comet-memory
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill defines rules for processing memory review data packets which may contain untrusted user text, presenting a potential indirect prompt injection surface.
- Ingestion points: Ingests external, untrusted context via the
comet.memory.review.v1packet as detailed in the input boundary section ofSKILL.md. - Boundary markers: Explicitly commands the model to reject prompt injections, secrets, credentials, and text asking to ignore rules or modify instructions.
- Capability inventory: The skill possesses no functional capabilities or tools. The frontmatter includes
disable-model-invocation: true, and the text explicitly states the reviewer must not write files or call tools. - Sanitization: Directs the model to treat prompt injection attempts as text data and execute a
skipaction rather than sanitizing or continuing execution.
Audit Metadata