - routes, forms, roles, jobs, webhooks, storage and admin surfaces, plus a stack fingerprint and trust-boundary map
- Stack pack - loads a framework pack before the sweep: Next.js/React, Express/Fastify/NestJS, Django/DRF, Laravel, FastAPI or WordPress/WooCommerce, each with idiomatic greps and a "do not flag" list that prevents false positives
- Checklist sweep - auth, sessions, access control/IDOR, input validation, injection sinks, XSS/CSRF/CORS, headers, secrets, debug endpoints, uploads, SSRF, caching, data layer, concurrency, business logic - every check answered, never silently skipped
- Per-endpoint cards - 12 questions per endpoint: authn actually applied, ownership predicates, mass assignment, validation, sinks, encoding, rate limits, idempotency, failure modes, caching, logging, contract match
- Core-flow edge matrix - 10 edge cases per critical flow (signup, login, reset, checkout, upload, admin): duplicates, concurrency, replays, boundaries - where double charges and lost updates actually live
- Report - dedupe by root cause, false-positive guards applied, severity + confidence, partial coverage stated explicitly
Guardrails
Non-intrusive by design: no exploitation, no brute force, no attack payloads, no destructive or state-changing requests outside a disposable test environment. Runtime probes are limited to read-only GET/HEAD/OPTIONS at max 2 requests per second, with benign markers and owned test accounts only. Secrets and PII are redacted. Intrusive checks are marked UNVERIFIED with exact instructions for a human instead of performed. Your code is not modified - patches are proposed, applied only if you ask.
Requirements
- Repo access to the web project (white-box mode needs nothing else)
curlfor the optional grey-box probe script- Optional: a staging URL and two owned test accounts for runtime confirmation
Known limitations
- Static-first: most checks are answered from code; runtime observation confirms but never replaces reading the code
- Black-box mode (URL only) has limited coverage, stated explicitly in the report
- No-match stacks (Rails, Spring, Go...) get the generic checklist with reduced precision, honestly noted
- Finds obvious holes and broken core components - it is not an exhaustive exploitation exercise
Not a penetration test: no exploitation, brute force, fuzzing or destructive requests - intrusive checks are marked UNVERIFIED with human instructions instead of performed. Black-box mode (URL only) has limited coverage, stated in the report. Runtime probes obey a fixed budget: read-only GET/HEAD/OPTIONS, max 2 requests per second and about 200 per run. Stack packs cover Next.js, Node, Django, Laravel, FastAPI and WordPress; other stacks get the generic checklist with reduced precision, honestly noted. The audit does not modify your project - patches are proposed in the report and applied only if you ask.
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 today
Requires
Permissions
Allowed Hosts
File Scopes
White-box mode only reads your repo and writes the report to ./audit/. Grey-box mode additionally runs passive_probe.sh: low-rate read-only GET/HEAD/OPTIONS requests (max 2 req/s, about 200 per run) against a host you own, plus at most one login to an owned test account. No exploitation, no payloads, no brute force.