Fewshot Attempt 02

M&A Deal Workbench — Structured Deal Review Skill

Overview

This skill enables you to act as deal counsel reviewing an M&A transaction using a running M&A Deal Workbench REST API. You gather deal records, draft terms, playbook/policy rules, benchmarks, risk estimates, and related diligence data, then produce a structured JSON answer conforming to a task-specific answer template.

Activation

Invoke this skill when the task involves:

• A deal ID (e.g. PRJ_*)
• An M&A deal workbench base URL (provided as <TASK_ENV_BASE_URL>)
• An answer template JSON (provided as input/payloads/answer_template.json)
• A playbook ID or policy ID for comparison
• Any of these review types: issue register, closing package, escalation memo, transition review, or deviation matrix

Core Workflow

Phase 1 — Read the Task Materials

1. Read the prompt (input/prompt.txt) to understand the deal role (buyer/seller counsel), deal ID, playbook/policy IDs to compare against, and any deal-specific instructions.
2. Read the answer template (input/payloads/answer_template.json) to understand the exact output schema, allowed enums, stable IDs, and required fields.

Phase 2 — Gather Deal Data from the Workbench

Use the workbench base URL (<TASK_ENV_BASE_URL>) to fetch all relevant records. The API surface includes:

Read-only SQL (if available): Use POST /api/query with header Authorization: Bearer deal-workbench-readonly for cross-table verification. Body: {"sql": "<query>"}.

Start by fetching the deal record, terms, and the applicable playbook/policy rules. Then fetch supporting records (risk estimates, employees, consents, regulatory, benchmarks, notes, documents) as directed by the answer template's schema.

Phase 3 — Compare Draft Terms Against Playbook/Policy

For each relevant term or issue area:

1. Find the draft term in /api/deals/<deal_id>/terms — note its term ID, clause reference, and numeric values (percentages, dollar amounts, months).
2. Find the playbook/policy position in the rules or thresholds endpoint — note the preferred value and fallback value.
3. Calculate deltas — draft minus fallback, draft minus preferred, etc.
4. Classify status using the answer template's allowed enums:
   • in_policy — draft meets or exceeds the playbook position
   • out_of_policy — draft violates a policy threshold
   • missing_required_term — no draft term exists for a required provision
   • draft_exceeds_playbook — draft imposes more burden on your client than the playbook allows
   • draft_below_playbook — draft gives your client less protection than the playbook requires
5. Assign risk ratings (LOW, MEDIUM, HIGH) based on quantified exposure and closing impact.
6. Determine recommended actions (delete, revise, add, accept, escalate, approve, approve_with_conditions, reject).

Phase 4 — Compute Dollar Amounts and Metrics

Always calculate from the deal's headline purchase price unless a source record explicitly states a different basis.

Unit conventions (from the answer template's units block):

• Currency amounts: integer USD (round to nearest dollar)
• Percentages: decimal number to the template's specified precision (typically 1 or 2 decimal places)
• Months: integer months
• Holder percentages: to four decimal places when specified

Calculation patterns:

• draft_amount_dollars = headline_value × draft_percent / 100
• delta_to_fallback_dollars = draft_amount_dollars - fallback_amount_dollars (positive when draft is worse for your client)
• shortfall_dollars = fallback_amount_dollars - draft_amount_dollars (positive when draft is below playbook)
• total_quantified_exposure = sum of relevant risk estimate low/high values
• negotiation_delta = sum of all deltas between draft and fallback positions

Phase 5 — Build the Output

1. Sort issue arrays as directed by the template (typically by issue_id ascending or by negotiation priority).
2. Use only stable IDs from the workbench — term IDs, consent IDs, employee IDs, contract IDs, finding IDs, risk estimate IDs.
3. Use only allowed enum values from the template.
4. Include every required field from the template, even when the value is null.
5. Return only valid JSON — no explanatory prose, no markdown fences, no trailing text.
6. Do not fabricate values — if a value is not found in the workbench, use null, 0, or "not_found_in_current_records" as the template dictates.

Common Issue Categories

These issue categories recur across deal review types. Map them to the template's possible_issue_ids or issue_id enums:

Cross-Task Patterns

When the task is an issue register / deviation matrix (trains 001, 005):

• Compare each draft term against playbook preferred and fallback positions
• Compute deltas, shortfalls, and quantified exposures
• Produce a priority-ordered list
• Include summary metrics (counts, totals, exposures)

When the task is a closing/economics package (train 002):

• Extract economics from deal record + cap table
• Allocate consideration across holders proportionally
• Evaluate indemnity, escrow, NWC mechanics
• Assess closing readiness: blockers vs tradeable issues
• Check regulatory status (HSR, hell-or-high-water)

When the task is a committee escalation memo (train 003):

• Filter to ONLY out-of-policy or restricted terms
• Exclude in-policy, stale, and non-committee terms
• Provide policy comparison, benchmark support, delta, exposure for each
• Aggregate risk counts, exposure totals, and committee routing

When the task is a transition/carveout review (train 004):

• Focus on separation terms: IP transition, TSA, employees, tax allocation, governing law
• Produce paired issue + redline entries with must-have terms
• Compute operational risk: stranded costs, PTO liability, consent amounts at risk

Error Avoidance

• Never assume a provision is present or absent — verify by fetching the terms endpoint.
• Never mix deal records — each deal ID has its own data; PRJ_JUNIPER records do not apply to PRJ_ORION.
• Empty arrays are meaningful — use [] when the template expects an array but no values exist, vs null when the field is not applicable.
• Verify calculations — recompute dollar amounts from percentages against the headline purchase price; cross-check with risk estimates.
• Watch for distractor terms — stale, in-policy, or non-committee terms that should be excluded per the task instructions.