maintain-llm-wiki

Pass

Audited by Gen Agent Trust Hub on Aug 24, 2026

Risk Level: SAFENO_CODE
Full Analysis
  • [SAFE]: The skill consists exclusively of markdown instructions and YAML configuration. It contains no executable scripts, binaries, or external code dependencies.
  • [NO_CODE]: The skill does not contain any code snippets or external dependencies to analyze.
  • [SAFE]: The system enforces data integrity by requiring SHA-256 hashing for all source snapshots and utilizing an append-only log to record operations.
  • [INDIRECT_PROMPT_INJECTION]: The skill defines a process for ingesting external source snapshots, creating a surface for potential indirect injection. This is addressed through structural defenses: 1. Ingestion points: Untrusted data is ingested as local snapshots via references/ingest.md. 2. Boundary markers: The operating contract requires clear demarcation between read-only evidence records and synthesized wiki content. 3. Capability inventory: The agent is restricted to file system operations within defined wiki and source roots, with no shell or network capabilities specified. 4. Sanitization: The skill relies on human-approved intake and integrity verification via hashes to prevent the processing of malicious or tampered data.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 24, 2026, 10:50 AM
Security Audit — agent-trust-hub — maintain-llm-wiki