Business harness builder

Business Harness Builder

Build an AI-ready project harness: a folder that helps humans and AI agents understand the goal, read the right context, produce consistent outputs, preserve evidence, and continue work across sessions.

Use This Skill When

• The user says "build a harness", "make this project AI-ready", "make a one-shot import package", "handoff this to another AI", or "turn these materials into a reusable collaboration workspace".
• The project is business-facing: consulting, research, whitepaper, product discovery, transformation, training, workshop, delivery playbook, or team onboarding.
• The user needs newcomers to follow a repeatable workflow without knowing how to operate AI tools deeply.

Do not use this skill for low-level coding-agent runtime design, MCP setup, provider configuration, or subagent orchestration unless the user explicitly wants a business project harness around those topics.

Core Idea

A harness is not a folder dump. It is a project operating system with four loops:

1. Goal loop: what are we trying to achieve?
2. Context loop: what should AI read first, later, or never?
3. Evidence loop: what facts support each conclusion?
4. Delivery loop: what output is due, how is it reviewed, and what changes next?

Workflow

1. Classify the project.

   • Identify the project type: consulting, research, whitepaper, product discovery, workshop, transformation, or general project.
   • Identify the audience: solo user, new team, client team, external partner, or public users.
   • Identify the expected artifact: document, slide deck, demo, plan, backlog, research report, or operating system.

2. Build the harness spine.

   • Create or propose these files:
     • 00_START_HERE.md
     • 01_MASTER_PROMPT.md
     • 02_PROJECT_BRIEF.md
     • 03_SOURCE_OF_TRUTH.md
     • 04_EVIDENCE_LOG.md
     • 05_DECISION_LOG.md
     • 06_STAGE_GATES.md
     • 07_TASK_BOARD.md
     • 08_REVIEW_LOOP.md
     • 09_HANDOFF_NOTE.md
   • Create folders:
     • 10_source_materials/
     • 20_work_outputs/
     • 30_templates/
     • 40_reviews/
     • 90_archive/

3. Choose the right templates.

   • For whitepaper projects, read references/business-project-patterns.md.
   • For consulting or FDE-like projects, read references/business-project-patterns.md and references/harness-principles.md.
   • For evaluation and review setup, read references/evaluation-gates.md.
   • For market positioning or open-source explanation, read references/market-landscape.md.

4. Write the master prompt.

   • Make the prompt tell AI who it is, what files to read first, what to produce, what not to assume, and how to update the evidence and decision logs.
   • Include a "stop and ask" rule when source materials are missing or facts are uncertain.

5. Add stage gates.

   • Use gates rather than vague progress:
     • G0 Intake complete
     • G1 Context mapped
     • G2 Evidence base usable
     • G3 First synthesis produced
     • G4 Review feedback resolved
     • G5 Handoff-ready
     • G6 Iteration archived

6. Add review rules.

   • Every major output must have an acceptance checklist.
   • Every conclusion must cite evidence or be labeled as an assumption.
   • Every unresolved question must be placed in 04_EVIDENCE_LOG.md or 07_TASK_BOARD.md.

7. Package for handoff.

   • Explain how a new person should import the folder into an AI tool.
   • Include a copyable first message from 01_MASTER_PROMPT.md.
   • Include "what good looks like" in 08_REVIEW_LOOP.md.

Output Standard

When producing a harness, return:

• Folder path or package path.
• What was created.
• How the user should start.
• Any missing inputs or risks.
• Suggested next iteration.

Keep language practical. Prefer concrete templates and checklists over abstract theory.