1. Gather OpenAPI/Swagger spec, auth docs, rate-limit page, changelog
2. Run Pre-Draft Intake to confirm scope and posture
3. Build Source Register (version-lock every source)
4. Extract constraints into structured tables
5. Draft exhibit sections with

   [BRACKETED]

   placeholders
6. Generate Traceability Matrix and Risk/Gap Log
7. Run Post-Draft Alignment with user

• Agreement context — exhibit placement, provider vs. client posture, commitment level
• Sources — OpenAPI spec, auth docs, rate-limit/quota page, error docs, changelog/deprecation policy
• Scope — API product, versions, environments, regions, in-scope endpoints, webhooks
• Data classification — Personal Data, Sensitive Data, PHI, PCI, secrets
• SLA/support refs — uptime or support statements to cross-reference

• Version and timestamp locked for each source

• servers[].url

  and environment labels captured

• components.securitySchemes

  and operation-level security identified

1. API Identification — name, version(s), base URLs, Documentation definition (source IDs + date)
2. Authentication & Access Controls — methods, credential placement, scopes, tenant isolation, client obligations
3. Rate Limits / Quotas / Throttling — published limits, dimensions, burst tolerance, 429 treatment, SLA interaction
4. Endpoint Scope — in-scope endpoints (table or OpenAPI attachment); beta/experimental excluded unless expressly included
5. Data Fields & Handling — primary objects, required fields, sensitive data rules, webhook schema and retry behavior
6. Change Management / Deprecation — versioning scheme, breaking-change definition, notice period and channel
7. Error Handling — error format, retriable vs. non-retriable errors
8. Support & Incidents — cross-reference to SLA/Support exhibit
9. Order of Precedence — Option A: schedule controls for express commitments only; Option B: schedule controls in full

1. Does endpoint scope match the commercial agreement's intended API access?
2. Should any constraints be elevated from descriptive/as-is to binding?
3. Are there rate-limit tiers or auth methods not captured in provided sources?
4. Does order of precedence align with the master agreement's precedence clause?

• Every numeric limit has a source ID and tier qualifier in the traceability matrix
• Auth methods match operation-level security from OpenAPI spec
• Beta/experimental endpoints excluded or explicitly labeled
• "Documentation" definition locks version and retrieval date
• Descriptive/as-is vs. binding commitments clearly distinguished
• Sensitive data fields flagged and mapped to DPA requirements
• No hard-coded numbers without source reference
• Risk/gap log captures all unresolved items
• Order of precedence consistent with master agreement
• All

  [BRACKETED]

  placeholders clearly marked

• Over-warranting: prefer "as of [date]" with change-control; never warrant undocumented behavior
• Unsourced numbers: every numeric limit needs a source ID and tier qualifier
• 429 and SLA: treat throttling explicitly in SLA calculations
• Beta endpoints: exclude unless expressly agreed
• Regulated data: flag PHI/PCI/PD categories and require appropriate addenda
• Uncertainty: mark with

  [VERIFY]

  for legal/engineering review

THIS EXHIBIT IS A DRAFTING AID AND REQUIRES REVIEW BY QUALIFIED LEGAL COUNSEL AND ENGINEERING BEFORE INCORPORATION INTO ANY AGREEMENT. IT DOES NOT CONSTITUTE LEGAL ADVICE.

• Description: Trimmed from 10 lines with keyword stuffing to a concise third-person summary with clear trigger guidance
• Removed "Why This Skill Exists": Replaced with a 2-sentence overview — the rationale is implicit in the workflow
• Collapsed Checkpoints A/B: Renamed to "Pre-Draft Intake" and "Post-Draft Alignment" with streamlined content
• Removed the

  tags

  field: Not part of the required frontmatter spec
• Eliminated the full exhibit template: Replaced the verbatim 50-line code block with a 9-item numbered list describing each section — the agent can generate the actual text
• Consolidated "Guidelines" and "Quality Audit": Merged into a "Quality Checklist" and a "Pitfalls" section
• Reduced from 225 lines to ~145 lines while preserving all domain-critical tables, checklists, and legal guardrails