ingest
Pass
Audited by Gen Agent Trust Hub on Jul 22, 2026
Risk Level: SAFECOMMAND_EXECUTION
Full Analysis
- [SAFE]: The skill incorporates security-positive logic by instructing the agent to 'Skip material that ... contains credentials or personal data'.
- [COMMAND_EXECUTION]: The skill utilizes local CLI tools
llm-wiki searchandllm-wiki syncto manage and index the wiki vault content. - [INDIRECT_PROMPT_INJECTION]: The skill has a surface for indirect prompt injection as it ingests untrusted user content, notes, and URLs.
- Ingestion points: Raw input is read and stored in the
sources/directory (SKILL.md, Workflow Step 2). - Boundary markers: No specific boundary markers or delimiters are defined for the source data ingestion.
- Capability inventory: The agent can write markdown files to the
wiki/andsources/directories and execute thellm-wiki synccommand. - Sanitization: The skill mitigates risks by requiring the agent to 'Summarize rather than copy source text' and filter for sensitive information.
Audit Metadata