Orchestrator.md
One investigation, many agents, one case file.
Drop it into an advanced LLM and it automatically spins up multiple specialist agents that work at the same time — one investigates people, another businesses, another reviews your case facts, another writes the report — then merges everything into one defensible case file. You hand it the case; it runs the team.
Runs on any advanced LLM. It coordinates whatever skills you own — the more you have, the more it can orchestrate. Updates ship via the $10/mo Update Subscription.
Not a prompt. A runnable investigation harness.
The methodology
Triage → smart routing → parallel fan-out → second-wave pivots → cross-correlation → synthesis. Follow it manually or hand it to an agent.
A runnable workflow script
Executes the fan-out with real sub-agents, persists a canonical case object to disk, and resumes idempotently — re-running only incomplete clusters.
Verification gates
High-impact findings get refute-tested; a final integrity agent audits the assembled case for hallucinated citations, uncited claims, and miscalibrated confidence.
A deliverables renderer
Turns the case object into a markdown case file, link-analysis CSVs for Gephi / Maltego / i2, and a PPTX briefing deck.
The agent stops working one skill at a time and starts running whole investigations.
Reads the case, plans the team
Triage maps your subject and seed selectors to the applicable workstream clusters — and drops anything outside lawful scope at intake, not after the fact. No prompt gymnastics from you.
Spawns agents that run at once
One specialist agent per cluster, all working simultaneously — each with its own context window and skill subset, so there's no serial bottleneck and no context thrash.
Merges into one case file
Entity resolution dedupes findings across agents, contradictions get flagged, a fact-check gate hunts hallucinated citations — then one report with full provenance.
Skip trace, aliases, social footprint, breach exposure, associates.
Filings, registered agents, beneficial ownership, sanctions, litigation.
Reviews your documents, builds a timeline, flags contradictions and gaps.
Assembles a fully cited case report — sources, confidence language, exhibits.
The same case, with and without Orchestrator.md.
"Here's a case file: a person of interest, the company they're tied to, and a stack of documents the client handed over. I need all of it run down, cross-checked, and written into one report I can defend."
~4 hours, fragmented, you're the project manager.
- T+0You prompt the person workup. It runs. You read it.
- T+60Now the company. Different prompt, fresh context — it's forgotten what the person workup found.
- T+120You hand-feed the documents and ask it to check them against the findings. The context window is straining; you reset the chat.
- T+240You stitch three disconnected threads into one report yourself — and you're the only thing cross-referencing them.
~30 minutes, four agents at once, one report.
- T+0Orchestrator triages the file and launches four specialist agents — all running at the same time.
- ↳ APeople agent: skip trace, aliases, social footprint, breach exposure, known associates.
- ↳ BBusiness agent: corporate filings, registered agents, beneficial ownership, sanctions/PEP, litigation history.
- ↳ CCase-facts agent: reviews the documents you provided, builds a timeline, flags contradictions and gaps to chase.
- ↳ DReport agent: assembles A, B, and C into a fully cited report as they finish — sources, confidence language, exhibits.
- T+30Orchestrator merges everything, dedupes shared findings, and flags the cross-reference the agents caught on their own: the person from A is the registered agent of the company in B. You audit one finished report instead of running four jobs by hand.
Note: parallel sub-agents need an advanced LLM that supports them (Claude with sub-agents/Tasks, GPT-class, etc.). Orchestrator coordinates whatever skills you own — buy it on its own and it orchestrates your library as it grows; pair it with the Complete Library and it ranges across all 659 skills. Add the MCP Toolkit and every agent gets real tool access too.