scope-guard-craftsman
Pass
Audited by Gen Agent Trust Hub on Sep 26, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONNO_CODE
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill establishes protocols for the agent to ingest and analyze existing codebase content. This creates an attack surface where malicious instructions could be embedded in repository files, such as code comments or documentation, to influence the agent's behavior during inspection. The skill partially mitigates this by requiring explicit evidence states and smallest-justified-change plans. \n
- Ingestion points: Repository source code, configuration files, and documentation (as referenced in SKILL.md and references/evidence-model.md). \n
- Boundary markers: The skill mandates the use of specific evidence labels ([PROVEN], [OBSERVED], etc.) and a 'First-Sentence Rule' for plans, which provide structural delimiters for the agent's reasoning. \n
- Capability inventory: The skill assumes the agent has access to file inspection, symbol searching, and test execution tools. \n
- Sanitization: No specific input sanitization or filtering logic is defined for the content being inspected. \n- [NO_CODE]: The skill consists purely of markdown-based instructions, examples, and documentation. It does not include scripts, binaries, or other executable code that could perform automated actions.
Audit Metadata