The Agent Skills Directory

[COMMAND_EXECUTION]: The skill performs extensive system administration by executing numerous local commands (e.g., dscl, ioreg, sysadminctl) and using SSH to distribute and run its own audit scripts on remote Macs within the fleet as described in SKILL.md.
[PERSISTENCE]: Scripts such as op-profile-audit.sh and octopool-audit.sh modify user shell configuration files (~/.profile and ~/.zprofile) to ensure that environment variables and binary shims are persistent across interactive and non-interactive sessions.
[DYNAMIC_EXECUTION]: Scripts claude-attribution-audit.sh and global-gitignore-audit.sh utilize Node.js to execute embedded JavaScript snippets from stdin to process and validate JSON configuration data.
[REMOTE_CODE_EXECUTION]: The skill is designed to update global npm packages and Homebrew formulae/casks, which involves fetching and executing code from official public software registries.
[EXTERNAL_DOWNLOADS]: The skill initiates network requests to perform git fetch operations and update software packages from well-known sources including GitHub, Homebrew, and the npm registry.
[INDIRECT_PROMPT_INJECTION]: The skill processes data from inventory files and system state which could potentially contain malicious instructions if those sources were compromised.
Ingestion points: Reads desired state from ~/Projects/manager/fleet/inventory.json and observes system metadata such as application names and bundle IDs.
Boundary markers: Does not utilize formal delimiters when interpolating inventory data into shell commands, though generated Brewfiles are constructed with basic character escaping.
Capability inventory: Extensive capabilities including broad file system writes, network access for updates, and remote command execution over SSH.
Sanitization: Implements JSON.parse for structured data ingestion and utilizes manual escaping functions in fleet-profile.mjs for profile rendering.

fleet-maintenance