One workflow engine for
every CLI coding agent
Send a coding agent one prompt, or chain prompts and scripts into a workflow that checks its own work. Every session stays searchable in one VS Code dashboard.
How it works
- 1
Prompt
Reference files, symbols, selections, and Git changes right in the editor — with completions, validation, and token estimates.
- 2
Steer
Watch and drive the native terminal directly, or let a structured workflow carry the step forward for you.
- 3
Automate
Chain prompts and scripts into a validated workflow. Retry on real output, mix harnesses, run several at once.
- 4
Revisit
Every run lands in a searchable, diffable history you can replay, export, or run again.
One dashboard, six harnesses
Each CLI coding agent you install and run through ctrlyoke is a harness. ctrlyoke calls it for you, so you can change harnesses, not habits.






Your CLI and your account. ctrlyoke just calls it.
ctrlyoke doesn't replace Claude Code, Codex, or Copilot CLI. It launches the copy you installed, signed in the way you signed it in.
You install it and sign in to it
ctrlyoke never signs in for you, and never reads or stores your harness login.
ctrlyoke remembers the flags
Each CLI's invocation, session resume, permission modes, and MCP and hook config are handled for you.
Your plan and its terms still apply
ctrlyoke adds an editor and a workflow engine around the agent. The agent itself, and what you pay for it, stay the same.
Build visually, or edit as YAML
Drag steps into parallel groups and pick a harness and model for each one, or open the same workflow as a file and type it with completions. Either way, it's plain YAML or JSON that lives in your repo.
A prompt language for your whole project
Reference files, symbols, commits, earlier steps, secrets, skills, and MCP servers right in the prompt. Completions find each one as you type, and ctrlyoke turns them into real context before the harness sees the prompt.
@src/routes/Files and folders
#sym:normalizeTagSymbols and line selections
~0f2072aCommits, staged and working-tree diffs
$previousOutput from earlier steps
!API_TOKENSecrets, passed as environment variables, never written into the prompt
/skill:commitSkills and MCP servers, per step
Plus spell check, image paste, token estimates that include the attached context, and a warning before you run a reference that won't resolve. See the full reference syntax.
Every session, searchable
ctrlyoke keeps every run on every harness: the prompts, the answers, the commands, and the diffs. Search all of it at once, reopen any run, and continue its session or run it again.



Every workflow you've run, with its harness, tokens, cost, and run count. Star the ones you come back to and they stay on top.
Search prompts, answers, commands, and diffs across every run and every harness. Here, how five agents handled the same rate-limiting task.
Open a result to land in that run with every match highlighted. Resume its session, or run the workflow again.
Every result, with undo
Whichever harness ran the step, its file edits show up the same way: a diff for each one, and undo for a single edit, a single file, or everything a workflow changed.


Every file edit in the output has its own diff and Undo button, whichever harness made it. Here, Antigravity's fix during review.
See every file a workflow changed in one place, with the steps that touched each one, and undo just the files you pick.
Your project, in any browser
The browser dashboard comes with a workspace viewer: a code editor with language servers, search, diffs, and Git. Check what an agent did and commit it without a VS Code window.


Browse and edit files with syntax highlighting and a real language server: Go to Definition, Find All References, Rename Symbol, and Format Document.
Review what the agent changed file by file, stage it, and commit, with the branch history alongside.
The same tools, whichever harness runs
The terminal, validation and your agents, skills and MCP servers work the same way on all six harnesses.
Native terminal, structured control
Watch and steer the real interactive CLI in an embedded terminal while ctrlyoke builds a searchable record beside it.
Checks that use your own scripts
Pre- and end-conditions gate each step with the scripts your project already trusts — continue, retry, skip, or stop.
Agents, skills and MCP servers, set up once
Manage agents, skills, MCP servers, and hooks once, then reuse them across every harness you run.
Code from your phone
Pro unlocks the remote dashboard. Scan the QR code and take control of a workflow running on your PC, even mid-step, from any browser or phone that can reach it: across your home or office network, or from anywhere over your own VPN. See how to connect from anywhere.
Install ctrlyoke
Free for individuals and companies under $10M revenue and 250 employees. Pro is optional.
No VS Code? ctrlyoke server opens the same dashboard in your browser.