crw-watch
Pass
Audited by Gen Agent Trust Hub on Oct 12, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted markdown and JSON content from web scrapes, which is then processed using a natural-language 'goal' field for AI-assisted diffing.\n
- Ingestion points: Untrusted data enters the agent context via the 'current' and 'previous' fields in the REST API request examples and the bash script in SKILL.md.\n
- Boundary markers: Scraped data is structured within JSON fields, but the prompt construction lacks explicit 'ignore embedded instructions' markers or specific delimiters for the untrusted content.\n
- Capability inventory: The skill utilizes 'Bash' for network operations (curl) and local file system reads (via jq and file redirections).\n
- Sanitization: The provided 'cron-check.sh' script demonstrates a secure practice by using 'jq -Rsc .' to properly escape untrusted strings before interpolating them into JSON payloads, which significantly mitigates the risk of command injection or malformed JSON structure.\n- [COMMAND_EXECUTION]: The skill uses 'Bash' to perform network requests and process data using 'curl' and 'jq'. These are standard development tools used for the skill's intended purpose of web monitoring and do not show signs of malicious usage.
Audit Metadata