llm-wiki
Pass
Audited by Gen Agent Trust Hub on Aug 16, 2026
Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests external content from files and URLs, presenting a surface for indirect prompt injection. It mitigates this via a mandatory evidence chain:\n
- Ingestion points: Document parsing in
scripts/ingest.pyaccepts user-provided files and URLs.\n - Boundary markers: Sources from URLs or suspicious content are tagged with
trusted: false.AGENTS.mdinstructs the AI to treat wiki content as data, not instructions.\n - Capability inventory: The skill performs file operations, git commits, and subprocess execution for search indexing.\n
- Sanitization:
scripts/ingest.pyscans for injection patterns using regex and warns the agent.\n- [COMMAND_EXECUTION]: The skill executes local Python scripts and shell commands (git,pip,npm) for wiki operations and dependency management.scripts/init-wiki.pyspecifically runs a local setup script for search functionality.\n- [EXTERNAL_DOWNLOADS]: The skill fetches external documents and installs packages from standard registries. These operations are core to its functionality.
Audit Metadata