The Agent Skills Directory

[SAFE]: The skill is an advisory workflow for requirements elicitation and does not perform any high-risk operations such as network requests or arbitrary code execution.\n- [COMMAND_EXECUTION]: The skill utilizes local git commands (e.g., git branch --show-current) and workspace search tools (Grep, Glob, Read) to ground its research in the project context. These operations are restricted to the local environment and intended for information gathering.\n- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted user input ($ARGUMENTS) and codebase content (via Explore/Grep tools). Ingestion points: user request arguments and file system tools. Boundary markers: The skill uses a restatement/diff phase where requirements are tagged as [user-stated] or [inferred] for clarification. Capability inventory: Capability is limited to writing a markdown brief to the local .agent/ directory. Sanitization: Verification is performed via a mandatory human-in-the-loop confirm-and-brief phase.

interview