# Plan

> Create a detailed TDD implementation plan with parallel tasks from a feature spec or description.

- Skill: `conava/plan` (Agent Skill)
- Install (CLI): `npx skillmds@latest add conava/plan`
- Raw SKILL.md: https://api.skillmd.com/api/skills/conava/plan/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: Conava (https://skillmd.com/u/conava)
- Updated: 2026-09-10
- Page: https://skillmd.com/skills/conava/plan

---


# Plan — Feature Spec to Implementation Plan

Launch the **planner** agent in the **foreground** with the user's feature specification or description.

## Interactive Loop

The planner may return a `STATUS: NEEDS_INPUT` response instead of a finished plan. This response can contain any combination of:
- **Design Questions** — architectural or behavioral choices that need user input
- **Code Issues Found** — bugs, flaws, or tech debt discovered in the blast radius that the user should decide whether to fix
- **Proactive Improvements** — issues the planner is unsure about including (bake into existing task, standalone task, or skip)

When this happens:

1. Present ALL sections to the user using **AskUserQuestion** — for design questions, map each to an option set with the planner's recommendation marked "(Recommended)". For code issues and improvements, present as include/skip choices.
2. **Resume** the planner agent (using the `resume` parameter with its agent ID) with the user's answers
3. Repeat if the planner returns more questions or findings — the planner is encouraged to iterate multiple times to refine the plan

## Phase Context (Multi-Phase Pipelines)

When the devline orchestrator invokes this skill during a multi-phase pipeline, it passes phase context to the planner:

- **Phase number** — which phase is currently being planned (e.g., 2 of 3)
- **Prior plan file paths** — paths to all already-completed phase plans (e.g., `.devline/plan-phase-1.md`)
- **Output path** — the planner writes to `.devline/plan-phase-N.md` instead of `.devline/plan.md`

Pass this context through when launching the planner agent. The planner uses it to scope the plan to the current phase and avoid re-planning prior phases' work.

When no phase context is provided (the default), the planner writes to `.devline/plan.md` exactly as today — this path is unchanged.

## After Planning

Once the planner has all answers, it will:
1. Write the full plan to `.devline/plan.md` (single-phase) or `.devline/plan-phase-N.md` (multi-phase)
2. Return a concise summary (architecture overview, tasks, key decisions)

Present the summary to the user for approval. The full plan lives at the output path — implementers read it directly from disk.

This skill does NOT automatically continue into implementation. To proceed, run `/devline:implement` with the approved plan, or `/devline` for the full pipeline.