The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is designed to execute local shell scripts, specifically scripts/probe_asset_credentials.sh and scripts/audit_reference_report.py. The credential probe script sources the user's shell configuration files (~/.bashrc, ~/.zshrc, ~/.bash_profile, ~/.zprofile) to evaluate the environment.
[DATA_EXFILTRATION]: The skill implements a 'Credential Probe' mechanism that checks for the existence of specific environment variables (TRIPO_API_KEY, GEMINI_API_KEY, ELEVENLABS_API_KEY). While it does not print the secret values, it instructs the agent to report the presence of these keys ('SET' or 'MISSING') in a markdown report, which constitutes reconnaissance and information disclosure regarding the user's available credentials.
[PROMPT_INJECTION]: The skill contains an 'Indirect Prompt Injection' surface through its 'Skill Path Ladder' mechanism. It instructs the agent to search for and load SKILL.md files from several directories, including those in the user's home directory (~/.claude/skills/, ~/.codex/skills/, ~/.agents/skills/). This allows the agent's behavior to be controlled by files placed in these locations by other processes or users.
Ingestion points: Relative and absolute file paths including ~/.claude/skills/<skill-name>/SKILL.md, ~/.codex/skills/<skill-name>/SKILL.md, and ~/.agents/skills/<skill-name>/SKILL.md.
Boundary markers: None. The skill instructs the agent to 'follow its workflow' for the loaded sibling skill.
Capability inventory: The agent can read files, execute shell commands (Node.js, Python 3, Bash), and potentially drive a browser via Playwright.
Sanitization: No validation or sanitization of the content from the loaded SKILL.md files is performed before the agent adopts their instructions.

threejs-game-director