The Agent Skills Directory

[PERSISTENCE]: Section 7 (Ambient context via session integrations) instructs developers and agents to register tools into the agent's session lifecycle using native hooks. It specifically targets configuration files that execute commands at startup.
Evidence: Mentions registering hooks in ~/.claude/settings.json (Claude Code), ~/.codex/hooks.json (Codex), and ~/.config/opencode/plugins/ (OpenCode).
Evidence: Encourages using SessionStart hooks to inject context via stdout on every session, ensuring persistent presence in the agent's environment.
[DATA_EXFILTRATION]: The skill identifies and encourages modification of sensitive platform-specific configuration files.
Evidence: Directions to read and modify ~/.claude/settings.json, which often contains user preferences and platform-specific data.
[COMMAND_EXECUTION]: The documentation provides explicit shell commands for agents to execute to install additional components.
Evidence: Instructions to run npx skills add <owner>/<repo> --skill <name> to modify the agent's capabilities.
[EXTERNAL_DOWNLOADS]: The skill promotes the installation of external agent skills from unverified repositories.
Evidence: Recommends installing skills via npx skills add, which fetches content from remote sources without integrity verification in the provided instructions.

axi