Writing Plans

Run before building anything estimated at >30 minutes.

Externally maintained specifications

If PRDs, RFCs, or other authoritative docs live outside oh-my-colab conventions, run document-intake first so .ohc/doc-sources.md and reconciliation are settled. Shorten Step 1 to unanswered questions, document↔repository conflicts, and ambiguous acceptance. Always include Sources and Traceability (assets/plan-template.md).

Step 1: Deep Interview (3–5 questions max)

Clarify before designing:

1. What is done? — the specific verifiable end state
2. What is NOT in scope? — explicit boundaries prevent scope creep
3. What are the constraints? — performance, compatibility, deadline
4. Are there existing patterns? — check PROJECT.md conventions
5. What could go wrong? — risks to address upfront

Use .ohc/PROJECT.md as the primary project context. Also use optional domain docs (CONTEXT.md / CONTEXT-MAP.md) when present. If the plan creates a durable architectural decision, record whether an ADR is needed.

Step 2: Approach Selection

Propose 2-3 approaches with tradeoffs. Get user confirmation on approach before writing the full plan. See brainstorming skill for format.

Step 3: Task Decomposition Rules

Each task must be:

• ≤2 hours — if longer, split it
• Single responsibility — one logical change
• Independently verifiable — has its own success criterion
• Clear file scope — lists which files will change
• Agent-assigned — specifies which agent does it

Step 4: Write and Save

Save to: .ohc/plans/{kebab-case-name}.md Use template from assets/plan-template.md.

Step 5: Gate

Do NOT proceed to BUILD without explicit user confirmation. "Plan ready. Does this look right? Reply 'go' to start building."

See references/plan-template.md for canonical format. See references/task-decomposition.md for splitting rules. See references/domain-docs.md for glossary and ADR-aware planning.