The Agent Skills Directory

[COMMAND_EXECUTION]: The skill performs command execution for validation and deployment of new skills.
It executes a local validation script located at skills/learn/scripts/validate_skill.py using the uv runner.
It uses shell commands to copy files to the persistent storage at ~/.claude/skills/, which is necessary for the skill's primary function of cross-session knowledge persistence.
[PROMPT_INJECTION]: The skill creates a surface for indirect prompt injection by transforming session data into persistent agent instructions.
Ingestion points: Content from the current session, including user workflows and workarounds, is ingested to create new SKILL.md files.
Boundary markers: The skill provides a structural template for new skills (Phase 4) but does not explicitly instruct the agent to sanitize or ignore potentially malicious instructions embedded in the session data.
Capability inventory: The skill uses Write to create new skill definitions and allows Bash(uv run python*) for generated scripts.
Sanitization: There is no explicit sanitization or validation of the session content before it is persisted into the new skill's instructions.

learn