Imported from sebastianspicker/rae (packages/orchestration/adapters/gemini/skills/orchestration-plan/SKILL.md). Install upstream with npx skills add sebastianspicker/rae --skill orchestration-plan. Copyright stays with the author.

/plan - Deterministic Execution Plan (Gemini Adapter)

Use this when

• /ar output is approved.
• The user requests /plan, implementation planning, or task decomposition.

Model tier

Use a balanced-reasoning model. Task decomposition is systematic and bounded; full high-reasoning is not required.

Semantic intent

• Predeclared verification: define tests and acceptance checks before coding starts.
• Context minimization: task groups should be specific enough that builders do not need extra narrative.

Input

• .pipeline/runs/<run-id>/design.json
• .pipeline/runs/<run-id>/review.json

Procedure

1. Build task groups

Partition work into independent groups (target 3-6 tasks/group, max 8). If a group falls outside 3-6 tasks, add scope_override.reason in that group and justify why. Assign builder_tier per group: high_reasoning for groups requiring architectural judgment or complex refactoring, balanced for structured implementation with reasoning, fast for straightforward implementation tasks.

2. Assign file ownership

Create explicit path->group mapping and remove overlaps.

3. Provide executable code patterns

For each task, include concrete patterns (imports/signatures/structure), not placeholders.

4. Define tests upfront

Each task needs named test cases containing:

• setup,
• assertion,
• expected outcome.

5. Define acceptance criteria

Attach non-negotiable completion checks per task.

6. Define verification commands

Include exact commands for lint/format-check/type/build/test checks.

7. Run pre-final drift check

Compare plan against design (via /pmatch flow or multi-model-review drift mode). Resolve structural drift before finalizing.

8. Build artifact

Write plan.json conforming to contracts/artifacts/execution-plan.schema.json:

• .pipeline/runs/<run-id>/plan.json

9. Gate evaluation

Validate:

• schema compliance,
• at least one test case per task,
• file ownership conflicts absent,
• no TODO/placeholders in required code patterns,
• verification command list present,
• task-group granularity in 3-6 range unless scope_override.reason is provided.

Write gate output to:

• .pipeline/runs/<run-id>/gates/plan-gate.json

Non-negotiables

• If builders must infer missing intent, planning is incomplete
• No placeholder implementation guidance
• Test design happens before build, not after