Reference — generated from the toolkit
Source of truth: toolkit/prompts/security/self-review-asvs.md. Edit it there; this page is regenerated on build.
Security self-review (ASVS)
Purpose: the agent reviews its own diff against ASVS L2 before raising the PR — security encoded once, invoked everywhere.
Prompt
Review THIS diff against OWASP ASVS Level 2. For each relevant control area (auth, access control, input validation, output encoding, data protection, logging), state pass/concern with the specific line. Flag: unparameterised queries, missing authz checks, secrets in code/logs, unvalidated LLM/tool output, PII in logs. Propose the minimal fix for each concern. Do not claim a control passes without pointing to the code that satisfies it.
Usage notes
- Runs before the PR is raised; the human still reviews at the merge gate.
Failure modes
- False "passes" (mitigated: "point to the code" requirement).