The Agent Skills Directory

[PERSISTENCE]: The skill implements a 'Supersede Kit' designed to modify the agent's global configuration to handle skill routing collisions. * Evidence: scripts/setup_supersede_hook.sh modifies the AI agent's settings.json file to register SessionStart hooks. * Evidence: scripts/supersede-routing-hook.sh is installed as a persistent shell hook in the agent's hooks/ directory.
[DATA_EXFILTRATION]: The skill contains logic to read sensitive local files containing conversation histories for the purpose of knowledge distillation. * Evidence: scripts/mine_conversation.py reads project sessions and history from ~/.claude/ and ~/.codex/ directories. * Mitigation: The script includes a Redactor class with patterns to strip API keys, bearer tokens, emails, and user home paths before processing.
[COMMAND_EXECUTION]: Multiple utility scripts execute system commands and external tools as part of the development and packaging workflow. * Evidence: scripts/run_eval.py and scripts/improve_description.py execute claude -p to evaluate and optimize skill triggers. * Evidence: scripts/audit_skill_regression.py and scripts/reference_net.sh execute various git commands to audit skill changes.
[EXTERNAL_DOWNLOADS]: The skill instructions and templates involve downloading verified security and development tools. * Evidence: references/prerequisites.md and scripts/install_<tool>.sh (template) include commands to fetch gitleaks and vercel-labs/skills from their respective official GitHub repositories.

skill-creator