SOW Generate

Generate a client-ready Statement of Work with three named scope packages. Operate in one of two modes depending on arguments.

Parse Arguments

Extract these values from $ARGUMENTS:

• [brief] (string, optional) — inline project brief text. If not provided, ask the user interactively.
• --team (boolean, default: false) — activate full 6-agent competing-hypotheses pipeline mode.
• --client=NAME (string, optional) — client name for the SOW header. If not provided, infer from the brief or ask the user.
• --budget=AMOUNT (string, optional) — maximum budget constraint (e.g., $50,000 or 50000). If not provided, pricing agent uses market rates.
• --weeks=N (integer, optional) — maximum timeline in weeks. If not provided, derive from scope.
• --output=PATH (string, default: ./sow-output/) — output directory for the generated SOW file.

Business Context (Optional)

Check if context files exist at _infrastructure/context/active/. If the directory contains .md files, read business-info.md, strategy.md, and current-data.md. Use this context to personalize output (e.g., prioritize known clients, use correct terminology, align with current strategy). If files don't exist, skip silently.

Step 0: Memory Context

Read the context-injection skill at _infrastructure/memory/context-injection/SKILL.md. Query for memories relevant to the current input (company, contacts, topics detected in arguments). If memories are returned, incorporate them into your working context for this execution.

Context Loading

Before generating in either mode:

1. Check if Notion MCP is available. If yes, search Notion for historical SOWs matching the client name, industry, or project type using notion-search with the query "SOW". Summarize any relevant historical scope, pricing, or risk patterns found. Note to the user if historical context was found and what it informed.
2. Check if a .sow-history file exists in the output directory. If it exists, read it and use its contents as additional calibration context for scope and pricing.
3. If neither Notion nor .sow-history is available, proceed without historical context and note this to the user.

Mode 1: Default (Single-Agent Quick SOW)

When --team is NOT present:

1. Read ALL 3 skills:

   • ${CLAUDE_PLUGIN_ROOT}/skills/sow/scope-definition/SKILL.md
   • ${CLAUDE_PLUGIN_ROOT}/skills/sow/sow-writing/SKILL.md
   • ${CLAUDE_PLUGIN_ROOT}/skills/sow/risk-assessment/SKILL.md

2. Collect the project brief:

   • If a [brief] argument was provided, use it as the project description.
   • If no brief was provided, ask the user for all of the following:
     • Client name and industry
     • Project description and goals
     • Key deliverables or features required
     • Budget constraint (or confirm none)
     • Timeline constraint in weeks (or confirm none)
     • Key priorities (e.g., speed, quality, cost, flexibility)
   • Do not proceed until at least a project description and client name are available.

3. Run context loading (see Context Loading section above).

4. Generate three SOW options internally applying the rules from all three skills:

   Option A — Conservative (Foundation Package)

   • Apply scope-definition skill: P90 confidence level, 20% scope buffer, include only explicitly stated requirements
   • Apply risk-assessment skill: score all risk dimensions, flag high-risk items, recommend mitigations
   • Apply pricing: 30% margin on estimated effort; derive from skills pricing tables

   Option B — Balanced (Growth Package)

   • Apply scope-definition skill: P75 confidence level, 10% scope buffer, include stated requirements plus one logical adjacent deliverable
   • Apply risk-assessment skill: score all risk dimensions, moderate risk tolerance
   • Apply pricing: 25% margin on estimated effort

   Option C — Ambitious (Transformation Package)

   • Apply scope-definition skill: P60 confidence level, 0% scope buffer, include full vision with stretch deliverables
   • Apply risk-assessment skill: score all risk dimensions, note elevated risk areas clearly
   • Apply pricing: 20% margin on estimated effort

   Respect --budget and --weeks constraints: if an option exceeds the constraint, adjust scope downward and note the change.

5. Determine the recommended option: the option whose risk score is Medium or below with the highest value delivery. Typically Option B unless budget or timeline constraints make Option A the only viable choice.

6. Create the output directory if it does not exist.

7. Derive the output filename as sow-[client-slug]-[YYYY-MM-DD].md where [client-slug] is the client name lowercased with spaces replaced by hyphens and [YYYY-MM-DD] is today's date.

8. Write the full SOW to ${output}/sow-[client-slug]-[YYYY-MM-DD].md using the three-option document format from the sow-writing skill. The document must include:

   • Cover page (client name, project name, date, prepared by)
   • Executive summary
   • All three option sections with deliverables, timeline, pricing, and risk summary
   • Comparison table across all three options
   • Terms and conditions section
   • Signature block

9. Present the output summary to the user:

## SOW Generated

**Client**: [Client Name]
**Output**: [output path]
**Options**: 3 ([Option A name] · [Option B name] · [Option C name])
**Recommended**: [Option B name] (Option B)

### Option Summary
| Option | Timeline | Price | Risk |
|--------|----------|-------|------|
| [Option A name] | N weeks | $XX,XXX | Low |
| **[Option B name] ✓** | N weeks | $XX,XXX | Medium |
| [Option C name] | N weeks | $XX,XXX | High |

_View the full SOW at [output path]_

10. Notion tracking (if Notion MCP is available):

11. Database discovery (ordered):

    • Search for "[FOS] Deliverables" database first (preferred)
    • If not found, fall back to "Founder OS HQ - Deliverables" database
    • If not found, fall back to legacy "SOW Generator - Outputs" database
    • If none exists, skip Notion tracking (do NOT create the database)

12. Resolve Company + Deal relations (only when writing to "[FOS] Deliverables" or "Founder OS HQ - Deliverables"):

    • Look up the client name in the CRM Pro "Companies" database
    • If a matching company is found, set the Company relation property
    • If a deal was identified during context search, set the Deal relation property
    • If no match is found, leave relation properties empty

13. Idempotent upsert: Check for existing record with same Client Name + Project Title. When using "[FOS] Deliverables" or "Founder OS HQ - Deliverables", also filter by Type = "SOW" to avoid collisions with other deliverable types.

14. Create or update the record:

    • Type: "SOW" (when writing to HQ Deliverables)
    • Title: Project name
    • Status: "Draft"
    • Amount: Recommended option's quoted price
    • File Path: Output file path
    • Sources Used: Data sources consulted
    • Generated At: Current date/time

Mode 2: Team Pipeline (--team)

When --team IS present:

1. Read ${CLAUDE_PLUGIN_ROOT}/agents/sow/config.json.

2. Collect the project brief:

   • If a [brief] argument was provided, use it.
   • If no brief was provided, ask the user for: client name, project description and goals, key deliverables, budget constraint, timeline constraint, and key priorities.
   • Do not proceed until at least a project description and client name are available.

3. Run context loading (see Context Loading section above).

4. Assemble a structured context object to pass to all Phase 1 agents:

   {
     "brief": "[project brief text]",
     "client": "[client name]",
     "budget_constraint": "[budget or null]",
     "weeks_constraint": "[weeks or null]",
     "historical_context": "[summary from Notion/.sow-history or null]",
     "skills_paths": {
       "scope_definition": "${CLAUDE_PLUGIN_ROOT}/skills/sow/scope-definition/SKILL.md",
       "sow_writing": "${CLAUDE_PLUGIN_ROOT}/skills/sow/sow-writing/SKILL.md",
       "risk_assessment": "${CLAUDE_PLUGIN_ROOT}/skills/sow/risk-assessment/SKILL.md"
     }
   }
   

5. Execute Phase 1 — spawn all three scope agents in parallel using the Task tool. Each agent reads its definition from its file path in teams/config.json and receives the structured context object. The agents are independent and must not communicate with each other.

   • scope-agent-a — conservative interpretation, P90 confidence, 20% buffer
   • scope-agent-b — balanced interpretation, P75 confidence, 10% buffer
   • scope-agent-c — ambitious interpretation, P60 confidence, 0% buffer

   Wait for all Phase 1 agents to complete before proceeding. If fewer than 2 agents succeed (per minimum_hypotheses_required in config), halt and report the failure. If exactly 2 of 3 succeed, proceed and note which agent failed.

6. Collect Phase 1 outputs (three scope proposals). Structure them as a combined proposals object.

7. Execute Phase 2 — spawn risk-agent and pricing-agent in parallel using the Task tool. Each agent reads its definition from its file path and receives the structured context object plus all Phase 1 proposals.

   • risk-agent — scores risk dimensions for each proposal
   • pricing-agent — estimates cost and validates value-for-money for each proposal

   Wait for both Phase 2 agents to complete before proceeding.

8. Execute Phase 3 — invoke sow-lead with the structured context object plus all Phase 1 and Phase 2 outputs. The lead agent reads its definition from ${CLAUDE_PLUGIN_ROOT}/agents/sow/sow-lead.md and synthesizes all inputs into the final SOW document. The lead writes the output file to the --output directory.

9. Present the pipeline completion summary:

## SOW Pipeline Complete

**Client**: [Client Name]
**Output**: [output path]
**Pipeline**: 6 agents (3 scope · 1 risk · 1 pricing · 1 lead)

### Agent Timing
| Agent | Duration | Status |
|-------|----------|--------|
| scope-agent-a | Xs | Done |
| scope-agent-b | Xs | Done |
| scope-agent-c | Xs | Done |
| risk-agent | Xs | Done |
| pricing-agent | Xs | Done |
| sow-lead | Xs | Done |

### Scoring Matrix
| Option | Scope Score | Risk Score | Pricing Score | Total |
|--------|-------------|------------|---------------|-------|
| [Option A name] | X/10 | X/10 | X/10 | X/30 |
| [Option B name] | X/10 | X/10 | X/10 | X/30 |
| [Option C name] | X/10 | X/10 | X/10 | X/30 |

**Recommended**: [Recommended option name]

_View the full SOW at [output path]_

Error Handling

• If no brief is provided and the user declines to provide one: halt with the message "A project brief is required to generate a SOW. Please provide a description of the project, goals, and any constraints."
• If Notion MCP is unavailable: skip historical context lookup, proceed with filesystem-only, note "Notion unavailable — proceeding without historical SOW context."
• If the output directory does not exist: create it with Write before writing the SOW file.
• If --budget is not provided: generate all three SOW options without a budget ceiling; the pricing agent uses market rates derived from the sow-writing skill pricing tables.
• If --weeks is not provided: derive timeline from scope estimates in each option; do not artificially constrain.
• In team mode, if a Phase 1 agent fails: log the failure, continue with remaining agents if at least 2 succeed.
• In team mode, if a Phase 2 agent fails: log the failure, have sow-lead note the missing analysis and complete synthesis with available data.

Final: Memory Update

Read the pattern-detection skill at _infrastructure/memory/pattern-detection/SKILL.md. Log this execution as an observation with: plugin name, primary action performed, key entities (companies, contacts), and output summary. Check for emerging patterns per the detection rules. If a memory reaches the adaptation threshold, append the notification to the output.

Usage Examples

/founder-os:sow:generate "Build a customer portal for Acme Corp with dashboard, auth, and reporting"
/founder-os:sow:generate --client="TechCo" --budget=75000 --weeks=16
/founder-os:sow:generate --team "Redesign the onboarding flow for a SaaS product"
/founder-os:sow:generate --team --client="StartupXYZ" --output=./proposals/
/founder-os:sow:generate                                  # Interactive: asks for brief