Developers · Agent harness
The harness for succession and M&A work.
General-purpose assistants know nothing about what a practice is worth, who may see a target's numbers, or when a move is proven. HnyComb runs that work as governed agent loops, with checkpoints, permission tiers and owner approvals, in the workbench or inside the assistant you already use.
Every tool call, sorted by tier
margin loop · engine 1.8.0
- The model asks for a tool, and every call passes the tier check.
- Read: runs now. The engine runs it, and figures come only from the engine.
- Propose: parks a card for the owner. Nothing runs, and pulls start only when the owner starts them.
- Approve: anything that changes the record waits at the gate until the owner clicks.
- Never: refused and logged. No handler exists.
- Each step is written once to the run record and never edited.
Run record · append only
written once, never edited
- 01 · readread_margin_facts
- 02 · readdraft_steps
- 03 · proposepropose_pnl_pull
- 04 · approveapply_steps
- 05 · neverrefused: send_email
- 06 · neverrefused: change_figure
01 / Agent harness
A loop, step by step.
The data check that runs before every compute. Clean data computes straight through. Here two sources disagree, so the loop stops for the owner.
- 01Read the practice, its sources and the latest saved run
- 02Detect issues deterministically: missing, unverified, inconsistent, currency, ambiguous, default value
- 03Word one plain question per issue; the model writes words only and fixed wording takes over if it fails
- 04Wait for the owner: pick a source, enter a value, start a pull, or compute anyway as provisional
- 05Compute on the owner's click; open items are saved with the run and listed on the report
Step log · illustrative
| seq | kind | tool | tier |
|---|---|---|---|
| 1 | tool | read_preflight | read |
| 2 | model | order_issues | |
| 3 | system | waiting_owner | |
| 4 | owner | answer:pick_source | |
| 5 | tool | write_value | execute_after_approval |
| 6 | tool | read_preflight | read |
| 7 | owner | compute | |
| 8 | tool | compute | execute_after_approval |
02 / Agent harness
Four permission tiers
Every tool a loop can call is registered with one tier. A call outside its tier is refused and the refusal is logged as a step.
| Tier | What happens | Examples |
|---|---|---|
| read | Runs freely | Practice record, source map, saved runs, data requirements |
| propose | Becomes a card for the owner; nothing runs | Pull from a connected ledger, use one source's figure, ask the target |
| execute_after_approval | Runs only after the owner's click | Start the pull, save a chosen value with its source, compute, add steps to the plan |
| never | Not callable; any attempt is refused and logged | Send or share without approval, change a figure on its own, mark a move verified, delete, contact a client |
03 / Agent harness
Six parts of a harness, built for this domain.
A harness is the system around a model that runs the loop, calls the tools, keeps state and enforces the rules. Here is what HnyComb does for each part.
- Agent loop
- HnyComb runs seven loops: the data check before every compute, the operating-margin loop, the client book loop, the billing loop, the growth loop, the hosted agent that works an assigned move on its cadence, and the deal loop for buyers and targets over A2A. The deal loop also answers a network's recruiting requests and practice reviews for the firm; on the network side, those requests are sent by hand from the Recruiting, Practice management and Succession desks. Each run has step, model-call and time caps, and stops when it is done, needs the owner or reaches a limit.
- Tool calling
- Tools run on the server against a deterministic seven-method engine. The model orders issues and words questions; it never states or computes a figure.
- Context management
- Skills are served fresh on every call. The model sees issue types and category order, never client rows or ledger lines.
- Checkpointing
- Every step is written once to a step log that cannot be edited, with an idempotency key. An interrupted run resumes from its last step and never repeats a write.
- Permissions and sandboxing
- Four permission tiers on every tool. Read-only pulls, only when you start them. A buyer's loop never pulls from a target's systems; it asks through a data request under the NDA.
- Guardrails
- Model text carries no digits. A compute stops on blocking gaps until they are answered or acknowledged. Only a saved run marks a move verified. Drafts are never sent.
04 / Agent harness
Loops today
Every move on a plan also has its own lifecycle: planned, in progress, done and verified against the next saved run.
validate
Before every compute, in the workbench and over MCP
Finds missing, conflicting, unverified and default inputs and asks only about those.
margin
On the operating-margin move, and re-checked on the move's cadence
Reads the P&L the owner pulls, shows the expense mix and the gap to the size-tier median, drafts steps the owner applies.
client book
On the retention, concentration and demographics moves, and re-checked on each move's cadence
Reads the household book the owner pulls or uploads in the browser, shows retention, concentration or demographics for that move, and drafts steps the owner applies. Household rows never leave the browser.
billing
On the revenue-per-client and recurring-revenue moves, and re-checked on each move's cadence
Reads the household billing the owner pulls or uploads, shows the effective rate, fee mix or recurring share for that move, and drafts steps the owner applies. Billing rows are never stored or logged; only the totals the owner sends are kept.
growth
On the organic growth move, and re-checked on the move's cadence
Separates net new assets from market change using flows the owner pulls or uploads, and drafts steps the owner applies. A stated growth rate is never treated as flows, and the loop never forecasts. Flow rows are never stored or logged; only the totals the owner sends are kept.
hosted agent
On an assigned move's cadence, or Run now
Scout, the hosted agent, works the next open step, drafts the deliverable the move's playbook names, and leaves every change as a card for the owner.
recruiting answers (deal loop)
While a recruiting request is open, on the advisor firm's side
The deal loop's answering side works a network's recruiting request under the recruiting NDA, drafting from the firm's own record; the firm decides what to share. The network sends the request by hand from its Recruiting desk, and accepting a reply never recomputes anything on its own.
practice review answers (deal loop)
While a practice review is open, on the member firm's side
The deal loop's answering side works a network's practice review under existing network consent, drafting from the member's own record. The network sends requests, suggested moves and professional offers by hand from Practice management or the Succession desk, where a review can carry a succession agenda; the member chooses Add to plan, Assign or Not now.
deal
While a diligence room is open, for the buyer and the target
The buyer's loop finds what is missing and proposes data requests and diligence; the target's loop drafts responses from its own record. Every send, share and accept waits for a person, over A2A.
05 / Agent harness
Loops between firms and networks.
The same harness runs work between two parties. Each side proposes, a person clicks, and the two sides connect only through a request. Neither side reads the other's run.
Deal diligence
A buyer proposes data requests and diligence. The target drafts answers from its own record. Every send, share and accept waits for a person, behind an NDA.
Recruiting
A network's recruiting desk asks an advisor firm for an educational estimate of its book. The recruit's loop drafts an answer; the recruit decides what to share, under the recruiting NDA.
Practice review
A network's practice desk requests figures, suggests moves and offers professionals to member firms. The member's loop drafts answers under existing network consent; the member decides what joins its own plan.
Agents never accept a suggested move or a professional's offer for the owner. See agent to agent for how these run between assistants.
06 / Agent harness
Over MCP.
The same rules hold when the work runs in Claude, ChatGPT or any MCP client.
- compute_valuation
- Runs the data check first. With blocking issues it returns the typed issues and questions instead of a valuation, unless the call passes
gaps_acknowledged: true, which saves the run as provisional. - validate_practice
- An explicit check.
format: "issues"returns each issue's type, a question and the sources that can fill it, so the assistant can work through them with the owner. - Pulls stay with the owner
- In your AI assistant, your own connectors do the reading, and HnyComb never receives their credentials. Only the totals you approve are saved, each with its source.
07 / Agent harness
Scout runs on the same harness.
Scout is the in-app assistant that works the next open step of an assigned move and leaves every change as a card for the owner to review and approve.
Meet Scout →Run it in your assistant
Connect HnyComb to Claude or ChatGPT and ask for a valuation. The plain-English versions of these rules are on How it works and Trust & data use.

