The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes local shell commands during the setup process to initialize its storage environment. Evidence: setup.md and memory-template.md contain bash scripts to create the ~/self-criticism/ directory structure. Evidence: setup.md includes commands to set restrictive file permissions (chmod 700, chmod 600) on the skill's data directory to ensure it is accessible only by the current user.
[PROMPT_INJECTION]: The skill exhibits a designed indirect prompt injection surface as it is intended to learn from user feedback. Ingestion points: User friction, corrections, and revision loops are captured to refine critique triggers (as specified in triggers.md and incidents.md). Boundary markers: No explicit delimiters are implemented to separate ingested feedback from system instructions. Capability inventory: The skill has instructions to write to local memory files and potentially modify workspace configuration files like SOUL.md and AGENTS.md (as outlined in setup.md). Sanitization: No validation or filtering is specified for ingested friction signals before they are persisted in memory.

Self-Criticism