web-security-review
Skill0 starsChanged 12 days agoMITView on GitHub
Security review of a web app, API or SPA for the two bugs that matter most in practice — leaked secrets (API keys, JWTs, tokens, DB URLs in code, .env, git history and built JS bundles) and broken access control (IDOR/BOLA, missing ownership checks, role checks only in the UI, privilege escalation) — plus JWT/session weaknesses. Works for developers checking their own code before release (findings + fixes) and for bug bounty hunters working an authorised scope (test plan + a report that survives triage). Use it whenever someone asks to check, audit or review an app's security, find leaked keys, look at JS bundles, test authorisation/roles/IDOR, inspect a JWT, prepare for a pentest, or write a HackerOne/Bugcrowd report — "проверь безопасность", "найди утечки ключей", "IDOR", "BAC", "напиши отчёт для HackerOne" — even if they don't say "security review".
- Reads credentials
1 other project uses a version of this.
Adopted by
Most recently taken up by
- ai-moto/web-craft-skillstheir own version, first seen Aug 23, 2026
Other versions1 project uses a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 1
- ai-moto/web-craft-skills★ 0same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repositoryOne other file it gives its agents.
- prompt-coachSkill
The file
We split this skill into the rules it gives an agent
3 rules. About 0 are common, 3 are distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈3
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 17
If it is neither (someone else's system with no programme or permission), help with defensive learning on a local or intentionally …
- Distinctive: line 69
Not a duplicate of something the programme says is known?
- Distinctive: line 72
If it fails these, say so plainly and suggest how to strengthen it (find the impact, chain it) or drop it.
≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Web Security Review
Two bug classes cause most real damage in web apps and most accepted bug-bounty reports: secrets that leaked to where they shouldn't be and access control that trusts the client. This skill goes deep on those instead of shallowly on everything.
0. Establish whose system this is (one question, then go)
The work is identical for both audiences; the output differs.
- Developer mode — it's the person's own code or their employer's. You can read source, run the app, change code. Output: findings with fixes, ordered by risk.
- Hunter mode — a bug bounty / VDP / pentest target. Only test what the program's scope allows, only with the person's own accounts, within rate limits. If they haven't said which program, ask for the scope and rules (or the program page) before planning any live testing. Output: a test plan and, when something is confirmed, a report.
If it is neither (someone else's system with no programme or permission), help with defensive learning on a local or intentionally vulnerable app instead, and say why: testing systems without authorisation is illegal in most countries and burns the person's reputation on platforms.
Reply in the person's language; write bug-bounty reports in English unless the programme says otherwise.
1. Secrets
Where to look
- Source and config:
.env*,config/, CI files (.github/workflows,.gitlab-ci.yml), Docker files, mobile configs, test fixtures. - Git history: secrets "deleted" in a later commit are still in history.
- Built front-end bundles (
dist/,build/,.next/static/, downloaded*.jsand*.js.map): front-end env vars (NEXT_PUBLIC_*,VITE_*,REACT_APP_*) are compiled into public files. This is the most common real-world leak. - Source maps (
.map) reveal original code, comments and internal endpoints.
Run the scanner (local files only; values are redacted in the output):
python3 <skill-dir>/scripts/scan_secrets.py <folder-or-files> [--git] [--show-public]
python3 <skill-dir>/scripts/jwt_inspect.py <token>In hunter mode, get the JS files of the in-scope app with normal browsing (DevTools → Sources, or save the files the page loads) and scan the saved copies. Don't crawl beyond scope.
Then think — the scanner only finds candidates. For each hit decide:
- Is it secret at all? Some keys are public by design (Stripe
pk_live, Google Maps / Firebase web keys, Sentry DSN, analytics write keys). They matter only if unrestricted or they unlock non-public data. Reporting them as-is is a classic N/A. - Is it live and what does it unlock? An expired JWT or a revoked test key has little impact. A long-lived admin JWT, a cloud key, a DB URL or a payment secret is critical. In hunter mode, prove impact with the least invasive call (e.g. a read-only "who am I" endpoint), never by reading or changing real user data, and stop as soon as impact is clear.
- What is the fix? Rotate first (the old value is compromised forever), then move it server-side, and add a CI check so it can't come back.
See references/secrets.md for per-provider notes and least-invasive verification.
2. Access control (BAC / IDOR / privilege escalation)
The core question for every request: does the server check that this user may do this to this object? The UI hiding a button is not a check.
Developer mode — code review. Map the routes/handlers, then for each one that takes an object ID or performs a privileged action, find the line that enforces ownership or role. No line → finding. Pay attention to:
- lookups by ID without scoping to the current user/tenant (
findById(id)vsfindOne({id, owner: user.id})); - role checks only in front-end code or only on some HTTP methods (GET protected, PUT not);
- mass assignment (the client can send
role,owner_id,is_admin,org_id); - secondary endpoints that bypass the main check: exports, downloads, search, bulk operations, GraphQL resolvers, websockets, older API versions, mobile API;
- multi-tenant scoping in background jobs and caches.
Hunter mode — two-account testing. Create account A (attacker) and B (victim) — and a low-role and high-role user if roles exist — all owned by the person. Capture B's requests, replay them with A's session, compare. references/access-control.md has the full test matrix (horizontal, vertical, method switching, ID formats, tenant boundaries, GraphQL, mass assignment) and how to show impact without touching real users.
3. JWT and sessions
Check with jwt_inspect.py and the app's behaviour: alg: none accepted, HMAC with a weak or leaked secret, kid/jku trusted from the token, no exp or very long life, tokens still valid after logout/password change, token in a URL or localStorage together with an XSS, role claims trusted without server lookup. Details: references/jwt-sessions.md.
4. Before calling something a finding
Reviewers reject most reports for the same few reasons. Run through this before writing it up:
- In scope (asset and vulnerability type)? Not in the programme's exclusions?
references/exclusions.mdlists the usual out-of-scope items (missing headers, rate limiting, self-XSS, enumeration, public keys…). - Reproducible by someone else from your steps alone?
- Impact demonstrated on your own accounts, not just theorised?
- Not a duplicate of something the programme says is known?
- For developers: is it exploitable in the deployed configuration, or only in dev?
If it fails these, say so plainly and suggest how to strengthen it (find the impact, chain it) or drop it. Sending weak reports costs a hunter their platform reputation (Signal/accuracy), which controls access to private programmes.
5. Output
Developer mode — a findings list, worst first:
## <Severity> — <short title>
Where: <file:line / endpoint>
What: <one paragraph>
Why it matters: <concrete consequence for this app>
Fix: <specific change, with a code snippet if useful>End with: what to rotate now, and one CI/automation step to prevent recurrence (e.g. run scan_secrets.py or gitleaks on every PR).
Hunter mode — a test plan while exploring; for a confirmed bug, a report using references/report-template.md. Keep real users' data, live secrets and tokens out of the report beyond the redacted minimum the programme needs.
Boundaries
This skill is for finding and fixing weaknesses with permission. Don't write exploits that go beyond demonstrating impact, don't brute-force credentials or secrets on live targets, don't run scanners or high-volume automation against programmes that forbid it, and don't access, keep or share other people's data. If a step would cross one of these, explain why and offer the safe alternative.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

