The Agent Skills Directory

[SAFE]: The skill contains no executable code, scripts, or automated commands. It is composed of markdown-based instructions and templates for human or agent workflows.
[SAFE]: The content emphasizes strong security and authorization boundaries, specifically listing operations that must never be delegated, such as managing secrets, database migrations, production deployments, and final verification of code.
[INDIRECT_PROMPT_INJECTION]: The skill outlines a system for delegating tasks via structured "briefs" to sub-agents. This creates a potential surface for indirect prompt injection if the data included in these briefs originates from untrusted external sources.
Ingestion points: The brief-writing process described in SKILL.md and templated in references/brief-template.md.
Boundary markers: Not explicitly enforced; the skill relies on manual construction of the brief files.
Capability inventory: The delegates mentioned (Claude sub-agents, Grok CLI, Codex CLI) typically possess file system and command-line execution capabilities.
Sanitization: No automated sanitization or escaping mechanisms are described; the skill focuses on high-level procedural guidance.

delegation