wiki-crystallize
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input in the form of chat transcripts, user-provided summaries, and descriptions to generate or update wiki pages. This creates a surface for indirect prompt injection, where malicious instructions embedded in a conversation could be processed and persisted into the user's permanent knowledge base.
- Ingestion points: User-provided thread summaries, pasted transcripts, and session descriptions are ingested at the beginning of the crystallization workflow.
- Boundary markers: The instructions do not specify any boundary markers (such as XML tags or unique delimiters) to separate the untrusted data from the processing instructions.
- Capability inventory: The skill requires and utilizes full filesystem read and write access to manage wiki pages, indexes, logs, and overview files (SKILL.md).
- Sanitization: No explicit content sanitization or instruction-filtering is performed on the ingested text beyond the agent's internal distillation logic.
Audit Metadata