research-synthesis

---
name: research-synthesis
description: Use when researching unfamiliar topics, comparing technology approaches, or building comprehensive understanding from multiple sources
effort: medium
---

**Tier-aware discipline** — before starting, check T3 (`search`), T2 (`memory_search`), and T1 (`scratch` search) widest to narrowest so you don't duplicate work already done; reuse a matching plan via `plan_search` before dispatching multiple agents. Before returning, write findings back at the tier matching their audience (`scratch` for siblings this session, `memory_put` for this project, `store_put` for permanent cross-project knowledge). Full checklist: [resources/tier-discipline.md](../../resources/tier-discipline.md) (shared across every skill that prescribes it — nexus-cnzei.6).

# Research Synthesis Skill

Delegates to the **deep-research-synthesizer** agent.

## Model Selection

Default: **sonnet** (`conexus/agents/deep-research-synthesizer.md` pins
`model: sonnet` in its frontmatter, and an agent's own declaration beats the
session default, so omitting `model` on the Agent tool gets sonnet). Pass
`model: haiku` explicitly for the cheap end of the table below:

| Task Shape | Model | When |
|-----------|-------|------|
| Single-topic lookup, ≤3 sources | haiku (pass it) | Quick research |
| Multi-source synthesis, >5 sources, or cross-project | sonnet (default) | Comprehensive research |

## When This Skill Activates

- Researching a new topic or technology
- Investigating best practices for an unfamiliar domain
- Comparing different approaches or frameworks
- Questions requiring synthesis from multiple sources
- When T3 search returns insufficient context for a decision

## Pre-Dispatch: Seed Link Context

Before dispatching the deep-research-synthesizer agent, seed T1 scratch with link targets so the auto-linker can create catalog links when the agent stores findings:

1. If the task references an RDR (pattern `RDR-\d+`) or a known document, resolve it: `mcp__plugin_conexus_nexus-catalog__search(query="RDR-NNN or document title")`
2. Check T1 scratch for `rdr-planning-context`
3. Write link context to scratch:
   ```
   mcp__plugin_conexus_nexus__scratch(action="put", content='{"targets": [{"tumbler": "<resolved-tumbler>", "link_type": "cites"}], "source_agent": "deep-research-synthesizer"}', tags="link-context")
   ```
4. If no RDR/document reference found, skip seeding (the auto-linker handles empty context gracefully)

## Agent Invocation

Use the Agent tool to invoke **deep-research-synthesizer**:

```markdown
## Relay: deep-research-synthesizer

**Task**: [what needs to be done]
**Bead**: [ID] or 'none'

### Input Artifacts
- Files: [relevant files]

### Deliverable
Research synthesis across multiple sources

### Quality Criteria
- [ ] All relevant sources consulted
- [ ] Key findings synthesized (not just listed)
- [ ] Recommendations provided with supporting evidence
```

For full relay structure and optional fields, see [RELAY_TEMPLATE.md](../../resources/agent-shared/RELAY_TEMPLATE.md).

## Research Methodology

The agent uses `mcp__plugin_conexus_sequential-thinking__sequentialthinking`:
1. Form hypothesis about what information is needed
2. Search nx store for existing knowledge — for scoped evidence, prefer `query(question=..., author=..., content_type=...)` over `search()` + manual scoping
3. Search web resources for current information
4. Analyze relevant code if applicable
5. Synthesize findings from all sources
6. Resolve contradictions between sources
7. Formulate actionable recommendations

## Success Criteria

- [ ] All relevant sources consulted (nx store, web, code)
- [ ] Key findings synthesized (not just listed)
- [ ] Contradictions identified and resolved
- [ ] Recommendations provided with supporting evidence
- [ ] Findings persisted to T3 store via `mcp__plugin_conexus_nexus__store_put` (RDR-080 — direct call, no agent spawn)
- [ ] Round artifacts persisted to T2 or T3

## Agent-Specific PRODUCE

- **Research Synthesis**: Store in T3 via store_put tool: content="# Research: {topic}\n{content}", collection="<subject>", title="research-{topic}-{date}", tags="research,{domain}"
- **Source Citations**: Include in document content (not separate)
- **Knowledge Gaps**: Create research beads for follow-up
- **Cross-Reference Maps**: Document relationships in T3 document content
- **Round Artifacts**: Use T1 scratch to track findings per research round:
  - scratch tool: action="put", content="# Round {N} findings\n{content}", tags="research,round-{N}"
  - scratch_manage tool: action="flag", entry_id="<id>", project="{project}", title="research-round-{N}.md"