Skillget

agentplane

agent
Guvenlik Denetimi
Gecti
Health Gecti
  • License — License: MIT
  • Description — Repository has a description
  • Active repo — Last push 0 days ago
  • Community trust — 39 GitHub stars
Code Gecti
  • Code scan — Scanned 12 files during light audit, no dangerous patterns found
Permissions Gecti
  • Permissions — No dangerous permissions requested
Purpose
This CLI framework provides a governed, auditable workflow layer for coding agents like Claude Code, Codex, and Cursor. It enforces a strict task, plan, and verification process locally within your Git repository to ensure all automated changes are trackable and transparent.

Security Assessment
Overall risk: Low. The tool operates entirely as a local CLI without requiring a hosted runtime, meaning it keeps your project's state inside your local Git repository. The automated code scan of 12 files found no dangerous patterns, and the tool requests no dangerous permissions. There are no hardcoded secrets in the scanned code. It does not appear to make unauthorized external network requests or execute hidden shell commands outside of standard local Git operations.

Quality Assessment
The project demonstrates strong quality and active maintenance. It was updated today (0 days ago), uses the permissive MIT license, and requires Node.js 20+. It features a clear, professional README with visible CI testing and coverage pipelines. With 39 GitHub stars, the project is relatively new and small but shows a healthy foundation and structured documentation for developers.

Verdict
Safe to use.
SUMMARY

Governed CLI framework for deterministic, auditable AI agent workflows

README.md

Agent Plane Header

AgentPlane

npm
Downloads
License: MIT
Node.js 20+
CLI Contract
Core CI
test:fast
coverage
release:parity
knip

Use coding agents without losing Git discipline.

AgentPlane is a local CLI that makes Claude Code, Codex, Cursor, Aider, and similar coding-agent
work auditable inside your Git repository:

task -> plan -> approve -> implement -> verify -> finish

No hosted runtime. No hidden state. Everything stays in your repo.

Install

npm i -g agentplane
agentplane init
agentplane quickstart

Requirements:

  • Node.js 20+
  • Git repository
  • Local terminal

Why it exists

Coding agents can change files. Teams still need to know what happened:

  • What task was the agent solving?
  • What plan was approved?
  • What changed in the repository?
  • What was verified?
  • Why was the task considered finished?

AgentPlane adds a visible workflow layer around agent work without replacing Git, your editor, or
your terminal.

What appears in your repository

AGENTS.md or CLAUDE.md   Policy gateway for the repository
.agentplane/            Repo-local workflow workspace
.agentplane/config.json Current workflow configuration
.agentplane/tasks/      Per-task records
.agentplane/WORKFLOW.md Materialized workflow contract
verification records    Stored in task docs
finish record           Linked to the Git revision

These artifacts make agent work inspectable. A reviewer can see what policy governed the repo, what
task was active, what plan was accepted, what checks ran, and how the task was closed.

How it works

agentplane task new --title "First task" --description "Describe the change" --priority med --owner DOCS --tag docs
agentplane task plan set <task-id> --text "Explain the plan" --updated-by DOCS
agentplane task plan approve <task-id> --by ORCHESTRATOR
agentplane task start-ready <task-id> --author DOCS --body "Start: ..."
# run Claude Code, Codex, Cursor, Aider, or edit manually
agentplane task verify-show <task-id>
agentplane verify <task-id> --ok --by DOCS --note "Checks passed"
agentplane finish <task-id> --author DOCS --body "Verified: checks passed." --result "One-line outcome" --commit <git-rev>

If your repository does not require explicit plan approval, skip task plan approve.

Without vs with AgentPlane

Without AgentPlane With AgentPlane
Prompt in chat Task is recorded
Agent edits files Plan is explicit
Human inspects diff Approval is visible
Context is scattered Verification is stored
Verification is implicit Finish creates closure evidence
Closure is manual Everything lives close to Git

Who it is for

  • Developers using Claude Code, Codex, Cursor, Aider, or local coding agents.
  • Maintainers who want agent changes to remain reviewable.
  • Teams that need task state, verification, and closure before merging agent-generated work.
  • Local-first builders who do not want a hosted agent runtime between their repo and their workflow.

What it is not

  • Not a hosted agent platform.
  • Not a prompt framework.
  • Not a replacement for Git.
  • Not a replacement for your editor.
  • Not a replacement for Claude Code, Codex, Cursor, or Aider.

Workflow modes

direct

Fast local loops in the current checkout. Good for solo work, prototypes, and short tasks.

branch_pr

Structured per-task branch and PR-style handoff. Good for teams, stricter review, and integration
boundaries.

Recipes

Start from the recipe that matches your current stack:

Each recipe includes when to use it, commands, expected repo artifacts, limitations, and a
copy-paste flow.

Documentation

Start here:

Developer surfaces:

Deep architecture terms such as framework control plane, harness contract, protocol surfaces, and
behavior precedence belong in the developer docs, not in the first-start path.

Technical proof

  • MIT licensed.
  • TypeScript codebase.
  • Local-first CLI.
  • Active release history.
  • No hosted control plane.

Contributing

Contributions are welcome. See CONTRIBUTING.md.

License

MIT

Yorumlar (0)

Sonuc bulunamadi