plan

name	plan
description	Decompose a feature into tasks with domain agent assignments. Use when starting a new feature, phase, or multi-step implementation. Triggers: plan, decompose, break down, feature planning.

Plan: Feature Decomposition

Steps

1. Gather context. Read the feature request and identify which NanoVec modules are affected (store, index, distance, simd, mcp, server).

2. Read current state. Check the existing code in affected modules to understand what exists and what needs to change:

   find src/ -name '*.rs' -type f | head -30
   cargo test --all-features 2>&1 | tail -5
   

3. Decompose into tasks. Break the feature into atomic, testable tasks. Each task must:

   • Have a single domain agent owner (core, simd, mcp, testing, docs)
   • Be completable independently (or have explicit dependencies)
   • Include acceptance criteria (what test must pass)
   • Follow TDD: specify the test before the implementation

4. Assign agents. Map each task to a domain agent based on the routing table:

   • VectorStore, RecordStore, KD-Tree, distance -> core
   • AVX2, NEON, portable_simd, assembly -> simd
   • MCP tools, rmcp, transport, JSON-RPC -> mcp
   • Tests, proptest, benchmarks, coverage -> testing
   • Documentation, CHANGELOG, API docs -> docs

5. Sequence tasks. Order tasks by dependencies. Common patterns:

   • Scalar implementation (core) before SIMD acceleration (simd)
   • Data structure (core) before MCP tool (mcp)
   • Implementation before tests (unless TDD, then reverse)
   • All code before documentation (docs)

6. Write the plan. Save to docs/plans/<feature-slug>-plan.md using this format:

   # Plan: <Feature Name>
   
   ## Summary
   <one paragraph>
   
   ## Tasks
   
   ### Task 1: <title>
   - **Agent:** <agent name>
   - **Files:** <files to create or modify>
   - **Depends on:** none | Task N
   - **Acceptance:** <test or verification command>
   
   ### Task 2: ...
   

7. Present for approval. Show the plan to the developer and wait for confirmation before execution.