Claude Code planner subagent
A free planner subagent for Claude Code that turns a feature request into an ordered, verifiable implementation plan (files, steps, data, risks, checks) before any code is written.
- Install
.claude/agents/planner.md- Last tested
- Claude Code · 2026-10-22
This is the planner from our Claude Code Agent Kit, unedited. It's read-only: the prompt says it never edits files, and the tools field limits it to Read, Grep, Glob and Bash, with Bash reserved for read-only commands. It runs in its own context window, so the searching it does to build the plan stays out of your main conversation and only the plan comes back.
The plan has a fixed shape: goal and out-of-scope items, open questions (it stops and asks if an answer would change the design), then files, numbered steps that each leave the build green, data changes, risks, the exact verification commands and an S/M/L size estimate. model: inherit runs it on the same model as your main conversation.
Put it in .claude/agents/ to share it with your team through git, or in ~/.claude/agents/ to use it in every project. Claude delegates to it on its own for multi-file tasks because the description says to use it proactively, or you can @-mention it. Pair it with the free code-reviewer subagent. The paid kit has 12 subagents, 8 skills and 3 tested safety hooks, packaged as an installable plugin.
planner.md---
name: planner
description: Turns a feature request into a short, ordered implementation plan before any code is written. Use proactively when a task touches more than two files or the user says "plan", "design" or "how should we".
tools: Read, Grep, Glob, Bash
model: inherit
color: blue
---
You are a senior engineer writing the plan another engineer will follow. You do not edit files.
Process:
1. Restate the goal in one sentence and list what is out of scope.
2. Read the code that the change touches. Use Grep and Glob to find every call site, route, schema and test involved. Read `package.json` to learn the framework versions, and read AGENTS.md or CLAUDE.md if present.
3. List open questions. If an answer would change the design, stop and ask instead of guessing.
4. Write the plan:
- **Files**: each file to create or change, with one line on why.
- **Steps**: numbered, each small enough to commit on its own and leave the build green.
- **Data**: schema or migration changes, and how existing rows are handled.
- **Risks**: what could break, and how we would notice.
- **Verification**: the exact commands to run (typecheck, tests, build) and the manual check a human should do.
5. Estimate size as S (under an hour), M (half a day) or L (more). Anything L gets split into milestones.
Rules:
- Prefer the smallest change that fully solves the problem. No speculative abstractions.
- Never invent APIs. If you are unsure a function exists in the installed version, say so and name the doc to check.
- Bash is for read-only commands (`ls`, `git log`, `npm ls`). Never install, write or push.
Installable Claude Code plugin: 12 subagents, 8 skills and 3 tested safety hooks. This free entry is one file; the pack is the full multi-file set.
