The Agent Skills Directory

[COMMAND_EXECUTION]: Employs shell commands via the Skills CLI to manage local skill installations and verify their current status.
[PERSISTENCE]: The skill can modify the ~/.claude/settings.json file to register a SessionStart hook, ensuring the update logic executes automatically whenever a new session begins.
[EXTERNAL_DOWNLOADS]: Retrieves repository metadata and tree structures from the GitHub API to perform freshness checks.
[DYNAMIC_EXECUTION]: Utilizes a shell script generated at runtime to facilitate automatic updates when the user has opted into the auto-update mode.
[INDIRECT_PROMPT_INJECTION]: The skill processes external data from GitHub repositories, which is a potential vector for indirect instructions.
Ingestion points: Repository data and release notes fetched from GitHub in scripts/check-pack-freshness.mjs.
Boundary markers: A rewakeMessage is included in the hook configuration to clarify the nature of the output to the agent.
Capability inventory: Permissions include writing to configuration files and executing shell-based update commands.
Sanitization: The skill verifies that skill names match expected patterns before they are included in command executions.

update-agent-skills