The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests data from the local filesystem (repository names discovered via the find command) which are then rendered in the UI and used as targets for system operations like opening files in an IDE. Maliciously named local directories could theoretically influence the agent's context or lead to unintended file interactions.
Ingestion points: src/search-repositories.tsx (via useExec call to find in SKILL.md)
Boundary markers: Absent
Capability inventory: Action.Open (Visual Studio Code, Terminal), Action.CopyToClipboard (in SKILL.md)
Sanitization: Absent; repository names and paths are used directly as strings for UI titles and command targets.
[COMMAND_EXECUTION]: The skill provides numerous examples and instructions for executing shell commands to manage system state, configure permissions, and discover resources.
Evidence: Shell command execution via useExec (find), standalone bash scripts (vm_stat), and documentation for CLI utilities (chmod +x, brew install, npx ray).
[EXTERNAL_DOWNLOADS]: The skill instructs the user or agent to fetch dependencies from well-known official registries.
Evidence: Mentions npm install for @raycast/api and @raycast/utils, and brew install node for environment setup.

raycast