From coworkpowers
Researches and plans knowledge work tasks like drafting communications, strategic decisions, meeting prep, or analysis projects before execution.
npx claudepluginhub nabeelhyatt/coworkpowers --plugin coworkpowersThis skill uses the workspace's default tool permissions.
You are orchestrating the **Research phase** of the Compound Knowledge Work loop. Your job is to thoroughly research a task and transform it into an actionable plan that makes execution straightforward.
Orchestrates execution of knowledge work plans using specialized agent mindsets for writing, analysis, decisions, meetings, and tasks. Activates post-planning on cues like 'execute the plan'.
Researches past work and structures knowledge work plans for strategy docs, campaign plans, content briefs, research synthesis, and operational playbooks.
Orchestrates research workflows from question definition to evidence-based findings documentation for technical, requirements, literature, and codebase topics.
Share bugs, ideas, or general feedback.
You are orchestrating the Research phase of the Compound Knowledge Work loop. Your job is to thoroughly research a task and transform it into an actionable plan that makes execution straightforward.
80% of compound knowledge work is in research and review. Do not rush this phase.
Before starting research, check what MCP tools and CLIs are available in the environment (e.g., email, calendar, meeting notes, CRM, company intel). Note which are available and pass this to all agents so they can use them proactively.
Read the task description carefully. If the user provided a document, file, or URL, read it thoroughly.
Classify the work type:
Determine stakes and complexity:
Before launching research, identify gaps and ask the user clarifying questions. This prevents wasted research on the wrong problem.
When to ask questions:
Question patterns (from superpowers):
Hard Gate for high-stakes work - You MUST get answers to:
Example questions:
When to skip this phase:
Before any new research, check what we already know. This is where compounding pays off.
In Claude Code (file-based retrieval):
.context/learnings/INDEX.md - scan for matching category, type, and tags..context/learnings/[category]/ for the relevant work type (e.g., communication/, decision/, meeting/).pattern and template insights that match the current task.type: preference insights in the relevant category - these capture style, tone, and detail preferences from previous work.If past learnings exist, incorporate them into the plan:
pattern) that worked beforefailure) that were documentedtemplate) if a relevant one existspreference) from previous work in this categorySurface what you found: "Based on previous [category] work, we learned [X] and have a template for [Y]."
Launch research agents in parallel based on the work type. Use the Task tool with appropriate subagents. Pass the stakes level to each agent so they can calibrate depth.
Scale agents to stakes:
| Stakes | Agents to Run |
|---|---|
| Low | context-gatherer only (brief scan, no web search) |
| Medium | context-gatherer + 1-2 agents from the work type table below |
| High | Full agent roster from the work type table below |
Agent roster by work type (for medium/high stakes):
| Work Type | Additional Agents |
|---|---|
| Communication | stakeholder-mapper, precedent-researcher |
| Decision | stakeholder-mapper, constraint-analyzer, precedent-researcher |
| Analysis | constraint-analyzer, precedent-researcher |
| Meeting | stakeholder-mapper, precedent-researcher |
| Coaching | stakeholder-mapper |
| Operations | constraint-analyzer, precedent-researcher |
For high-stakes work, always add:
After parallel research returns, assess whether you need additional external research.
Always research externally when:
Use the plan-synthesizer agent mindset to create the final plan.
Plan Output Format:
# Plan: [Task Name]
## Task Summary
[2-3 sentences: What we're doing, why it matters, what success looks like]
## Context Brief
[Key context gathered - what we know, what matters, what's sensitive]
## Prior Learnings Applied
[What we found from past compound phases, organized by type:
- **Patterns**: [approaches that worked before]
- **Failures**: [mistakes to avoid]
- **Templates**: [reusable structures available]
- **Preferences**: [style/tone/detail preferences for this category]
If none found, state "No prior learnings found for this work type."]
## Stakeholder Considerations
[Who's affected, their interests, anticipated reactions]
## Constraints
[Time, political, resource, or other constraints identified]
## Relevant Precedents
[What's been done before in similar situations, what worked/didn't]
## Approach
[Step-by-step plan for execution]
### Step 1: [Action]
- **What**: [Specific action]
- **Why**: [Rationale]
- **Agent**: [Which work-phase agent to use]
- **Inputs needed**: [What information is required]
### Step 2: [Action]
[Same structure]
## Review Plan
[Which review agents should evaluate the output, and why]
| Reviewer | Why Needed |
|----------|------------|
| [Agent] | [What they'll catch] |
## Success Criteria
- [ ] [Measurable outcome 1]
- [ ] [Measurable outcome 2]
- [ ] [Measurable outcome 3]
## Risks & Mitigations
| Risk | Likelihood | Mitigation |
|------|------------|------------|
| [Risk] | [H/M/L] | [How to address] |
## Timeline
[Suggested sequencing and any time constraints]
Present the plan to the user. Ask if they want to:
When the plan is approved, move to execution: /coworkpowers:workflow-work