Security Engineer
Application security, end to end: from "what can go wrong with this design" to "we were breached, what now". Every leaf holds its full procedure — nothing here hands work to another skill.
When to use
Any task whose subject is an attacker: designing a system, reviewing a diff, choosing a verification target, hardening an endpoint, triaging a CVE, storing a secret, or cleaning up after an incident. Test phrase: "is this endpoint secure?"
When not to use
Writing prose docs, building or fixing release pipelines, operating infrastructure day to day, or attacking systems you do not own — no leaf here covers them.
The professional workflow
Threat-model the design → pick the verification level → build the defences (validate input, encode output, parameterise, authenticate, authorise per object, bound consumption) → store and rotate secrets → prove it with tests and scans → log the right events → know the incident first hour before you need it.
Precedence rule
The deepest loaded instruction governs its uniquely-claimed scope; every
scope:string is claimed by exactly one file in this tree; no leaf may widen a prohibition declared by an ancestor — leaves narrow, never override upward.
Subskill index
| Leaf | Task | Trigger |
|---|---|---|
subskills/00-rsi.md |
RSI event log | After any leaf in this skill fires |
subskills/threat-model.md |
Threat-model a system or feature | "What can go wrong here?" |
subskills/secure-design.md |
Design with trust boundaries and safe defaults | "How should this fail closed?" |
subskills/verification-level.md |
Choose and apply an ASVS verification target | "What level must this app meet?" |
subskills/input-output.md |
Validate input and encode output | "How do I handle untrusted data?" |
subskills/injection.md |
Stop SQL, command and template injection | "Make this query injection-proof" |
subskills/browser-security.md |
Defend the browser side: XSS, CSRF, headers, cookies | "Stop XSS or CSRF here" |
subskills/authentication.md |
Passwords, MFA, passkeys, recovery | "How do users log in and recover?" |
subskills/sessions-tokens.md |
Sessions, JWTs, OAuth tokens, webhooks | "How long should a session live?" |
subskills/authorization.md |
Enforce who may do what to each object | "Stop user A reading user B's data" |
subskills/api-security.md |
Harden an API against the API Top 10 | "Is this endpoint bounded and safe?" |
subskills/secrets.md |
Store, scan, rotate, respond to leaks | "Where does this key live? It leaked — now what?" |
subskills/crypto-use.md |
Use vetted cryptography correctly | "How do I encrypt this? Which hash?" |
subskills/ssrf.md |
Block server-side request forgery | "The server fetches a user URL — make it safe" |
subskills/file-upload.md |
Accept and serve uploads safely | "Let users upload files safely" |
subskills/deserialization.md |
Parse untrusted data safely | "This library deserialises — is that safe?" |
subskills/supply-chain.md |
Pin, sign and vet dependencies and builds | "Trust what I install and ship" |
subskills/cve-triage.md |
Prioritise vulnerable dependencies | "A dependency has a CVE — how urgent?" |
subskills/security-testing.md |
Gate code with security scans in CI | "Catch security bugs before merge" |
subskills/llm-prompt-injection.md |
Defend LLM features against injection | "Can a prompt hijack the model?" |
subskills/agent-tools.md |
Constrain what an agent's tools may do | "What may this agent call?" |
subskills/privacy-data.md |
Handle personal data lawfully | "My duties for this user data?" |
subskills/abuse-rate-limits.md |
Rate-limit and defeat automated abuse | "Bots hammer signup — throttle" |
subskills/security-logging.md |
Log security events, never secrets | "What events must I log and alert on?" |
subskills/security-incident.md |
First hour of a suspected breach | "We were breached — what first?" |
subskills/vuln-disclosure.md |
Receive and handle external reports | "A researcher reported a bug — process?" |
subskills/cloud-iac.md |
Harden cloud accounts and IaC | "Is this cloud account and IaC safe?" |
subskills/containers.md |
Build and run containers safely | "Harden this image and runtime" |
subskills/security-review.md |
Review one diff against the target level | "Is this change safe to merge?" |
Hard lines (human-gated acts)
Propose, never execute: deploying to production, deleting data, dropping schema, force-pushing, publishing, sending or submitting, releasing a version, rotating or revoking credentials, disclosing a vulnerability externally, or claiming authorship of AI-generated text.
Sources
- OWASP Top 10 — 2025 edition current (announced 2025-11-06, final Jan 2026); 2021 superseded. https://owasp.org/Top10/2025/
- OWASP API Security Top 10 — 2023 current; no newer official edition. https://owasp.org/API-Security/editions/2023/en/0x04-release-notes
- OWASP ASVS 5.0.0 (2025-05-30): 14→17 chapters, full renumbering, v4 IDs stale; L2 the target for personal or payment data. https://asvs.dev
- OWASP Top 10 for LLM Applications 2025 and Top 10 for Agentic Applications (final 2025-12-09). ⚠️ the Agentic list was read from secondaries.
- NIST SP 800-53 Rev 5, SP 800-207, SP 800-63B-4 (final July 2025), SP 800-218 SSDF (1.1 final; 1.2 draft), SP 800-57, FIPS 203/204/205, IR 8547.
- CWE Top 25 2025 and the CISA KEV list; NVD/CVSS 4.0 for scoring.
See Also: each leaf names its siblings where a boundary sits.