agent-md-maker
Pass
Audited by Gen Agent Trust Hub on Oct 2, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted repository files to generate instruction artifacts, creating a potential attack surface for indirect prompt injection.\n
- Ingestion points: The skill reads project manifests, source code, and documentation files during the project discovery phase (rules/project-discovery.md).\n
- Boundary markers: Instructions explicitly mandate that retrieved content be treated as evidence rather than authority, and that user instructions outrank repository data (SKILL.md).\n
- Capability inventory: The agent has access to
read,search,edit, andexecutetools.\n - Sanitization: The validation rubric requires rejecting instructions embedded in retrieved content and grounding all generated rules in verifiable project artifacts (rules/validation.md).\n- [COMMAND_EXECUTION]: The skill instructs the agent to execute local commands for environment verification and file management.\n
- Evidence: Git commands (e.g.,
git ls-files,git add) and repository-specific verification scripts (e.g.,npm test) are used to validate state and manage symlinks (rules/instruction-design.md).\n - Safety: Potentially dangerous commands involving network, credentials, or production effects are strictly prohibited without explicit user authorization (SKILL.md).
Audit Metadata