docs-guard
Pass
Audited by Gen Agent Trust Hub on Jul 5, 2026
Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSCREDENTIALS_UNSAFE
Full Analysis
- [PROMPT_INJECTION]: The skill is designed to process external documentation surfaces such as READMEs, tutorials, and API references which serve as untrusted ingestion points for the agent.
- Ingestion points: Documentation files specified in SKILL.md including docstrings, changelogs, and tutorials.
- Boundary markers: The instructions lack explicit delimiters to isolate untrusted documentation content from the agent's core logic.
- Capability inventory: The skill utilizes capabilities to read the local codebase, execute CLI help commands, and run code samples as part of the verification process in verification.md.
- Sanitization: No automated sanitization or filtering of documentation content is described; the skill relies on the agent's manual cross-referencing with the source code.
- [COMMAND_EXECUTION]: The verification procedure in verification.md and code-samples.md encourages the agent to execute shell commands to verify CLI flags (running --help) and to execute code samples found in the documentation to ensure their validity within the project environment.
- [EXTERNAL_DOWNLOADS]: The skill contains informational references to research papers and documentation style guides from established and well-known organizations including Google, IETF (RFC 5737), ACM, and USENIX. These URLs are provided for context and do not involve automated runtime downloads or script execution.
- [CREDENTIALS_UNSAFE]: The skill provides explicit safety guidelines in code-samples.md regarding the handling of secrets, instructing the agent to avoid real-looking tokens and to use RFC-reserved IP addresses and specific placeholders like YOUR_API_KEY to prevent accidental data exposure.
Audit Metadata