Orcaops

Git versions your code. We version your intent.

As your coding agent works, Orcaops captures the plan, decisions, progress, checks, and open questions in a local, versioned task record — without replacing your agent, planning tools, or development workflow.

Documentation: https://docs.orcaops.ai/ · Website: https://orcaops.ai

What it does

• Keep your workflow. Continue using the coding agent and planning method you already have. Orcaops installs skills into supported agents and captures the task as the work happens, with no separate reporting step.
• Watch the work, not just the answer. orcaops watch opens a terminal UI over the captured plan, progress, decisions, findings, and Task Review as they develop. Leave comments directly on the diff for your agent to read, answer, or address.
• Don't take "done" on faith. Evaluators compare the finished code and recorded evidence with the task the agent captured, surfacing missing work, scope drift, unsupported claims, and unresolved questions while they can still be addressed.
• Make every task compound. Each completed task leaves behind decisions, constraints, rejected approaches, and evidence that future agents can search and build on instead of rediscovering it — or undoing deliberate choices.

How it works

Describe a task to your agent the way you normally would. The installed Orcaops skills run the capture lifecycle underneath:

plan → checkpoint(s) → finish

• Plan records the intended outcome, boundaries, decisions, and acceptance criteria before implementation. If the work changes the plan, the agent revises the same record rather than starting over.
• Checkpoints bracket each coherent unit of work. Orcaops attributes changed lines by diffing the worktree between open and close, so provenance comes from real diffs, not from the agent's account of them.
• Finish runs the final evaluator pass, pauses on warnings that need a human, records the summary, and renders the reviewer-facing digest.

Together these form a task record: the plan and its revisions, checkpoints, decisions, verification, evaluator results, uncertainty, and the final outcome. The CLI calls one captured task thread an artifact, and a branch can hold several.

When the branch is ready, ask your agent to generate a Task Review. A forensic lane reviews the diff without seeing the agent's account, an account lane organizes the captured intent without inspecting code, and Orcaops validates and merges both into one review you read in orcaops watch. It is review material for a human, not a merge verdict.

Getting started

You need Node.js 22.14 or newer on macOS or Linux (Windows via WSL2), and a git repository with at least one commit. orcaops watch, the Task Review terminal UI, installs with the CLI as a compiled companion for your platform; nothing else is required.

npm i -g @orcaops/cli
orcaops --version

# from the repository root
orcaops init

orcaops.ai is the project site and where accounts live. npm is the only place the CLI is published; the site points at the command above rather than offering a second way in.

init detects your coding agents and installs the matching Orcaops skills. The default personal setup changes nothing that git tracks: skills go to each agent's global location and .orcaops/ is hidden through git's local exclude, so teammates see no difference.

Then open your agent in the repository and describe a task normally:

Add pagination to the activity feed and update its tests.

When you want to review the branch, ask for the review and open Watch:

Generate a Task Review for this branch.

orcaops watch

Supported agents: Claude Code, Codex, Cursor, OpenCode, AiderDesk, GitHub Copilot, and Antigravity CLI.

The Getting started guide walks through the full setup, including exactly what init changes and how to undo it.

Local by default

Capture, evaluators, search, provenance, and orcaops watch all operate on local data. Network access starts only when you log in for an optional Cloud workflow, or when a tool you configured — such as an LLM-backed evaluator — calls its own provider. Evaluator packs are opt-in; orcaops init installs none. See Local data and privacy.

Teams and Cloud

orcaops update --scope project materializes shared skills, instructions, and evaluator configuration for the whole team to review and commit. Orcaops Cloud adds cross-repository history, web plan approval, and pull-request review on top of the local workflow; connect with orcaops login. See Adopt as a team and Cloud collaboration.

Documentation

The hosted site is at https://docs.orcaops.ai/. The same pages live in this repository under apps/docs/content/.

Start locally

• Getting started
• Import git history
• Local data and privacy

Work through your agent

• Working with your agent
• Skills
• Task Review and Watch
• Evaluators

Teams and Cloud

• Adopt as a team
• Cloud collaboration
• Authentication
• Plan review

Reference and extensions

• Configuration
• Agent integrations
• Session hooks
• Capture and data
• Troubleshooting
• Command reference
• Glossary
• Authoring evaluator packs
• Task Review protocol

Security

See SECURITY.md for the trust model — in particular, what a checked-in repository configuration is and is not allowed to authorize, and how consent for evaluator packs works.

Development

This is a pnpm workspace:

pnpm install
pnpm build
pnpm test

pnpm format:check, pnpm lint, pnpm typecheck, and pnpm typecheck:tests mirror CI. See TESTING.md for the test taxonomy — where a new test belongs and how to keep it discovered and typechecked — and docs/dependency-policy.md before adding a dependency.