The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs the agent to offer the installation of the Grok CLI using a remote shell script from x.ai's official domain.
[COMMAND_EXECUTION]: The agent is directed to execute various CLI tools including grok, codex, and npm to manage implementation tasks.
[DATA_EXFILTRATION]: Repository context, specifications, and code prompts are sent to external AI vendors (xAI and OpenAI). The skill includes mandatory safeguards requiring the agent to disclose what is being sent and obtain user consent before the first dispatch.
[PROMPT_INJECTION]: The skill is exposed to indirect prompt injection from malicious or erroneous code generated by subagents. It defines a robust evidence chain for managing this risk: 1) Ingestion points: Subagent reports, diffs, and log files mentioned in SKILL.md and references/native-workflows.md. 2) Boundary markers: Mandatory use of SPEC files and structured 'FINAL REPORT' sections in worker prompts. 3) Capability inventory: Shell command execution, file system writes, and git management. 4) Sanitization: Detailed instructions for 'Adversarial Review' to detect shims or environment-evasion workarounds, and a requirement to re-run all tests outside the sandboxed worker environment.

coordinator