Skip to content

Bleu #1857

@Nirvaan05

Description

@Nirvaan05

Display Name

Bleu

Category

Agent Skills

Sub-Category

None

Primary Link

https://github.com/Nirvaan05/Bleu-plugin

Author Name

Nirvaan

Author Link

https://github.com/Nirvaan05

License

MIT

Other License

No response

Description

Planning plugin for Claude Code. Turns an idea into a file-backed markdown wiki (vision, architecture, ADRs, action points) that survives /clear and hands off to GSD, Superpowers, or raw Claude Code.

Validate Claims

Install the plugin, ask Claude to blueprint a small system, and inspect the generated blueprint/ directory: you should see plan/ docs, action-points/ with a dependency graph, decisions/ with MADR-style ADRs, and SESSION.md/NEXT.md. Then run /clear and ask it to resume. It should re-orient itself from those files alone, in roughly 2k tokens, without you re-explaining anything. The on-disk structure and the clean resume after /clear are the whole point, and both are verifiable in a single session.

Specific Task(s)

Ask Claude to plan a URL shortener service with custom slugs, rate limiting, and click analytics. It is small enough to finish in one session, but has enough moving parts to exercise component breakdown, action points with dependencies, and a couple of architectural decisions.

Specific Prompt(s)

First prompt: use bleu to blueprint a URL shortener with custom slugs, rate limiting, and click analytics

Once the blueprint/ workspace exists, run: /clear

Then in the fresh session, say: where did we leave off?

Claude should read SESSION.md, NEXT.md, and the journal tail, then state the current phase and the next action without re-reading the whole plan.

Additional Comments

Plan-only by design: it writes no application code, only markdown under blueprint/, so everything it produces is auditable on disk. No network requests beyond your own Claude Code web tooling. MIT licensed, v1.0.0 released. Not meant for small tweaks; it targets substantial systems planned across multiple sessions.

These lean into the two claims a reviewer can verify fastest (on-disk structure + resume-after-/clear), pick a task small enough to satisfy their "one session" bar, and the comments preempt their security questions (no network, plan-only, auditable). The Validate/Task/Prompt fields address the reviewer directly because that matches their own examples - only the Description field had the "do not address the reader" rule.

Recommendation Checklist

  • I have checked that this resource hasn't already been submitted
  • It has been over one week since the first public commit to the repo I am recommending
  • All provided links are working and publicly accessible
  • I do NOT have any other open issues in this repository
  • I am primarily composed of human-y stuff and not electrical circuits

Metadata

Metadata

Assignees

No one assigned

    Labels

    resource-submissionThis Issue submits a new resource to the listvalidation-passedResource has passed initial validation

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions