The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes a sub-agent (explore) to perform codebase mapping and fact-gathering. This is used to inform the interview process and avoid asking the user for information already present in the source code.
[PROMPT_INJECTION]: The skill is subject to indirect prompt injection because it processes user-provided ideas and question responses, which are then used to generate specifications for autonomous agents like autopilot.
Ingestion points: User-provided arguments in {{ARGUMENTS}} and responses to AskUserQuestion calls in SKILL.md.
Boundary markers: The skill does not define specific delimiters for user input within generated prompts, relying instead on the agent's ability to distinguish intent.
Capability inventory: The skill can read codebase metadata (explore), write to the filesystem (Write), and trigger other high-capability skills (Skill).
Sanitization: The skill incorporates a normalization step (Step 3.6) that summarizes oversized initial context to ensure it fits within prompt budgets and limits the impact of large payloads.
[DATA_EXPOSURE]: The skill reads configuration files from the user's home directory (~/.claude/settings.json, ~/.config/claude-omc/config.jsonc) to resolve platform-specific settings and ambiguity thresholds. This is part of its standard integration with the agent's host environment.

deep-interview