securing-applications
Skill0 starsChanged 5 months agoMITView on GitHub
Use when building or reviewing anything that handles untrusted input, authentication, authorization, secrets, user data, PII/privacy, payments, file uploads, CORS/headers, external dependencies, LLM/AI features, or deployment config — and as a mandatory gate before shipping or claiming a feature done. Covers OWASP Top 10:2025 + LLM risks + data privacy. Stack- and agent-agnostic.
In this repository4 other files it gives its agents.
The file
We split this skill into the rules it gives an agent
6 rules. About 0 are common, 5 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈5
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 12
Core principle: Never trust input, never trust the caller, never trust defaults.
- Distinctive: line 41
No secrets in client-side code, logs, error messages, or URLs.
- Distinctive: line 46
Prompt injection: treat ALL model input (user text, retrieved docs, tool output) as untrusted.
- Distinctive: line 47
Treat model OUTPUT as untrusted: never exec/eval it, never render unsanitized, never trust it for authz decisions.
- Very specific: line 56
Retention: delete data past its retention window (cleanup jobs — see operating-production-services).
- Distinctive: line 60
Do NOT mark a feature/site done until (tick = tested with evidence, not assumed):
≈ “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
Securing Applications
Overview
Security is a gate, not a feature — it blocks "done", it isn't an optional add-on. Built on OWASP Top 10:2025 (current edition). Most breaches are boring: missing access checks, misconfig, leaked secrets, unpatched dependencies. Cover the basics relentlessly.
Core principle: Never trust input, never trust the caller, never trust defaults. Verify at every boundary.
Announce when securing/reviewing: "Using securing-applications: checking against OWASP Top 10:2025."
When to Use
- Building: auth, input handling, file upload, payments, admin features, anything touching user data.
- Configuring: deploy, server, cloud, CORS, headers, env.
- Reviewing: before shipping any feature; pre-launch; dependency updates.
OWASP Top 10:2025 — risk → defense → how to check
| # | Risk | Defense | Check |
| A01 | Broken Access Control (#1) | Authorize EVERY request server-side per resource & action; deny by default; never trust client role/ID; check ownership per object (BOLA); reject unexpected fields (mass assignment); validate outbound URLs (SSRF — allowlist, block link-local & cloud metadata 169.254.169.254) | Access another user's object by ID across ≥2 users → 403; unauthenticated request → deny; POST an extra isAdmin field → ignored; SSRF to metadata IP → blocked |
| A02 | Security Misconfiguration | Harden defaults; full security headers (CSP, HSTS, X-Content-Type-Options, X-Frame-Options/frame-ancestors, Referrer-Policy, Permissions-Policy, COOP/COEP); disable debug/dir-listing; least-privilege cloud IAM; CORS: never reflect arbitrary Origin, never * + credentials | Scan headers (securityheaders.com-class); review prod config diff; no 0.0.0.0 open ports, no default creds; CORS not wildcard-with-creds |
| A03 | Software Supply Chain Failures (rising) | Pin deps (lockfile, frozen install in CI); scan for known CVEs; verify package integrity; pin CI actions by SHA; generate SBOM; beware typosquats/compromised maintainers | Run dependency audit (e.g. npm audit --audit-level=high, pip-audit); CI fails on critical CVE; review new deps before adding |
| A04 | Cryptographic Failures | TLS everywhere; hash passwords with argon2id (tuned params) or bcrypt; encrypt sensitive data at rest; no homemade crypto; rotate keys; monitor cert expiry | Grep for weak algos (md5/sha1 for passwords); confirm HTTPS-only + HSTS; secrets encrypted at rest |
| A05 | Injection (SQL/NoSQL/cmd/XSS) | Parameterized queries/ORM; validate+sanitize input; escape/encode output (XSS); never string-concat queries/commands; CSRF: SameSite=Lax/Strict cookies + per-request token on state-changing routes; file uploads: validate magic bytes, AV-scan, store non-executable off-origin | Injection payloads blocked; no string-built SQL; output encoded; CSRF token required on writes; malicious upload rejected |
| A06 | Insecure Design | Threat-model before building; security requirements as acceptance criteria; rate-limit as anti-abuse (credential stuffing, scraping); abuse-case thinking | Documented abuse cases for auth + payments exist; limits/quotas designed in |
| A07 | Authentication Failures | Strong auth (MFA option); secure session mgmt; lockout/throttle on login; no creds in URL; rotate/expire tokens; JWT: validate alg/aud/iss/exp server-side, reject alg=none, plan revocation | Brute-force login → blocked; sessions expire; tokens not in logs; tampered/alg=none JWT → rejected |
| A08 | Software/Data Integrity Failures | Verify update/deploy signatures; protect CI/CD pipeline; no unsigned/auto-updates from untrusted source; validate deserialized data | Pipeline access controlled; artifacts signed; no untrusted deserialization |
| A09 | Security Logging & Alerting Failures | Log auth events, access-control failures, server errors (structured); alert on anomalies; protect & retain logs; NEVER log secrets/PII | Are failed logins logged + alerted? Secrets absent from logs? |
| A10 | Mishandling of Exceptional Conditions (new) | Fail closed (deny on error, never grant); handle all error paths; no stack traces/internal detail to users; consistent error responses | Force errors → app denies access, shows generic message, doesn't crash open |
Secrets — non-negotiable
- Secrets in environment/secret-manager, never in source or commits. Scan history before pushing.
- Separate dev/staging/prod credentials. Rotate on exposure and on a schedule.
- No secrets in client-side code, logs, error messages, or URLs.
LLM / AI-app risks (if the app calls an LLM)
OWASP LLM Top 10 territory — the build checklist's "tested" is not "safe" here:
- Prompt injection: treat ALL model input (user text, retrieved docs, tool output) as untrusted; never let it override system instructions.
- Treat model OUTPUT as untrusted: never exec/eval it, never render unsanitized, never trust it for authz decisions.
- Tool/function calls: least-privilege, allowlist actions, confirm destructive ops; a compromised prompt must not reach dangerous tools.
- Data exfiltration: guard against the model leaking secrets/PII via tool calls or output.
Data privacy & consent (owner: this skill)
- PII inventory: know what personal data you store and why.
- Data-subject rights (DSAR): deletion + export paths exist (GDPR/CCPA).
- Consent: cookie/tracking consent gate before non-essential tracking; honor opt-out.
- Retention: delete data past its retention window (cleanup jobs — see
operating-production-services).
Pre-Ship Security Gate
Do NOT mark a feature/site done until (tick = tested with evidence, not assumed):
- Every user-reachable action has a server-side authorization check; tested across ≥2 users + unauthenticated (A01)
- BOLA / mass-assignment / SSRF checked (A01)
- All input validated server-side; all output encoded (A05)
- CSRF defense on state-changing routes (A05)
- File uploads validated + scanned + stored non-executable (A05)
- CORS not wildcard-with-credentials; full security headers set (A02)
- Secrets out of source; prod config hardened (A02, A04)
- Dependencies scanned, no unpatched critical CVEs (A03)
- Auth: throttling, session expiry, password hashing (argon2id/bcrypt), JWT validated (A07, A04)
- Errors fail closed; no internal detail leaked (A10)
- Security events logged + alertable; no secrets in logs (A09)
- If LLM app: prompt-injection + untrusted-output + tool-scope handled
- Privacy: PII inventory + DSAR + consent + retention addressed
Verify (neutral — map to your stack)
- "Dependency audit" = your ecosystem's CVE scanner.
- "Header/config scan" = a security-header checker + manual prod-config review.
- "Injection/access test" = attempt the attack yourself (or via a helper agent) and confirm it's blocked — evidence, not assumption.
- For a structured review on Claude Code, the
/security-reviewcommand if available.
Companion skills
- Overall delivery gate →
disciplined-delivery(security gate references this skill) - Web build →
shipping-production-websites(pillar 5 = this skill's depth)
Source
OWASP Top 10:2025 — owasp.org/Top10/2025 (current edition; A03 Supply Chain rising, A10 Mishandling Exceptional Conditions new).
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.

