AgentPlane records what an AI agent was asked to do, which plan was approved, what changed, which checks ran, and why the work is safe to merge or revert - all as reviewable artifacts inside your Git repository.
task → plan → approve → implement → verify → finish
AI agents can produce real code changes faster than teams can reconstruct why those changes happened. A pull request shows what changed. It rarely shows what the agent was asked to do, which plan was approved, which checks actually ran, or why the work is safe to merge.
Quickstart shows the repo-visible artifact shape before you let an agent touch product code. Reviewers do not need to trust an agent transcript; they can inspect files, commits, checks, hashes, and the ACR.
$agentplane task new --title "Demo task" --owner <agent-id> --tag docs$agentplane task plan set <task-id> --text "Inspect the artifact."$agentplane task start-ready <task-id> --author <agent-id> --body "Start: inspect artifacts."$agentplane task verify-show <task-id>$ls .agentplane/tasks/<task-id>/Recipes are optional signed packages for teams that want reusable agent profiles, prompt modules, skills, or repository mapping assets.
Start with the task -> plan -> verify -> ACR flow first; add recipes only when you need reusable behavior.
Claude Code, Codex, Cursor, Aider, Git, CI, and PR review keep their jobs. AgentPlane records the missing evidence around their work.
They generate or edit code; AgentPlane records the review trail around the work.
Policy text tells agents what to do; AgentPlane turns policy into lifecycle artifacts and gates.
Git stores the final diff; AgentPlane stores task intent, plan, verification, and closure.
They move workflow state elsewhere; AgentPlane keeps the durable record in the repository.
Use any coding agent. Keep the evidence in Git.
Record the scope before the agent edits files.
Keep the intended route visible before implementation.
Use Claude Code, Codex, Cursor, Aider, or another coding agent.
Attach checks, notes, CI results, findings, commit metadata, and ACR.
AgentPlane does not replace your agent, editor, terminal, or Git.
.
|-- AGENTS.md
`-- .agentplane/
|-- WORKFLOW.md
`-- tasks/<id>/
|-- README.md
|-- pr/
`-- acr.jsonAgentPlane creates ordinary repository artifacts that can be inspected, reviewed, and committed with the rest of your project.
Every meaningful agent change should leave behind one machine-readable record: task intent, approved or waived plan, agent and toolchain identity, Git base and work commits, policy decisions, verification results, and merge readiness.
$agentplane acr generate <task-id> --work-commit HEAD --write$agentplane acr validate <task-id> --mode local$agentplane acr check <task-id> --mode ci$agentplane acr explain <task-id>Humans read `acr explain`. CI reads `acr check`. External tools read `acr.json`.
Read the ACR specAgents are no longer just drafting snippets. They edit real repos, open branches, run checks, change docs, touch release paths, and leave reviewers with incomplete context. The next failure mode is not whether the agent can write code; it is whether the team can reconstruct, verify, and govern what happened.
Start with the 90-second local task and see the first evidence artifact before the abstractions.
Open the workflow guide for your agent and keep the work reviewable after the session is gone.
Add task evidence and ACR validation before merge.
See how AgentPlane differs from coding agents, AGENTS.md, Git-only workflows, CI, and hosted dashboards.
Read the recipe model after you understand the core evidence trail.