Blueprint Plan Generator (blueprint-generate)
Ends the Q&A phase and turns the context your blueprint session gathered into an executable implementation plan your coding agent can run in one shot.
Evidence shows a minimal-privilege skill: it only creates blueprint/<slug> and writes markdown there, explicitly forbids modifying other files before user confirmation, with no network calls or sensitive-data handling. Deducted for: no rollback guidance, unverified publisher attribution, and undefined behavior when the template name cannot be matched.
The six steps are self-consistent, all referenced files exist, slug-collision handling and progress indicators are specified, and the happy path is plausible. Deducted for: no failure feedback if template lookup fails, thin fallback when Q&A context is missing, no tests or abnormal-input handling; reliability is unverified by execution.
The README clearly scopes best-fit scenarios (greenfield, large features) and non-fit cases (visual frontends, small refactors); trigger semantics are clear (invoke after blueprint Q&A). Deducted for: the skill assumes a preceding blueprint session with underspecified standalone-invocation boundaries; no Chinese-language notes though function is locale-independent.
Docs are well layered (SKILL.md steps plus reference files); README covers install, workflow, template extension, and limitations; MIT license is explicit and contributors are named. Deducted for: no version numbers or changelog, templates. must be manually synced across two skill directories (drift risk), and no declared maintenance/update path.
The core task — synthesizing Q&A into a structured markdown plan — has a clear, directly handable output format for coding agents. Deducted for: static review cannot verify output quality; the value claim rests on non-auditable README testimonials; no tested comparison against alternatives.
The skill files are auditable primary material with concrete, inspectable templates and flows. Deducted for: no CI or committed tests covering key paths, unverifiable user quotes, and no independent reproduction under static review.
- This is a static source review only; nothing was executed and confidence is low.
- No failure feedback is defined when the recalled template name does not match templates., risking silent misformatted plans.
- templates. must be manually synced across the blueprint and blueprint-generate directories, creating drift risk.
- The skill presumes a preceding blueprint Q&A session; standalone invocation depends on the agent inferring context.
- README testimonials are marketing claims and do not constitute verifiable effectiveness evidence.
What does this skill do, and when should you use it?
blueprint-generate is one of two skills bundled in the imbue-ai/blueprint repository, serving as the closing step of the Blueprint planning copilot. It is invoked after the blueprint Q&A skill has gathered enough context, resolving the chosen template, refined prompt, and feature name from the conversation. It creates a uniquely named plan directory and writes a markdown plan to blueprint/<slug>/plan-<slug>.md. After writing, it supports further in-chat refinement rounds. It is a pure instruction-based skill that needs only shell and filesystem access, with no Claude Code-specific extensions.
The skill runs six steps: recalls the template name from the Q&A conversation and resolves it via references/templates.; generates a 2-5 word kebab-case slug (max 50 chars, appending -2, -3 on collisions); creates the blueprint/<slug> directory with mkdir -p; adopts the most recent refined prompt from the Q&A (generating one per references/refine-prompt.md if missing); writes the plan per references/write-plan.md while appending a ✓ Explore ✓ Plan ● Write ○ Refine progress line to every Write-phase message; then enters refinement per references/refinement.md.
- A developer already running the blueprint Q&A session who wants to immediately produce an implementation plan once enough context is gathered
- A team that wants a markdown plan document for a large new feature to hand to a coding agent
- Users of multiple agent tools (Claude Code, Codex CLI, Gemini CLI, etc.) who need a platform-agnostic plan file
- A developer iterating on multiple features who needs uniquely named plan files that never overwrite older ones
- A user who prefers writing the plan first, then asking "what are the open questions?" and refining over several rounds
What are this skill's strengths and limitations?
- Outputs a plain markdown plan file that any coding agent can consume
- Works with the blueprint Q&A skill so plans reflect your own design decisions rather than agent guesses
- Auto-generates a unique slug, preventing overwrites of existing plan directories
- Pure instruction-based skill requiring only shell and filesystem access, no extra dependencies
- Includes a refinement phase and progress indicator for tracking Explore/Plan/Write/Refine status
- Cannot be used standalone; it requires a prior blueprint Q&A session
- Template customization requires editing templates. in both skill directories to stay in sync, which is easy to forget
- The repository provides no test suite specific to this skill
- The README does not report verification on harnesses beyond Claude Code, Codex CLI, and Gemini CLI
How do you install this skill?
Install the whole skill collection with npx skills add imbue-ai/blueprint; Claude Code users can alternatively run curl -fsSL https://raw.githubusercontent.com/imbue-ai/blueprint/main/install-skills.sh | bash, which copies the skills into ~/.claude/skills/ (or $CLAUDE_CONFIG_DIR/skills). The skill lives at skills/blueprint-generate/SKILL.md in the repository.
How do you use this skill?
First run /blueprint <task description> to complete the Q&A session. Once you have covered enough ground, run /blueprint-generate to end the Q&A phase and write the plan to blueprint/<slug>/plan-<slug>.md. Then refine in chat, ask "what are the open questions?" to surface gaps, and hand the plan file to your coding agent for execution.
How does this skill compare with similar options?
The README compares Blueprint against Claude Code's plan mode and spec generators like Spec-kit and open-spec: Claude Code plan mode asks brief questions optimized to unblock the agent, whereas Blueprint asks questions to understand you; Spec-kit-style tools have the agent make choices first and ask you to review a long spec, whereas Blueprint puts your input first.