Frink skill
Frink runs up to four agents in their own worktrees and turns one sentence into a Flow automation with its own editor skill.
- 176 ★ on the repository
- Last commit
- MIT
Broad access
Why this level
- Launches Claude Code, Codex, Cursor and other agents on the user's behalf and gives them separate git worktrees on disk
- A Flow runs commands, writes files and opens a pull request on a schedule or webhook without a person present at trigger time
- Connects to Linear, Sentry, PostHog, Vercel and other services as chat tools and triggers, receiving their data
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add skills/frinkDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- Launches Claude Code, Codex, Cursor and other agents on the user's behalf and gives them separate git worktrees on disk
- A Flow runs commands, writes files and opens a pull request on a schedule or webhook without a person present at trigger time
- Connects to Linear, Sentry, PostHog, Vercel and other services as chat tools and triggers, receiving their data
This is third-party code. Review the repository files before installing.
What it does
Frink is a desktop app for macOS and Linux that runs Claude Code, Codex, Cursor, GitHub Copilot or OpenCode through the sign-in you already have, with no Frink account of its own. The window splits into up to four panes, each with its own agent, model and separate git worktree, so parallel tasks never touch the same files. A Flow is a scenario of agent, command, condition and approval steps that starts on a schedule, a webhook, or an event from Linear, Shortcut, ClickUp, Sentry, PostHog or Vercel. You describe the routine in a sentence, Frink builds it as a step graph you can read and edit before the first run. Finished or pending work sits in a shared Work Queue on the computer and in a beta iPhone app. Frink ships its own frink-flows skill for editing a Flow: it teaches the agent the graph contract, node types and the one-Start-Task-per-Flow rule, and wires in through dedicated MCP tools such as frink_flows_get, frink_flows_patch and frink_flows_list_catalog instead of hand-editing the YAML on disk. Every file change the agent makes shows as a line diff in the built-in editor, with a terminal and the option to roll the chat and code back to an earlier point.
Who it is for. Developers who run several agents on one machine and want recurring checks and test fixes to become an automated scenario that waits for approval before a risky step.
Good fit when
- Four independent tasks in one project need to run in parallel, each in its own git worktree, without waiting in a queue
- A routine like running tests on weekdays at 9am, fixing failures and pausing before a PR needs to be described in a sentence, not code
- A Linear ticket or a Sentry error should start an agent scenario by itself instead of waiting for a developer to open a terminal
- Agent changes need a line-by-line diff and a way to roll a chat back together with its code to an earlier point
- Finished work and agent questions need one queue visible from a phone instead of scattered terminal tabs
Not a fit when
- A team subscription with shared billing across members is needed: Frink always runs through each person's own sign-in on their own computer
- The team's main agent is a Windows-only tool: Frink ships builds only for macOS and Linux
- Flows, MCP servers and hooks are needed inside OpenCode: these do not work yet in OpenCode's preview mode
- The task is one short terminal command: spinning up a graph and a worktree for it is overkill
Example request
Build a Flow: on weekdays at 9am run the tests, and if they fail have an agent fix them and wait for my approval before opening a PRLimitations
- Frink never reads or stores an agent token: it only checks that a `claude` or `codex` sign-in already exists and launches the same program the terminal would use.
- OpenCode in Frink is marked preview: chats work, but Flows, building a Flow from a chat, MCP servers and hooks are not available yet, and OpenCode's own slash commands are not supported.
- MCP credentials and API keys are saved only when a desktop keyring (GNOME Keyring or KWallet on Linux) is available; without one Frink does not save them.
- Official builds send anonymous usage counts and crash reports; the full list of what is sent is documented in the telemetry guide.
How to disable. Quit the app and remove it like any other macOS or Linux program (drag the .app to the trash, delete the AppImage, or run sudo apt remove frink for a .deb install). Frink leaves no background daemon, and your Claude Code or Codex sign-in stays untouched.
FAQ
Does Frink need its own account to run agents?
No, no Frink account is needed. The app uses the sign-in you already have for Claude Code, Codex, OpenCode, GitHub Copilot or Cursor and never reads or stores the agent's token itself.
How do four agents work on one project without conflicts?
Each of up to four panes runs in its own git worktree, so agents never touch the same files at once, and each pane's diff and result are reviewed separately.
Can a Flow be edited by hand in YAML?
No, Frink explicitly says not to edit the Flow file on disk: changes go through the frink-flows skill and its MCP tools, such as frink_flows_patch, or a patch can break the graph's topology.
Does Frink work with OpenCode the same way it works with Claude Code?
Not yet fully. OpenCode chats work, but Flows, building a Flow from a chat, MCP servers and hooks are marked preview and not supported yet.
What does the Work Queue do?
The Work Queue collects every run from every project in one place: what is running, what is ready for review and what is waiting for approval, with a badge when an agent has a question or a Flow pauses, reachable from an iPhone in beta.
Discussion
No questions or reviews yet.
Related
SkillsDeclarative DAG workflows for coding agents: isolated phases, graph verification, resume and replay of runs.
Flowtrace turns an agent's work into steps stored as files in a git repository, so you can follow, check, fix one by one and re-run them on new input
Skill Flow groups skills from many sources and deploys them to many agents, including Claude Code, Cursor and Codex.
Maestro-Flow builds a chain from analysis to tests for a goal and runs Claude Code, Codex, Gemini and other agents through it
A multi-agent workflow engine in Go: one YAML file describes agents, steps and dependencies, run with one command.
A visual workflow builder for agents: draw a flow on a canvas and export it as skills and commands for Claude Code, Codex, Cursor and others
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Repository stars | Main difference |
|---|---|---|---|
| Frink (this card) | yes | 176 | |
| CLI Agent Orchestrator | no | 1,548 | a Flow graph with triggers from Linear and Sentry plus a shared review queue, not just running several CLIs in parallel |
| Awesome Agent Orchestrators | no | 2,413 | a ready desktop app you install on your computer, not a list of links to orchestrators |
README in short
The front page covers four capabilities: parallel agents in their own worktrees, building a Flow from a single sentence, starting a Flow from a ticket, error or webhook, and one shared queue for finished work. It separately lists the supported agents (Claude Code, Codex, OpenCode, GitHub Copilot, Cursor), install instructions for macOS and Linux with AppImage and .deb builds, and links to user docs on sessions, Flows, plugins and MCP.