orchestrate-research

Este extracto es una copia guardada durante la revisión. La fuente externa contiene la versión completa y actual.

---
name: orchestrate-research
description: "Stack orchestrator for research-skills. Reads what's already done in `research/` and `.agents/skill-artifacts/`, parses your intent, and proposes the next 1–3 skills in the research pipeline (icp-research → market-research / diagnose → prioritize → funnel-planner). Use when you don't know which research skill to invoke, or want a guided run through the full audience/market/strategy workflow. Not for executing the work itself — it routes to the skill that does. Not for cross-stack workflows (use orchestrate-meta or invoke skills directly). Renamed from `start-research` in v3.0.0."
argument-hint: "[free-form ask, or empty to be guided]"
allowed-tools: Read Grep Glob Bash
user-invocable: true
license: MIT
metadata:
  author: hungv47
  version: "1.0.0"
  budget: fast
  estimated-cost: "$0.03-0.10"
promptSignals:
  phrases:
    - "where do i start with research"
    - "i want to research"
    - "help me plan research"
    - "what skill should i use for research"
    - "start research"
    - "begin research"
    - "research workflow"
  allOf:
    - [where, start, research]
    - [what, skill, research]
  anyOf:
    - "research workflow"
    - "research pipeline"
    - "guide me through research"
  noneOf:
    - "code review"
    - "marketing campaign"
    - "landing page"
  minScore: 5
routing:
  intent-tags:
    - research-orchestration
    - workflow-routing
    - stack-entry-point
    - research-guide
  position: orchestrator
  lifecycle: pipeline
  produces:
    - .agents/experience/research-workflow.md
  side-effects:
    - manifest-sync
  consumes:
    - research/product-context.md
    - research/icp-research.md
    - research/market-research.md
    - .agents/skill-artifacts/meta/records/[date]-diagnose-[slug].md
    - .agents/skill-artifacts/meta/sketches/prioritize-[slug].md
    - .agents/skill-artifacts/meta/records/targets-[slug].md
    - .agents/experience/*.md
  requires: []
  defers-to:
    - skill: icp-research
      when: "no audience clarity yet — entry point of the research pipeline"
    - skill: market-research
      when: "audience clear, market landscape unclear"
    - skill: diagnose
      when: "known problem or metric decline to root-cause"
    - skill: prioritize
      when: "options landscape needs ranking"
    - skill: funnel-planner
      when: "revenue model needs numeric targets"
  parallel-with: []
  interactive: true
  estimated-complexity: low
---

# Orchestrate Research

*Meta — Stack orchestrator. The entry point for the research-skills stack when you don't know what to invoke.*

**Core Job:** read what's been done, infer where you are in the research pipeline, propose the next skill.

**Core Question:** "Given what's already in `research/` and `.agents/skill-artifacts/`, plus what you just asked, what's the highest-leverage research skill to run next?"

This skill does NOT execute research work. It is a router and progress-tracker. The actual research is done by the skill it routes you to (icp-research, market-research, diagnose, prioritize, funnel-planner).

---

## When To Use

- You just installed the research-skills plugin and don't know what to type.
- You're mid-project and forget which skill is next.
- You have a vague research need ("understand my market", "figure out who buys this", "why are conversions tanking") and want a guided routing.
- You want to resume a workflow across sessions — re-running `/orchestrate-research` re-reads state and resumes from the next step.

## When NOT To Use

- You already know which skill to run — just run it directly. This skill is overhead for users who know the stack.
- You want cross-stack guidance (e.g., "research → marketing → product"). Use `/orchestrate-meta` or compose conversationally.
- You want to execute research yourself rather than route. This skill only proposes and hands off.

---

## How It Works

**Tier note (`metadata.budget: fast`):** This is a pure router — no sub-agent dispatch, no critic ga