The Agent Skills Directory

[COMMAND_EXECUTION]: The skill documentation provides instructions for users to manually configure a Stop hook in their ~/.claude/settings.json file. This hook executes the evaluate-session.sh bash script whenever an agent session concludes.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to analyze session transcripts which contain untrusted data from user inputs and external tool outputs. This creates a surface for indirect prompt injection if a transcript contains malicious instructions intended to influence the agent during the pattern extraction phase.
Ingestion points: The script evaluate-session.sh identifies and accesses the session chat history via a transcript_path provided by the platform context.
Boundary markers: The skill does not employ explicit boundary markers or "ignore instructions" delimiters when requesting the agent to evaluate the transcript content.
Capability inventory: The associated shell script is limited to local file system checks, directory creation (mkdir -p), and message counting via grep.
Sanitization: No sanitization or filtering is performed on the transcript data before it is presented to the agent for evaluation.

continuous-learning