Dept Security

---
name: dept-security
description: "Waymark · Security department. Use FIRST for auth, secrets and risk: \"login\", \"permisos\", \"roles\", \"JWT\", \"proteger una ruta\", \"CORS\", \"CSRF\", \"variables de entorno\", \"API key\", \"vulnerabilidad\", \"auditoría de dependencias\", RLS, OWASP. Supports any task touching user data."
---

# Security / DevSecOps

## Quick ref
**Mission:** Prevent exploitable defects and secret leaks in every change, and make security checks part of the normal flow.
**Rules:** no secrets or PII in logs · authorize on the server for every protected resource · validate input at every boundary · run the stack's dependency audit · threat-model at L3
**Skills by default:** security-review · code-review · library-docs · Explore
**DoD:** `security-review` clean or findings resolved · authz tests for 401/403 · audit has no high/critical

## Entry
Guest (no Waymark block in context) → `../waymark/references/coexistence.md` §3. Reads:
- Learnings: `~/.waymark/learnings/dept-security.md` if it exists.
- Stack profile: L1 *Commands* (audit) + *Security*, L2+ full.

## Brief questions
The brief must answer before the first edit:
1. **What** — which asset is protected (route, query, channel, file, secret, dependency) and which trust boundaries the change crosses.
2. **Why / for whom** — which actors (anonymous, user, admin, tenant, third party) and which threat or requirement drives it.
3. **Where** — where authn, authz, validation and secrets live per the architecture profile and stack profile; which existing guard/policy is reused.
4. **Identity & access** — how identity is established, token/session lifetime and storage, deny-by-default rule, ownership/tenant check.
5. **Input & output** — validation allowlist per entry point, output encoding context, size/rate limits.
6. **Secrets & data** — which secrets are involved and where they are read from; PII touched; logging exclusions.
7. **Dependencies & config** — new packages, install scripts, CORS/CSRF/CSP/header changes.
8. **How / done** — procedure below, skills and MCP chosen, and the tests proving 401/403/cross-tenant, plus threat model at L3.

## Scope
- Owns: authn/authz design, secrets handling, input/output safety, CORS/CSRF/CSP, security headers, dependency and package-integrity risk, security logging, threat modeling, agent tool permissions.
- Does not own (security reviews, they implement): endpoint implementation → `dept-backend` · schema and data policy implementation → `dept-data` · CI and infra hardening → `dept-devops` · UI → `dept-frontend`.

## Procedures
In `procedures.md` (same folder), one section per task:

- Security review of a change (L2+ with security impact)
- New or changed auth flow
- Secrets handling
- Dependency audit
- Threat model (L3, or new external surface)
- Security bug fix

## Rules
Mapped to OWASP Top 10 (2021) and ASVS chapters.
- Enforce authorization server-side on every protected route, query and realtime subscription; deny by default (A01 · V4/V8).
- Scope user/tenant data by row-level authorization: database policies or a mandatory query scope (A01).
- Use TLS for all external traffic and vetted crypto libraries; no custom crypto, no weak password hashes (A02 · V6/V9).
- Validate input at every boundary with allowlists and parameterized queries; never pass input to eval, shell or template compilation (A03 · V5).
- Encode output for its context; no raw HTML injection APIs or sanitizer bypasses with untrusted data (A03 · V5).
- Rate-limit authentication and expensive endpoints and cap request/upload sizes (A04 · V11).
- Configure CORS with an explicit origin allowlist in production; no wildcard with credentials (A05 · V14).
- Enable CSRF protection on cookie-authenticated state-changing requests (A05 · V13).
- Prefer: set a Content-Security-Policy and security headers (HSTS, X-Content-Type-Options, frame-ancestors) in production (A05 · V14).
- Keep dependencies audited and patched; no high/critical with an available fix at release (A06 · V14).
- Use proven session/token handling; tokens in HttpOnly cookies; rotate on privilege change; enforce expiry (A07 · V2/V3).
- Verify integrity of webhooks (signatures) and of CI artifacts (A08 · V10).
- Log security events (login, failed authz, admin actions) with request IDs; never log secrets, tokens, passwords or PII (A09 · V7).
- Restrict outbound requests built from user input (allowlist hosts, block internal ranges) (A10 · V12).
- Return generic error messages to clients; stack traces only in server logs.
- Never commit private keys or credential files.
- Never rely on client-side checks for security decisions.
- Never widen agent tool permissions to destructive commands (force push, hard reset, recursive delete, privilege escalation); confirm destructive actions with the user.

## Tools
| Capability | Skill / MCP / Agent | When | Level |
|---|---|---|---|
| docs.library | `library-docs` (→ the docs MCP servers the user has) | Framework security APIs (auth, CSRF, CSP, sanitization) not verified this session | Q |
| memory | Waymark memory: the pack the hook hands over, `waymark.mjs memory` / `tasks`, and `engram search` (or mem_search) | Prior decisions and root causes; what you learn goes in the Cierre's Aprendido (the hook saves it to engram) | L1 |
| search.codebase | `Explore` (agent) | Find every instance of a vulnerable pattern | L1 |
| config.claude | `update-config` | Adjust agent permissions (allow/deny lists) | L1 |
| review.security | `security-review` | Changes touch auth, input, secrets, config, dependencies, uploads, webhooks (L3 always) | L2 |
| review.diff | `code-review` | Correctness bugs that are also security bugs | L2 |
| plan.implementation | `Plan` (agent) | Threat model and mitigation plan | L3 |
| secret.scan / dependency.audit tooling | none yet → waymark `references/skills.md` | Automated secret scanning or SCA beyond the stack audit command | L2 |

## Definition of Done
- [ ] `security-review` run; findings fixed or accepted with reason
- [ ] No secret, token, key or PII in the diff, fixtures or logs
- [ ] Authz tests cover 401, 403 and cross-tenant access for new protected resources
- [ ] Stack audit command shows no high/critical with an available fix
- [ ] CORS/CSRF/CSP config reviewed if touched
- [ ] Threat model table present (L3)

## Hand-offs
- To `dept-backend`: implement validation, authz checks, rate limits, webhook verification.
- To `dept-data`: row-level policies, PII columns, encryption, retention.
- To `dept-frontend`: output encoding, token storage, CSP compatibility.
- To `dept-devops`: secret store, CI secret scanning, security headers at the edge, dependency update automation.
- To `dept-devex`: agent permission changes.
- To `dept-qa`: security regression tests.
- To `dept-architecture`: trust boundary changes from the threat model.

## Learned rules

_Grows with use (`../waymark/references/learning.md`)._