make-evergreen-note
Pass
Audited by Gen Agent Trust Hub on Sep 1, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes external, untrusted data in the form of existing user notes and hunches during the grounding and synthesis phases. This creates a surface where embedded instructions in those notes could potentially influence the agent's output.
- Ingestion points: Notes explicitly named by the user, hunches provided during the session, and pending candidates retrieved from the backlog (SKILL.md).
- Boundary markers: The instructions do not specify explicit delimiters for distinguishing user notes from the agent's system instructions, although they mandate strict adherence to retrieved notes during grounding.
- Capability inventory: The skill allows the agent to execute a local binary (.slipbox/bin/slipbox) and perform file operations (SKILL.md), including writing drafts (draft.md) and staging mutations (mutations.json).
- Sanitization: The workflow incorporates /write-checks for note validation and a 'purity check' to ensure synthesized content is distinct from source claims.
Audit Metadata