How to use the make-plan agent skill: install and examples

What the make-plan skill makes the agent do

Make-plan changes how the agent approaches multi-step work. Instead of jumping straight into code, the agent first gathers facts from docs and existing examples, often through subagents. It then writes a phased plan where each phase cites specific sources and includes verification checks.

The plan is built to be handed across fresh chat sessions, so each phase is meant to stand on its own. After writing the plan file, the agent presents a short summary in a fixed format.

• Starts with a documentation discovery phase (Phase 0) that identifies the actual APIs and signatures available.
• Each implementation phase names what to copy from docs and where to find it.
• Each phase includes a verification checklist and anti-pattern guards.
• Ends with a final verification phase.

How to install the make-plan skill

Download the make-plan ZIP from this page. The skill comes as a single file, SKILL.md, inside the package.

To install it for yourself across projects, unzip it into ~/.claude/skills/. To install it for one project only, unzip it into that project's .claude/skills/ folder.

If you use the Claude apps, you can instead upload the skill under Settings, then Capabilities, then Skills.

• Personal install: ~/.claude/skills/
• Project install: the .claude/skills/ folder inside your project
• Claude apps: Settings → Capabilities → Skills, then upload

How to trigger make-plan with natural requests

You do not need a special command. The skill's description says it applies when you ask for a plan for a feature, a task, or a multi-step implementation. Ask for a plan in ordinary language and the agent should pick it up.

The source says the skill is especially meant for use before executing with the do skill, so asking for a plan first and then asking for execution is the intended flow.

• Plan a feature for adding CSV export to our reports page, using our existing API docs.
• Make a phased plan for migrating our auth middleware to the new session API.
• Before we build it, plan the multi-step change to the billing webhook handler.
• Create an implementation plan for the notifications feature that we can finish across several chat sessions.

What a make-plan plan contains

Every plan opens with a primary goal written as a user-visible outcome, not a mechanism. Each phase then has a short note explaining how it serves that goal. If a phase cannot be justified against the goal, the skill says it does not belong in the plan.

The skill also asks the agent to require reading documentation rather than assuming it. Plans are expected to cite specific files and lines, and to flag methods that do not exist or deprecated parameters.

• Primary goal: one or two sentences describing what done looks like.
• Phase 0: documentation discovery, with an Allowed APIs list citing sources.
• Implementation phases: what to implement, documentation references, verification checklist, anti-pattern guards.
• Final phase: verify against documentation, grep for known bad patterns, run tests.

What make-plan needs, and safety notes

The skill's listing states that it requires nothing beyond itself. It does not run scripts and does not use the network. The package contains only SKILL.md.

Because it runs no scripts and makes no network calls, the skill itself does not change files or fetch anything. Its output is a plan document and a chat summary. The skill does not execute the plan. Anything the agent reads or writes depends on the tools your agent environment already allows.

Review the plan before you run any follow-up work, since the plan is only as accurate as the sources the agent consulted.

• Requires: nothing listed
• Runs scripts: no
• Uses network: no
• Output: a plan file and a short summary, not executed code

Tips for getting useful plans

Give the agent access to the docs, examples, or codebase you want it to cite. The skill's plans depend on sources, so a request with no reference material will produce weaker grounding.

Use make-plan for work that spans several sessions. Each phase is written so a fresh chat can pick it up without the earlier conversation.

If your work starts from a bug or feature backlog rather than a fresh idea, the source points to a sibling skill, oh-my-issues, which clusters issues by root cause into plan documents. Make-plan then works on one plan slice at a time.

• Name the docs or example files you want used.
• Keep the plan file; the chat summary is only a short overview.
• Skip make-plan for one-line fixes and quick questions, since the process is heavy.