juxt/claude-plugins/a-star

Claude Code and agent skill by juxt/claude-plugins

Trust tier 2
  • community
LLM review pending
Source check pending

Checking that the SKILL.md file is still in the repository.

The a-star process, and its entry point: pick a card up, make it ready, and offer the next dart or drive from its ready sub-goals. Every other a-star skill loads this one first — it carries the loop, the roles, where sub-goals come from, the structure/behaviour split, what a-star records, and what it needs the project to say. Use when the user says "/a-star <card>", "pick up <card>", "start work on <card>", "refine this card", or resumes a card from an earlier session.

Install

npx vskill@latest install juxt/claude-plugins/a-star

SKILL.md

a-star

Interpret MUST, MUST NOT, SHOULD, SHOULD NOT, MAY, etc. per RFC 2119.

  1. Dart to an end-to-end as quickly as possible.
  2. Use what the dart found to land atomic, comprehensible, correct, compliant changes. Go to 1 as required.

Throwing a drive away is always an option: revert it, mark its sub-goal not done, and dart again with what it taught. A drive that fails before it lands steps back per a-star:drive.

Roles

  • You, the main session, are the super agent.

You triage what darts raise, and write everything that lands, including the tests.

  • The dart agent reaches the goal end-to-end and reports. Its code never lands.

A user's question is a question, not a go-ahead. "Is that a good idea?" wants an answer, not the step it asks about.

Sub-goals

  • Sub-goals come from darts, not from reading the code.
  • A sub-goal is one tidy or one advance, landed by a drive.
  • Each commit is one atomic unit of change, its own tests and docs included.

The user reviews a branch commit by commit, understanding and agreeing with each before the next; a commit whose tests or docs arrive later can't be agreed with on its own.

  • Together, a card's sub-goals MUST complete it.
  • The landed code keeps the data structures a dart got right, and leaves out what it built around the ones it didn't.

A structure change and a behaviour change never share a commit: a tidy changes structure, an advance changes behaviour, and a-star:drive lands either.

What to record

In whatever the session tracks its work with:

  • a sub-goal, and what it depends on;
  • a decision — a call you aren't sure of is an open question instead;
  • an open question, and what it blocks;
  • a sub-goal done, with its commits;
  • what's in hand — the sub-goal being worked, and anything that interrupted it;
  • whether the user is watching (a-star:primary, a-star:secondary).

A session starts primary, unless nobody can watch it — headless, cloud or scheduled — when it starts secondary.

Take the next step from the ready sub-goals.

/a-star <card>

  1. Pick the card up per the project's conventions — for example, reading the issue and its neighbourhood, agreeing why it's being done now, then assigning it and marking it in progress.
  1. Make it ready. A card is ready when a dart can start from it alone:
  • its root is agreed — goal, invariants and out of scope, on the tracker issue;
  • no open question bears on the card itself;
  • the project's readiness conventions hold.

Where it isn't, agree the root with the user — what the card is for, never how it gets there. In secondary, record the questions and put the card down instead. Before it ends, update the tracker issue with what was agreed: every decision that bears on the problem or the agreed scope, not only the root.

  1. Move to the next step, per the mode — a dart where the card has no sub-goals yet or part of its goal is still unwalked, else a drive on a ready sub-goal.

Primary offers it and waits for a yes; secondary takes it.

What the project says

Everything not here follows the project's conventions — writing, tracking, planning, picking a card up and putting it down, the base branch, ship/show/ask, the definition of done, coding standards, and how a commit is written. What goes in a commit is a-star's: one atomic unit of change.

The project MUST answer three choices of a-star's own:

  • How a tidy lands — feature branch or main; what to verify before committing; when to push; whether it's done at commit or once CI is green.
  • How an advance lands — what to verify before committing, and whether to push the feature branch as work lands.
  • Branch cleanup — what happens to a branch once its work has landed.

Raise an unanswered one as a question; don't fill it with a default. The exception is branch cleanup: absent an answer, delete local branches and worktrees whose work has fully landed, and only offer to delete remote ones.

Security

  • Security scanPassed, 100 of 100Pattern checks on the skill's files.
  • Model reviewNot runA model review is scheduled for the next rescan.

Quality evaluation

  • Task evaluationNot runNo evaluation yet. Run one in Skill Studio.

Source

Popularity

7-day trend score
0

Not refreshed yet.

Works with

Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.

  • Claude Code
  • Cursor
  • GitHub Copilot
  • Windsurf
  • Codex
  • Gemini CLI
  • Cline
  • Amp
  • Roo Code
  • Goose
  • Continue
  • Kilo Code
  • Junie
  • Trae
  • OpenHands
  • Qwen Code
  • AI Agent
  • Tabnine Chat
  • Sourcegraph Cody
  • Amazon Q
  • Firebase Genkit
  • Sweep
  • Devin
  • Phind
  • Mentat
  • Claude Desktop
  • Aider
  • GPT Pilot
  • Plandex
  • bolt.new
  • v0
  • Replit Agent
  • Anse
  • TypingMind
  • BetterChatGPT
  • LibreChat
  • Warp AI
  • Pieces
  • Fig AI
Embed badge
vskill VERIFIED badge
[![vskill VERIFIED](https://verified-skill.com/api/v1/skills/juxt/claude-plugins/a-star/badge)](https://verified-skill.com/skills/juxt/claude-plugins/a-star)

Something wrong with this skill or its page? Tell us.