The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes a bash script (evaluate-session.sh) triggered by a session stop hook. The script uses standard utilities such as grep, sed, and jq to read configuration files and evaluate transcript metadata.
[INDIRECT_PROMPT_INJECTION]: The skill ingests and analyzes full session transcripts, which represents a surface for indirect prompt injection from user-controlled dialogue. 1. Ingestion points: Session transcripts are accessed via paths passed to the script (evaluate-session.sh) standard input. 2. Boundary markers: No delimiters or specific instructions are implemented in the script logic to prevent the agent from following instructions embedded within the transcript. 3. Capability inventory: File system access (mkdir) and reading transcripts (cat, grep) in evaluate-session.sh, followed by agent evaluation. 4. Sanitization: The implementation verifies file existence but does not sanitize the transcript content before signaling the agent to perform extraction.

continuous-learning