New skills: LLM/agent-integration security, secrets, privacy/PII, frontend XSS/CSP, API authorization

[richardmhope]

What are you proposing? A new skill (several candidates).

What problem does it solve?

These areas fit a security and code-review collection, but no current skill covers them or only a sub-bullet touches them:

• LLM / agent-integration security: prompt injection through retrieved content, over-privileged tool calls, missing human confirmation on side effects, MCP server configuration, and model output used as code, SQL or shell. Source: OWASP Top 10 for LLM Applications. This is probably the most in-demand gap for a collection aimed at coding-assistant users.
• Secrets: hard-coded credentials, secrets in config/test fixtures/CI, weak secret generation, rotation after exposure, and .gitignore/.dockerignore gaps. Several skills touch this today with inconsistent severities (see the severity-alignment issue).
• Privacy / PII: data minimisation, PII in analytics and telemetry, retention and deletion paths, cross-border transfer hooks.
• Frontend: XSS sinks by framework (dangerouslySetInnerHTML, v-html, innerHTML), CSP, postMessage origin checks, token storage in localStorage.
• API authorization: object- and function-level authorization, mass assignment, rate limiting (OWASP API Security Top 10). security-review's V8 covers part of this.
• Container / Dockerfile: only partly covered by iac-review today.

Proposed approach

Handle each skill as its own PR: structural template, fetched and cited sources, a golden-diff eval fixture, and ideally a clean-diff fixture (see the eval issues).

[richardmhope]

Status check:

• Done:
  • llm-integration-review (Add llm-integration-review skill for LLM and agent integrations #123)
  • frontend-security-review (Add frontend-security-review and privacy-review skills #127, including XSS, CSP, and container-adjacent browser headers)
  • privacy-review (Add frontend-security-review and privacy-review skills #127)
• Remaining:
  • Secrets review: mostly covered already. Align severity, classification and output format across review skills #122 made the canonical "live credential exposed" rating critical, with a single owner per location (security-review, logging, ci-workflow-review, iac-review, and now frontend-security-review for client bundles). A dedicated skill would add generation, rotation and .gitignore/.dockerignore hygiene.
  • API authorization review: overlaps security-review V8 and 15.3 (object and function-level authorization, mass assignment). It is only worth splitting out if it goes deeper on the OWASP API Security Top 10.
  • Container/Dockerfile: still only partly covered by iac-review.

Leaving this open for those decisions.

---

Generated by Claude Code