continuous-learning
Pass
Audited by Gen Agent Trust Hub on Aug 5, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill implements a session evaluator using a local bash script (
evaluate-session.sh) intended to be executed via Claude Code's 'Stop' hook mechanism. The script's operations are limited to reading session metadata, counting messages, and signaling the agent to perform analysis. - [PROMPT_INJECTION]: The skill is susceptible to Indirect Prompt Injection (Category 8) due to its core functionality.
- Ingestion points: The skill processes the session transcript (provided via the
transcript_pathinevaluate-session.sh) to extract patterns. - Boundary markers: No explicit boundary markers or 'ignore' instructions are used when reading the transcript data.
- Capability inventory: The skill has the capability to write new agent instructions (skills) to the local filesystem at
~/.claude/skills/learned/based on the processed data. - Sanitization: There is no evidence of content sanitization or filtering to prevent malicious instructions embedded in a session from being 'learned' and persisted as a future skill.
Audit Metadata