lode-experience-distillation

Pass

Audited by Gen Agent Trust Hub on May 8, 2026

Risk Level: SAFE
Full Analysis
  • [SAFE]: The skill's operations are restricted to the local environment, specifically analyzing JSON logs within a user-defined 'knowledge vault'. It does not perform network operations or access sensitive system credentials such as SSH keys.
  • [COMMAND_EXECUTION]: The skill invokes a local script (scripts/distill_candidates.py) to extract candidates from work entries. The script uses standard file I/O and keyword-based logic to categorize data, avoiding dangerous dynamic execution functions.
  • [PROMPT_INJECTION]: As the skill summarizes external work entries, it contains a surface for indirect prompt injection where malicious log data could attempt to influence rule generation.
  • Ingestion points: The skill reads JSON files from the {vault}/raw/weeks/ directory.
  • Boundary markers: The skill and its associated contract (references/distillation-contract.md) enforce a 'proposal-first' workflow, ensuring the agent only suggests changes that the user must review.
  • Capability inventory: The skill is capable of reading local vault files and proposing updates to repository documentation and agent behavioral rules.
  • Sanitization: The Python script extracts raw text from JSON fields without specific filtering for prompt injection markers, relying on the user-approval checkpoint for safety.
Audit Metadata
Risk Level
SAFE
Analyzed
May 8, 2026, 06:48 AM
Security Audit — agent-trust-hub — lode-experience-distillation