Planning copilot for coding agents. Blueprint asks the right questions before you write code, then hands your agent a plan it can execute in one shot.

Agent-agnostic skills compatible with skills.sh. Works with Claude Code, Codex CLI, Gemini CLI, Pi agent, and other compatible harnesses.

Most coding agents rush to code or guess at the plan. Blueprint slows down just enough to ask the right questions. It reads your codebase and asks multiple-choice questions you can answer easily. The output is a markdown plan any coding agent can execute.

You stay in control without doing the tedious work. Blueprint splits planning between you and the agent. The initial idea comes from you. Enumerating all the considerations and choices comes from the agent. Decisions come from you. You skip the tedious parts of planning but stay in the driver's seat.

Questions that make you think. Blueprint asks questions that surface real design choices — the kind that engage you and make you think about what you actually want. It surfaces things you wouldn't have thought to ask about.

Easy to answer. Each round is a small chunk of work. Questions start broad and get more specific as the plan takes shape.

Scales to however much planning you want. You repeat this small loop until you decide to stop. If you don't want to plan much, you have a quick out — generate the plan after one round. If you want to plan extensively, the agent helps you keep exploring while keeping track of everything. This keeps you in control in two ways: you make the decisions, and you decide how much to plan.

"Catches things that I didn't think to think about."

"A way to get ideas out of your mind and into a spec."

"I'm never going back to not using it again."

npx skills add imbue-ai/blueprint

In your agent, invoke the skill with a short description of the task.

/blueprint Add a caching layer to reduce API calls

Blueprint asks you to pick a template, explores your codebase, and starts asking questions. Answer what matters. Skip what you don't care about. When you have covered enough ground, generate the plan.

/blueprint-generate

The plan is written to blueprint/<slug>/plan-<slug>.md. From there, chat to refine, ask "what are the open questions?" to surface gaps, or hand the file to your coding agent.

1. Run blueprint <description>. Pick a template. The agent explores your codebase and asks the first round of questions.
2. Answer questions. Follow-ups come naturally based on your answers.
3. Run blueprint-generate. The plan lands at blueprint/<slug>/plan-<slug>.md.
4. Refine in chat. Ask "what are the open questions?" to surface what is still ambiguous.
5. Continue refining for as many rounds as you want.
6. Hand the plan to your coding agent.

Two built-in templates ship by default.

• Default. Overview, Expected behavior, Implementation plan, Implementation phases, Testing strategy, Open questions.
• Concise. Overview, Expected behavior, Changes.

You can also describe a custom template inline when prompted.

Edit templates.json in both blueprint/references/ and blueprint-generate/references/ so the two skills stay in sync. Each entry has three fields.

• name. Short label shown during template selection.
• description. One-line summary shown next to the name.
• prompt. String or array of strings describing the plan structure the agent should follow.

Example:

{
  "name": "feature",
  "description": "Plan for a user-facing feature",
  "prompt": [
    "The plan should contain the following sections in order.",
    "",
    "- Overview: motivation for the feature and what users will be able to do",
    "- User experience: walkthrough of the primary flow and key edge cases",
    "- Implementation: files, modules, and data types to add or change, and what each does",
    "- Testing: how to verify the feature, including unit tests, integration tests, and edge cases",
    "- Open questions: unresolved design decisions or trade-offs"
  ]
}

Best fit. Greenfield projects. Large new features on existing codebases. Incremental changes big enough to warrant a plan. Research experiments. New models, systems, or subsystems.

Less ideal. Frontends where most decisions are visual. Small refactors. Debug-polish work.

• A compatible agent harness (Claude Code, Codex CLI, Gemini CLI, Pi agent, etc.)
• A workspace the agent can read
• npx to run the install command

Claude Code plan mode. Optimized to unblock the agent. Questions are brief. Blueprint asks questions to understand you, not to unblock itself.

Spec-kit, open-spec, other spec generators. The agent makes its own choices, then asks you to review a long spec. Blueprint reverses that order. Your input comes first.

Prefer a sidebar? Blueprint ships as a VS Code extension that works in VS Code, Cursor, and Windsurf. The extension is open source at imbue-ai/blueprint-vscode.

Follow along with what we are building.

• @Imbue_AI on X
• Subscribe to our newsletter
• Read the blog

• Introducing Blueprint

• Nayana Bannur
• Nikos Plugachev