astropham1808

Astro Agent Skills — Git skill for Claude Code

Git community

Codex skills and Claude Code plugins for verifiable agent-assisted delivery.

How to install Astro Agent Skills

This entry records only its repository, not the path inside it, so there is no exact command to give. Open astropham1808/astro-agent-skills and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Astro Agent Skills does

Codex skills and Claude Code plugins for verifiable agent-assisted delivery. Plan the repo, run one story through an implementer plus an independent read-only reviewer in an isolated Git worktree, gate on a project-owned verifier, and keep PR and merge human. Includes agent-toast notifications for WSL, macOS, and Linux.

Alternatives in Git

  • Git Worktree Setup — Create isolated git worktrees with instant feedback and background Rust verification 25.5k ★
  • Git Status — Read the Files and run the Commands and summarize the current state of the git repository 3.4k ★
  • Codexia — Agent Workstation for Codex CLI + Claude Code — with task scheduler, git worktree & remote control, skills man 638 ★

README

astro-agent-skills

Agent skills that make one story go from a request to a reviewed branch without anyone pretending the work is done. Claude Code implements, Codex CLI reviews in a read-only sandbox, an executable verifier decides what "done" means, and a human owns the pull request and the merge.

Ships as Codex skills and as a Claude Code plugin marketplace. No stack lock-in, no provider lock-in, no hidden framework assumptions.

Why this exists

Agent-assisted delivery breaks in the same few places every time.

The agent says the work is complete when nothing ran. The reviewer is the same session that wrote the code, so it agrees with itself. A refactor wanders into files nobody asked about. The agent pushes, opens a PR, or merges while you are still reading the diff. Two agents edit the same working tree and one of them loses.

These skills close each of those holes with a mechanism rather than a suggestion: verification is an exit code, review runs in a separate read-only session, each story gets its own Git worktree, and every external or destructive action stops at a human gate.

What it gives you

Four skills covering the story lifecycle, plus a notification plugin.

Category Skill What it owns
Planning and setup project-setup Classify the product, plan the repo, scaffold the verification contract.
Foundation disciplined-coding Scoped diffs, stated assumptions, verifiable success criteria.
Delivery orchestration start-story-multi-agent, codex-multi-agents-flow One story: implement, verify, independent review, one bounded fix pass.
Repository lifecycle close-story, close-story-worktree Tear down the worktree and branch after the merge, never before.
Platform integration agent-toast Optional and separate. Desktop notification when Claude finishes. Hooks only, no skills.

Every Codex skill has a Claude Code counterpart. The two delivery-orchestration skills stay separate because they automa