Eleven deal agents, inside a broker's authority

The agent catalogue: eleven deal agents across diligence, placement and bind — **The agent catalogue.** Eleven agents across three steps, each with its actions, the level each starts at, and what it never invents.

Analyze runs eleven agents in order on the deal. Each one has one job, reads a defined part of the record, reasons over what it read with Claude, and prepares work for a person. None of them binds, sends or signs.

Three steps, eleven agents

flowchart LR subgraph DIL["Diligence"] A1["Data room"] --> A2["Reps"] --> A3["Diligence gap"] --> A4["Exclusion"] --> A5["Intelligence"] end subgraph PLA["Placement"] B1["Submission"] --> B2["Market"] --> B3["Quote"] --> B4["UW call"] end subgraph BIN["Bind and control"] C1["Binder"] --> C2["Supervisor"] end A5 --> B1 B4 --> C1

Agent	Reads	Prepares for a person	Never invents
Data room	The data room against the placement checklist	A request to the deal team for what is missing	A document the data room doesn’t hold
Reps	The agreement’s reps, qualifiers and schedules	—	A rep or a qualifier
Diligence gap	Reps against the diligence reports	—	—
Exclusion	Schedules and quantified diligence findings	—	An amount
Intelligence	InsightPulse company 360, industry, news	—	—
Submission	The deal and the reviewed findings	Send the submission to markets	—
Market	Each market’s appetite	Approach the shortlisted markets	—
Quote	The comparison engine’s rows	Select the recommended quote	A premium or a rate
UW call	The data room, for underwriters’ questions	—	—
Binder	Binder vs quote; policy vs binder	Accept the binder or policy as checked	—
Supervisor	Placement controls and authority	A practice leader’s sign-off	—

One agent step

flowchart LR G["Gather: facts from the
record and the engines,
each with its source"] --> RZ["Reason: Claude answers
in a fixed shape,
citing facts by id"] RZ --> GC{"Every number
in the facts?"} GC -->|yes| PR["Prepare: finding or
decision at its level"] GC -->|no| RB["Discard; the agent's
own findings stand"] RB --> PR PR --> TR["Trace: model, tokens,
cost, seconds"]

The six diligence agents work the rw_placement playbook: each item’s check finds what the documents and intelligence say, then the agent reasons over it. The placement agents take their figures from the engines — market matching, quote comparison, binder checking — and reason over those. With no model configured, every agent’s own findings are the answer and say so.

The market agent's page with recent runs and one run's trace — **The market agent's page (illustrative).** Recent runs with model calls and cost, and one run's trace — the facts, the model step with tokens and cost, and what was prepared.

Each agent acts as itself

Every step runs under the agent’s own identity at its own autonomy level, and every row it writes is stamped agent:<code>. The audit trail can always say which agent did what, on whose Analyze.

sequenceDiagram participant B as Broker participant P as Pipeline participant AG as Agent participant AU as Autonomy participant L as Lense trace B->>P: Analyze loop each of eleven agents P->>AU: Effective level for this action and sector AU-->>P: L1 or L2, capped at L2 P->>AG: Run as agent identity AG->>L: Step trace AG-->>P: Insight and prepared decisions end P-->>B: Narrative and what waits on you