steelman
Pass
Audited by Gen Agent Trust Hub on Sep 4, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill's core workflow involving the analysis of repository files and external web documentation based on user input creates an attack surface for indirect prompt injection. Maliciously crafted content in these ingested sources could influence the agent's reasoning or lead it to execute unintended commands during the verification phase.
- Ingestion points: In SKILL.md, the agent is instructed to read referenced files, use grep to find patterns, and consult external web documentation.
- Boundary markers: There are no instructions in the skill to utilize delimiters or safety headers that would prevent the agent from being influenced by instructions embedded within the gathered data.
- Capability inventory: The skill leverages capabilities including file reading, symbol searching (grep), and the execution of shell commands and tests (e.g., "run tests/commands if cheap") as defined in SKILL.md.
- Sanitization: The instructions do not define any validation, filtering, or escaping processes for the content retrieved from the filesystem or the web before it is used.
Audit Metadata