agkan-subtask-direct
Pass
Audited by Gen Agent Trust Hub on Sep 22, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data, specifically task titles and bodies retrieved via
agkan task get, and interpolates this content into a sub-agent prompt for implementation review. - Ingestion points:
SKILL.md(Step 6 and Step 7) retrieves task metadata and bodies from theagkantool. - Boundary markers: The sub-agent prompt in Step 7 does not use specific delimiters or instructions to ignore potential instructions embedded within the task title or body.
- Capability inventory: The skill possesses capabilities including
git commit,git push, andagkan task update(to mark tasks as done). - Sanitization: While
jqis used for data extraction, the interpolated content is not sanitized or escaped before being placed into the prompt context. - [SAFE]: The skill incorporates proactive security guidance by explicitly warning against using
git add -Aorgit add ., which helps prevent the accidental commit of sensitive files like.envor credentials.
Audit Metadata