continuous-learning

Pass

Audited by Gen Agent Trust Hub on Jun 17, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill utilizes a shell script (evaluate-session.sh) intended to be configured as a 'Stop' hook in the environment settings. The script performs local file system checks, reads session counts from transcript files, and creates directories for storing learned patterns.
  • [PROMPT_INJECTION]: The skill facilitates an indirect prompt injection surface by design. It ingest session transcripts containing untrusted user messages to automatically generate 'learned skills' (instructions) for future AI sessions.
  • Ingestion points: Reads the session transcript file identified by the transcript_path provided in the hook input.
  • Boundary markers: None identified in the script or configuration to separate user data from instruction extraction logic.
  • Capability inventory: The skill has the ability to write new skill files to the user's local filesystem (~/.claude/skills/learned/).
  • Sanitization: There is no evidence of sanitization or filtering of the transcript content before it is analyzed for pattern extraction.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 17, 2026, 11:53 AM
Security Audit — agent-trust-hub — continuous-learning