Restate the research question in one sentence
List the options to compare (at minimum 2)
Define the decision criteria (cost, complexity, maintenance, performance, security, compatibility)
Identify documentation sources:
- Use Context7 for libraries, frameworks, SDKs, APIs, CLI tools, and cloud services
- Link vendor docs
- Reference existing repo docs or decisions

Phase 2: Research

Spawn read-only Explore subagents in parallel (one per option or topic area):

Fetch current documentation via Context7:

npx ctx7@latest library "<library name>" "<research question>"
npx ctx7@latest docs <libraryId> "<specific question>"

Read official docs, source code, examples
Identify: capabilities, limitations, failure modes, maintenance status, community health
Note expectation gaps between docs and reality

Phase 3: Compare and Recommend

Build a comparison matrix against the decision criteria
State a clear recommendation with rationale
List tradeoffs and risks for each option
Identify follow-up implementation issues if adopting

Phase 4: Write Findings

Write to docs/research/<date>-<slug>.md:

# Research: <Question>

**Date**: <ISO date>
**Decision**: <Recommended option>
**Status**: <recommended | needs-discussion | blocked>

## Question
<Restatement>

## Options Compared

### Option A: <name>
- **Pros**: ...
- **Cons**: ...
- **Risks**: ...

### Option B: <name>
- **Pros**: ...
- **Cons**: ...
- **Risks**: ...

## Comparison Matrix

| Criterion | Option A | Option B |
|-----------|----------|----------|
| ...       | ...      | ...      |

## Recommendation
<Clear recommendation with rationale>

## Follow-up
- [ ] Implementation issue if adopting
- [ ] Migration plan if replacing existing

## Sources
- Context7: <library IDs used>
- Vendor docs: <URLs>
- Repo docs: <paths>

Rules

No production code changes. Research only.
Always use Context7 for library/framework/SDK questions — do not rely on training data.
Capture findings even if the recommendation is "don't adopt."
File follow-up implementation issues for adopted recommendations.
Time-box the research. If a question cannot be answered within the time box, document what is known and what remains unknown.

research-spike

Research Spike

Usage

Process

Phase 1: Frame the Question