The Agent Skills Directory

[DATA_EXPOSURE]: The skill accesses local agent transcripts and session logs (e.g., in ~/.claude/projects/ or ~/.codex/sessions/) to calculate skill usage frequency. This access is limited to reading log files for the primary purpose of the skill, with no network transmission of the gathered data.
[COMMAND_EXECUTION]: The skill instructions direct the agent to perform directory deletions and execute shell commands (e.g., npm run lint, npm run test) only after the user has explicitly approved the recommended cleanup shortlist.
[INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrusted content from agent logs which could contain malicious injection attempts.
Ingestion points: The scripts/k_skill_cleaner.py helper reads .jsonl and .log files from local agent data directories.
Boundary markers: Absent; the script aggregates results into a JSON report that the agent then interprets to make recommendations.
Capability inventory: The agent possesses the ability to delete folders and execute package-level scripts (npm) based on the generated report.
Sanitization: The script performs structural validation by only recognizing directories that contain a SKILL.md file and by excluding sensitive system or project root directories (e.g., .git, node_modules).

k-skill-cleaner