The Agent Skills Directory

[COMMAND_EXECUTION]: The references/patterns.md file provides a code example for a SubprocessOutput component that uses child_process.spawn to execute external commands and capture their output. This is a powerful capability that requires careful input validation to prevent command injection.
[PROMPT_INJECTION]: The SKILL.md file contains behavioral instructions under the 'Companion check' and 'Unlock the full skill graph' sections that direct the agent to perform environment discovery upon activation. It instructs the agent to run ls commands on internal configuration paths (e.g., ~/.claude/skills/, ~/.agent/skills/) and npx skills ls --json to determine which other skills are installed, potentially leading to automated installation prompts.
[EXTERNAL_DOWNLOADS]: The skill documentation promotes the use of npx skills add to install the skill and its companions from a remote source (AbsolutelySkilled/AbsolutelySkilled). It also demonstrates the use of npx create-ink-app for project scaffolding, which downloads and executes remote templates.
[DATA_EXFILTRATION]: The skill instructs the agent to probe local filesystem paths where agent configuration is stored (~/.claude/skills/, etc.). While used to check for missing companion skills, this pattern of system discovery exposes the user's environment to the agent's logic.

react-ink