write-agent-skill
Pass
Audited by Gen Agent Trust Hub on Oct 6, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to audit and write agent skills, which involves reading and processing markdown files and scripts provided in the workspace. This creates a surface where the agent could ingest instructions embedded in the audited files.
- Ingestion points: The agent reads existing skill files (SKILL.md, references) and eval files (eval_queries.json, evals.json) provided in the workspace.
- Boundary markers: The skill uses templates and specific structural rules, but does not define explicit delimiters to separate the content being audited from the agent's instructions.
- Capability inventory: Uses python3 to run a validation script scripts/check_reference_structure.py on the workspace files.
- Sanitization: No specific sanitization logic is implemented for the content of the files being audited.
- [SAFE]: The included validation script scripts/check_reference_structure.py is written in Python and uses only the standard library (argparse, json, re, sys, pathlib). It performs local file path resolution and regular expression matching to verify the structure and linking of skill documentation without network access.
- [SAFE]: All external references and documentation links point to well-known technology organizations and research institutions, including Anthropic, OpenAI, Microsoft, Google, Vercel, and Arxiv.
Audit Metadata