The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes 'git diff' or 'git diff --cached' to compare file changes made during a skill's execution against expected outcomes defined in the SKILL.md. This is a local operation used for diagnostic purposes.
[INDIRECT_PROMPT_INJECTION]: The skill analyzes conversation history and user feedback to identify failures or ambiguities. This creates a surface where malicious instructions embedded in the processed data could potentially influence the optimization proposals. Ingestion points: Conversation history, file diffs, and manual user feedback. Boundary markers: The skill uses a structured diagnostic framework (e.g., 'Specificity gap', 'Ambiguity') to categorize findings and separate them from general conversation. Capability inventory: The skill has the ability to read and write to local markdown files (SKILL.md, LESSONS.md) and execute git commands. Sanitization: The skill implements a redaction mechanism that automatically masks patterns matching API keys and tokens in its output and the persistent LESSONS.md file.

skill-optimizer