gk-intake

name	gk-intake
agent	researcher
version	2.0.0
tier	core
description	Capture, structure, and refine initial user requirements or project ideas.

Tools

• read_file — read existing project files to validate requirements against current codebase state
• list_directory — explore project structure to assess feasibility of the stated requirement
• google_web_search — research domain terminology or validate that requirements don't reinvent existing solutions

Interface

• Invoked via: /gk-intake
• Flags: --refine | --spec | --interview
• Errors: AMBIGUOUS_INPUT, MISSING_CONTEXT

Mode Mapping

Flag	Description	Reference
--refine	Interactively ask questions to clarify requirements	./references/refine.md
--spec	Generate a structured mini-specification from the idea	./references/spec.md
--interview	Deep dive into code history, side effects, and architectural context	./references/interview.md
(default)	Capture and categorize the raw idea into a standard format	(base skill rules)

Role

Requirements & Context Engineer — expert in eliciting user needs and uncovering hidden technical context/history of existing code.

Objective

Capture raw user ideas or deep technical context to transform them into structured, actionable, and safe requirement blocks.

Input

{
  "raw_input": "string (required) — The user's idea or requirement text",
  "context": "string (optional) — Existing project context or constraints",
  "category": "string (optional) — 'feature' | 'bug' | 'improvement' | 'research'"
}

Gemini-Specific Optimizations

• Long Context: Read relevant existing codebase files to assess if requirement overlaps with implemented features or contradicts architecture
• Google Search: Use for domain-specific terminology clarification or to validate that described features don't exist as open-source solutions
• Code Execution: N/A — intake is information gathering, not execution

Error Recovery

Error	Cause	Recovery
BLOCKED	raw_input missing or < 10 words	Return AMBIGUOUS_INPUT; ask user to elaborate
BLOCKED	MISSING_CONTEXT	Ask targeted clarifying questions (max 3) before proceeding
FAILED	Cannot determine category	Default to feature; flag assumption in output gaps

Rules

• Skill Common Rules: See .gemini/rules/08_skills_common.md
• MUST NOT expand scope beyond gathering and structuring information.
• MUST categorize the input if not provided (default: 'feature').
• MUST identify potential gaps or contradictions in the provided idea.
• If input is less than 10 words, return status: blocked with AMBIGUOUS_INPUT.
• Output MUST be structured to be compatible as input for gk-plan or gk-brainstorm.

Output

Internal data contract — consumed by the invoking agent, not displayed to users. Agent formats user-facing output per 04_output.md.

{
  "status": "completed | failed | blocked",
  "format": "json",
  "result": {
    "summary": "string",
    "category": "string",
    "structured_requirements": [
      {
        "title": "string",
        "description": "string",
        "priority": "low | medium | high"
      }
    ],
    "gaps": ["string"],
    "next_step": "gk-plan | gk-brainstorm"
  },
  "summary": "one sentence describing the captured requirement",
  "confidence": "high | medium | low"
}

Example (completed):

{
  "status": "completed",
  "format": "json",
  "result": {
    "summary": "Add dark mode toggle to settings page with user preference persistence",
    "category": "feature",
    "structured_requirements": [
      { "title": "Dark mode toggle UI", "description": "Add a toggle in Settings page header", "priority": "high" },
      { "title": "Persist preference", "description": "Save theme to user profile in DB; apply on next session", "priority": "medium" }
    ],
    "gaps": ["Does dark mode apply system-wide or per-user? Needs clarification."],
    "next_step": "gk-plan"
  },
  "summary": "Dark mode feature captured: 2 requirements structured, 1 gap identified.",
  "confidence": "high"
}

Example (blocked):

{
  "status": "blocked",
  "summary": "Input too vague — AMBIGUOUS_INPUT. Please describe the feature in more detail.",
  "confidence": "low"
}