docs-expert
Pass
Audited by Gen Agent Trust Hub on Apr 12, 2026
Risk Level: SAFE
Full Analysis
- [COMMAND_EXECUTION]: The skill executes local Python scripts (scripts/bootstrap_doc_qa.py, scripts/check_brand_guidelines.py, scripts/check_readability.py) to perform documentation audits and setup. These scripts include defensive programming, such as validate_repo_write_target in the bootstrap script to prevent symlink traversal and directory escape attacks.
- [PROMPT_INJECTION]: The references/evals.yaml file contains strings that match prompt injection patterns (e.g., 'IGNORE PRIOR INSTRUCTIONS'). These are explicitly identified as test cases within the skill's evaluation suite and do not represent instructions for the agent's primary behavior.
- [DATA_EXFILTRATION]: Indirect prompt injection risk assessment: 1. Ingestion points: Reads repository Markdown, code, and config files (SKILL.md, references/readme-reality-audit.md). 2. Boundary markers: Implements 'Reader Testing' using isolated agent sessions (references/DOC_COAUTHORING.md). 3. Capability inventory: Writes configuration files and brand assets to the local repository (scripts/bootstrap_doc_qa.py). 4. Sanitization: Explicitly instructs redacting secrets and PII from outputs (SKILL.md).
- [EXTERNAL_DOWNLOADS]: The skill's README templates (in references/readme-crafting.md) include examples of installation via curl | bash. These are provided as user-facing documentation patterns rather than agent commands, and the skill explicitly categorizes this as a risky practice in its anti-patterns table, recommending verified package managers instead.
Audit Metadata