The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes a bash script (evaluate-session.sh) as a 'Stop' hook. The script is securely written, using proper quoting for the transcript_path and LEARNED_SKILLS_PATH variables to prevent command injection from untrusted input.\n- [PROMPT_INJECTION]: The skill processes session transcripts to 'learn' patterns, which creates a surface for indirect prompt injection from data within those transcripts. This is the intended primary purpose of the skill and no specific malicious payloads were detected.\n
Ingestion points: Session transcripts provided by the agent platform via the transcript_path in the hook input.\n
Boundary markers: Absent; the skill relies on the agent platform's default context handling for the transcript.\n
Capability inventory: Shell script execution (mkdir, grep) and the platform's ability to save persistent skills to the local filesystem.\n
Sanitization: Validates minimum session length before signaling the agent to perform pattern extraction.\n- [SAFE]: No external network activity, hardcoded credentials, or unauthorized persistence mechanisms were found. The skill's configuration and logic align with its stated goal of project-specific learning without introducing excessive privileges.

continuous-learning