debug-task
Pass
Audited by Gen Agent Trust Hub on Nov 2, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external sources, including task specifications, acceptance criteria, and failure output (documented in SKILL.md under Ownership and inputs). There are no explicit structural boundary markers or delimiters defined to separate this untrusted data from the agent's instructions. The skill grants the agent capabilities to execute reproduction commands and perform file operations (described in SKILL.md and references/diagnosis-guide.md). While the instructions advise redacting secrets and private data, they do not provide formal sanitization or escaping mechanisms for the ingested content.
- [EXTERNAL_DOWNLOADS]: The skill contains references to an external repository on GitHub to document its adaptation provenance and licensing (in references/upstream.md).
Audit Metadata