The Agent Skills Directory

Local Command Execution: The skill executes framework-specific diagnostic tools such as setup_preflight and skill-token-count using python3 and uv to perform pre-optimization checks and token count measurements.
Dynamic Script Generation: The extract-code pass relocates embedded code blocks into standalone script files; while this involves creating new executable files, the process is designed to be byte-identical and requires maintainer approval.
Indirect Prompt Injection Surface: The skill ingests external Markdown content from other agent skills for analysis and transformation, which introduces a potential attack surface. 1. Ingestion points: Reads target skill files (SKILL.md) provided via user input or automated sweeps. 2. Boundary markers: The skill relies on human review and paragraph-by-paragraph processing rather than programmatic delimiters to manage ingested content. 3. Capability inventory: Employs subprocess execution (python3, uv) and file system write/move operations (git mv). 4. Sanitization: Relies on external validation tools and human approval for all proposed changes before they are committed.

optimize-skill