SOW generator

Credit to OneWave AI · From GitHub OneWave-AI/claude-skills · License: MIT · Not tested by us · Added October 29, 2026

# Statement of Work Generator

Generate complete, client-ready Statements of Work from a project brief, meeting the standards of Big Four consulting firms, top-tier agencies, and enterprise procurement. Produce polished documents ready for client review with minimal edits, never rough drafts or templates with blanks.

## Contents

- `references/required-inputs.md` -- inputs to gather and pre-generation client research
- `references/document-structure.md` -- the 16-section SOW structure with templates for every section
- `references/pricing-templates.md` -- fixed-price, time-and-materials, and retainer pricing models (Section 11)
- `references/legal-terms.md` -- confidentiality, IP, warranties, liability, termination, and other legal clauses (Section 14)
- `references/change-management.md` -- change request and change order process (Section 12)
- `references/risk-management.md` -- risk register and escalation path (Section 13)

## Workflow

1. Gather the required inputs. Read `references/required-inputs.md`. If any required item is missing, ask for it explicitly. Do not guess at critical commercial terms.
2. Research the client. Use WebSearch per `references/required-inputs.md` to gather company overview, recent news, technology footprint, and regulatory environment. If research yields nothing, proceed on user-provided inputs and flag assumptions explicitly. Never fabricate company information.
3. Select the pricing model from `references/pricing-templates.md` based on engagement type (default: fixed-price).
4. Generate all 16 sections following `references/document-structure.md`, pulling Section 11 from pricing-templates, Section 12 from change-management, Section 13 from risk-management, and Section 14 from legal-terms. Replace every bracketed placeholder with real values from inputs and research. Leave brackets only where the user explicitly deferred. Mark gaps with `[ACTION REQUIRED: ...]`.
5. Enforce traceability: every deliverable traces to a scope item, every milestone references deliverables, every payment trigger references a milestone, and all cross-referenced IDs (D-, M-, DEP-, R-, CR-, CO-) are correct.
6. Verify against the quality checklist below, then write the finished SOW to `sow.md` in the working directory (or a user-specified path).

## Output Requirements

- Comprehensive: typically 500-800 lines for a standard engagement, longer for complex multi-phase projects.
- Clean Markdown with proper heading hierarchy, tables, and numbered lists.
- Professional, precise, unambiguous third-person prose. Define any non-standard jargon on first use.
- No emojis anywhere in the document.
- Include a footer note that the document is a template and should be reviewed by qualified legal counsel before execution. Never provide legal advice.

## Quality Checklist

Verify before finalizing:

- All user-provided inputs are reflected accurately
- Client research is incorporated into Background and Context
- Every in-scope item has corresponding deliverables
- Every deliverable has measurable acceptance criteria
- Every milestone has a date, associated deliverables, and payment trigger (if applicable)
- Out-of-scope section is specific to this engagement, not generic
- Assumptions are realistic and comprehensive; dependencies have owners and deadlines
- Risk register contains engagement-specific risks, not boilerplate
- Payment schedule sums to the total engagement fee
- Legal sections are complete and internally consistent
- All cross-references resolve correctly
- No placeholder brackets remain except those explicitly deferred by the user
- Document reads as a cohesive narrative, not a filled-in template
- Tone is professional and consistent throughout

## Notes

- For regulated industries (healthcare, finance, government), include industry-specific compliance sections (Section 15) and note additional legal review requirements.
- Ask clarifying questions whenever the scope is ambiguous enough that the SOW could be interpreted in materially different ways. Ambiguity in an SOW is a professional failure.

Fill in anything in [brackets] or {braces} with your own details first.

This is my prompt

Wrote this prompt? Tell me, and I will put your name on it. I check every claim by hand.

Report or remove
What is wrong?

More Everything else prompts

Everything else

Investor research

When the user wants to identify, evaluate, or prioritize potential investors for a fundraising round. Also activates when the user asks "who should I pitch?", "find me investors", "build an investor list", or mentions...

You needAny AI

Credit to Shawn Pang

Everything else

Market landing

Perform a comprehensive Conversion Rate Optimization (CRO) analysis on any landing page. This skill produces a section-by-section teardown with prioritized, actionable fixes that directly impact conversion rates. - User...

You needAny AI

Credit to Zubair Trabzada

Everything elseLinkedIn

Market launch

Generate a complete, week-by-week launch playbook for any product, service, or feature launch. This skill produces a tactical plan with templates, checklists, email sequences, social posts, and metrics tracking --...

You needLinkedInAny AI

Credit to Zubair Trabzada

Our tested version

Everything else

Market proposal

Generate a professional, client-ready marketing services proposal. This skill produces a complete proposal document that positions the agency/consultant as the clear choice, frames pricing with anchoring and tiered...

You needAny AI

Credit to Zubair Trabzada

Everything else

Commercial lease analyzer

Decode a commercial lease before signing or renewing -- true all-in occupancy cost (base rent plus NNN/CAM plus escalations), the clauses that hurt small tenants (personal guarantees, relocation, restoration), and the...

You needAny AI

Credit to OneWave AI

Everything else

CSV excel merger

Merges, appends, joins, and deduplicates multiple CSV, TSV, and Excel files (including multi-sheet workbooks) with pandas - mapping mismatched column names to one schema, normalizing keys, resolving conflicting values,...

You needAny AI

Credit to OneWave AI

Our tested version

All Everything else promptsAll prompts

Want a prompt for your own work? Ask for one. Every prompt here is free.

Want this built for you? Get it built.