Ma Deal Workbench Review

M&A Deal Workbench Review

Reusable operating rules distilled from the train tasks. These rules apply to any workbench task that names a deal, points at <TASK_ENV_BASE_URL>, and asks for JSON conforming to an answer_template.json. They contain no task-specific final values — fill every value from the live workbench at runtime.

When to use

Trigger this skill when the task:

• assigns a side (buyer-side / seller-side / committee counsel) and a deal_id (the PRJ_* pattern);
• references a deal workbench at <TASK_ENV_BASE_URL> and lists API entry points;
• asks for a structured JSON deliverable conforming to input/payloads/answer_template.json, with no narrative outside the JSON.

0. Resolve workbench access (network config comes from ONE place)

• Read environment_access.md in the working directory. It is the only source for:
  • the base URL — the GDPEVO_ENV_BASE_URL value that <TASK_ENV_BASE_URL> in the prompt maps to;
  • the read-only SQL token for POST /api/query;
  • the allow-list of endpoints.
• Do not hardcode the base URL or token, and do not invent endpoints outside the allow-list. Re-read this file each run.

1. Parse the engagement

• Side → buyer-side vs seller-side. This selects the playbook (PB_BUYER_A / PB_SELLER_A pattern — use the exact playbook_id named in the prompt) and the protective direction every term is judged against.
• deal_id → take it verbatim from the prompt. Never substitute a similarly-named project's records.
• Transaction type → APA, SPA, or carveout APA (drives which mechanics apply: cap-table/holder allocation for SPA, Section 1060 + transfer-tax + TSA for carveout, etc.).
• Deliverable shape → read answer_template.json first; it defines every required field, enum, and stable ID list. The template is the contract.

2. Gather the deal record

Pull the records the prompt lists and the records the template's fields imply. Minimum set (see references/workbench_endpoints.md for what each returns):

• deal record + terms, side-appropriate playbook rules, policy thresholds (if committee/escalation);
• risk-estimates, employees, consents, material-contracts, regulatory, diligence-findings, benchmarks, cap-table, notes, documents — as the template requires.
• Use POST /api/query (token from environment_access.md) only for cross-table joins; it is read-only and secondary to the GET endpoints.

3. Compare draft against the correct playbook / policy

• Map each current draft term to its matching playbook rule or policy threshold.
• Classify with the template's issue_status enums: in_policy, out_of_policy, missing_required_term, draft_exceeds_playbook, draft_below_playbook.
• Draft silence is an issue when your side's position requires an affirmative provision — record it as missing_required_term (e.g., seller needs an escrow release trigger; buyer needs an HSR clearance condition; carveout needs a Section 1060 allocation).
• Exclude distractors per the task's scope: drop stale, in-policy, or out-of-scope terms unless the task explicitly asks for them (e.g., a committee escalation lists only out-of-policy / restricted terms).

4. Quantify from the correct base

• Compute dollar amounts from the deal's headline purchase price unless a source explicitly states a different basis (upfront cash, identified findings, etc.).
• For each issue derive draft vs. preferred vs. fallback amounts/months/percent and the delta to fallback.
• Use the workbench's risk estimates for exposure low/high ranges and cite the source estimate_id.

5. Formatting discipline (follow the precision the prompt/template states)

• Currency → integer USD dollars.
• Percent points → decimal to the precision stated in that task's prompt/template (commonly two decimals; sometimes one decimal or whole points; holder/security percentages may require four decimals).
• Months → integers. Dates → YYYY-MM-DD.
• Stable source IDs only — reuse term_id, consent_id, contract_id, finding_id, employee_id, holder names, and estimate_id exactly as the workbench returns them. Never invent IDs.
• Every enum and every stable ID must come from the answer template's allowed_enums / possible_issue_ids / stable_*_ids lists.

6. Assemble the deliverable

• Build one JSON object matching answer_template.json exactly: top-level fields, nested objects, and array-element shapes.
• Sort as instructed: issues by issue_id ascending or by counsel workflow; provide priority_order from highest to lowest negotiation priority.
• Include the aggregate/summary block the template requires: issue and risk counts, quantified exposure low/high, negotiation deltas, consent/employee/PTO totals, closing-blocker counts, final-position / overall-posture / closing-readiness classification.

7. Output

• Return only the JSON object — no prose, no markdown fences, no commentary.
• Validate it parses and conforms to the template before finishing.

References

• references/workbench_endpoints.md — endpoint catalog and the fields each returns.
• references/output_cheatsheet.md — per-field type/precision and enum quick reference.