writing-plans
Pass
Audited by Gen Agent Trust Hub on Mar 29, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: No malicious patterns, obfuscation, or unauthorized data access were detected in the skill's instructions or templates. The skill follows established software engineering best practices such as DRY, YAGNI, and TDD.
- [INDIRECT_PROMPT_INJECTION]: The skill processes external specifications or requirements provided as input. While this represents a surface for indirect prompt injection where a malicious specification could attempt to influence the generated plan, the skill incorporates a multi-agent review loop (using the
plan-document-reviewer-prompt.mdtemplate) which acts as a verification and security control. - Ingestion points: Reads a specification or requirement document in
SKILL.mdto generate implementation plans. - Boundary markers: The skill does not explicitly define markers to isolate input text from instructions.
- Capability inventory: Instructions involve generating plans that recommend file writes and shell commands (e.g.,
pytest,git). Actual execution is deferred to other recommended sub-skills. - Sanitization: Relies on LLM reasoning and the review subagent logic rather than programmatic sanitization or escaping.
Audit Metadata