The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to ingest session history and user feedback to modify the agent's permanent instruction set.
Ingestion points: Processes user corrections, manual workarounds, and session outcomes directly from the current interaction history.
Boundary markers: The skill requires explicit user confirmation before writing and defaults to showing a diff proposal first.
Capability inventory: The skill possesses the capability to create and modify .md and .py files within the skill library.
Sanitization: The skill relies entirely on human review of the generated content and lacks automated filters to prevent the injection of malicious instructions into the long-term skill library.
[DYNAMIC_EXECUTION]: The skill facilitates the generation and storage of new Python scripts at runtime.
Pattern: The 'Sidecar Offload' and 'Write' procedures allow the agent to extract code blocks into new files (e.g., scripts/<name>.py) for future execution.
Context: These scripts are generated from session data, which could lead to the persistence of malicious code if the distillation process is manipulated by a user.

skill-distillation