self-learn

Pass

Audited by Gen Agent Trust Hub on Nov 8, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrusted data from git commit logs, session notes, and trajectory stores to generate 'learnings' and update project governance.
  • Ingestion points: Reads from .aidoc/learning/learnings.md, harness-specific session checkpoints, git log, and trajectory stores.
  • Boundary markers: The instructions include a verification step ('audit citations you author yourself', 'verify every candidate against source') to minimize errors, but the agent may still follow instructions embedded in these external sources.
  • Capability inventory: The skill can write to consolidated learning files, modify governance documents (like CLAUDE.md or AGENTS.md), and submit feedback/issues.
  • Sanitization: Includes a hygiene step to validate that no secrets or PII are stored in the entries.
  • [COMMAND_EXECUTION]: The skill utilizes repository inspection commands such as git log and grep to extract evidence and verify claims during the consolidation process.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 8, 2026, 11:07 PM