Skip to main content
T4 epic · research / consulting / VC — plan-then-execute. A fast model writes the plan, an agent works the web under explicit budgets, a thinking model synthesizes. Three stages, three different cost profiles, every intermediate on disk.

​
The job

« Get me up to speed on X by Thursday » is two days of tabs, or it’s this pipeline: four sharp queries planned, researched with verbatim quotes and source tracking, synthesized into an executive brief that separates what’s true from what’s contested.

​
The shape

​
The file

t4-deep-research-brief.nika.yaml
nika: v1
workflow: deep-research-brief
description: "plan → budgeted research agent → thinking synthesis → brief on disk"

model: anthropic/claude-sonnet-4-6

vars:
  topic:
    type: string
    required: true
    description: "What to research"

tasks:
  - id: plan
    infer:
      model: anthropic/claude-haiku-4-5    # planning is a fast-model job
      prompt: "Break '${{ vars.topic }}' into 4 sharp research queries · no overlap."
      schema:
        type: object
        required: [queries]
        properties:
          queries:
            type: array
            items: { type: string }

  - id: investigate
    depends_on: [plan]
    agent:
      system: |
        You are a rigorous researcher. Work the queries one by one ·
        fetch sources · keep verbatim quotes · note what each source
        actually supports. Call nika:done when the plan is exhausted.
      prompt: "Research plan · ${{ tasks.plan.output.queries }}"
      tools:
        - "nika:fetch"                 # the web
        - "nika:write"                 # scratch notes
        - "nika:done"                  # the loop-completion sentinel
      max_turns: 25
      max_tokens_total: 150000
      schema:
        type: object
        required: [findings, sources]
        properties:
          findings:
            type: array
            items: { type: string }
          sources:
            type: array
            items: { type: string }

  - id: brief
    depends_on: [investigate]
    infer:
      prompt: |
        Findings · ${{ tasks.investigate.output.findings }}
        Sources · ${{ tasks.investigate.output.sources }}
        Write the executive brief · what's true · what's contested ·
        what it means · what to do. Two pages max.
      thinking:
        enabled: true
        budget_tokens: 8000

  - id: save
    depends_on: [brief]
    invoke:
      tool: "nika:write"
      args:
        path: "./research/${{ vars.topic }}.md"
        content: "${{ tasks.brief.output }}"
        create_dirs: true

outputs:
  brief: ${{ tasks.brief.output }}
  sources:
    value: ${{ tasks.investigate.output.sources }}
    type: array
    description: "Every source the agent actually used"

​
How it works

1

Right model, right stage

Planning is a haiku-class job (model: override on the task); synthesis gets a thinking: budget. The envelope default covers the middle. One file, three cost profiles.
2

The agent has a leash

max_turns: 25 + max_tokens_total: 150000 bound the loop; nika:done lets it exit cleanly when the plan is exhausted. The final message must match the schema — findings AND sources.
3

Callable by other workflows

The typed outputs: (brief, sources) make this a building block — a parent workflow can consume the contract without parsing prose.

​
Constructs you just used

ConstructWhereReference
per-task model: overrideplanProviders
agent budgets + nika:doneinvestigateThe 4 verbs
thinking:briefThe 4 verbs
typed outputs:envelope tailYAML syntax

​
Make it yours

  • Force source-grounding harder: have brief cite ${{ tasks.investigate.output.sources }} inline per claim.
  • Competitive variant: seed plan with Competitor radar’s digest instead of a bare topic.
  • Weekly cadence: pair with your scheduler and diff this week’s brief against last week’s with nika:json_diff.

Next · Incident war room

Parallel evidence gathering, a settle-and-recheck loop, and a postmortem that refuses to claim recovery it can’t prove.