The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local commands through Python's subprocess module to perform its auditing functions.
scripts/fitness.py uses git archive and git rev-parse to retrieve and analyze skills from local git repositories.
scripts/trigger.py executes the claude CLI tool to perform trigger tests that measure how the agent selects skills in response to specific prompts.- [INDIRECT_PROMPT_INJECTION]: The skill analyzes session transcripts which contain potentially untrusted content from previous user and agent interactions.
Ingestion points: The skill reads Claude Code session history from ~/.claude/projects/*.jsonl in scripts/fitness.py to identify recurring code patterns.
Boundary markers: Data is ingested as structured JSON from log files; the skill treats transcript content as data for analysis.
Capability inventory: The skill can execute local git and claude commands and write to user-defined local baseline files.
Sanitization: Code snippets found in transcripts are analyzed statically using the Python ast module rather than being executed.

skill-fitness