deep research · claude code primitives

founder/research/agent

An end-to-end research pipeline built from nothing but subagents, one skill, and one MCP. Press play to watch a single run move down the tiers — no orchestration code, no SDK.

elapsed0:00
cost$0.00
modelready
step 1 / 8  ·  ready
Press play to watch the founder's data flow down the pipeline, tier by tier.
INPUT research [a founder] lead agent · CLAUDE.md — routes every tier · Task · Bash · Read 01 IDENTITY CHECK CONDITIONAL · MCP: parallel_search identity-checker HAIKU resolves ambiguous names — e.g. Alexander Wang (Scale AI) vs. the designer — waits for your pick 02 SOURCE RESEARCH PARALLEL ×4 · MCP: parallel_search biography books & biographies HAIKU wikipedia facts & chronology HAIKU news media & events HAIKU social interviews & posts HAIKU files/{founder}/source_research/ 03 CATEGORY ANALYSIS PARALLEL ×4 · reads source_research/ career roles · founding · exits SONNET financial valuations · net worth SONNET personal-dev skills · mentors · growth SONNET verification cross-check · confidence SONNET files/{founder}/analysis/ 04 ENHANCE & INTEGRATE SEQUENTIAL ×2 context-enhancer SONNET adds historical & causal context → enhancement/context.md integration-agent SONNET builds one timeline with Life Phases → integration/timeline.md 05 VISUALIZATION PLANNING SEQUENTIAL visualization-planner SONNET reads Life Phases + the 400-word image ceiling, then splits the timeline into 2–6 cohesive frames 06 IMAGE GENERATION PARALLEL ×N · SKILL: timeline-visualization image-gen → frame 1 HAIKU image-gen → frame 2 HAIKU image-gen → frame N HAIKU swap OPTIONAL SWAP skill → Higgsfield MCP no AI-SDK · no key in cwd files/{founder}/output/*.png 07 REPORT SEQUENTIAL · FINAL report-writer OPUS the only Opus call — embeds every image, writes the phase narrative, synthesizes the final document OUTPUT report.md + timeline images
14
subagents
~31m
per founder
$4.89
per run