woostack-reflect
Pass
Audited by Gen Agent Trust Hub on Oct 5, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted conversation transcripts and tool evidence, creating an inherent attack surface for indirect injections. However, it implements robust mitigations.
- Ingestion points: It analyzes a snapshot of the visible active conversation and tool evidence as defined in SKILL.md.
- Boundary markers: The skill operates on an immutable snapshot and includes explicit instructions to ignore any commands or URLs embedded in the evidence.
- Capability inventory: It has the capacity to propose instruction changes and file upstream issues to external repositories, though only after user approval.
- Sanitization: The instructions strictly mandate the removal of secrets, credentials, and raw transcript data from all outputs and issue bodies.
Audit Metadata