Start coding with idapt Code
Found this helpful? Share it:
Found this helpful? Share it:
idapt Code is a curated skill in normal Chat. You name a computer, a project, and an outcome. The agent reads the code, edits it, runs the project's own checks, and reports what changed. There is no separate coding app, no repository picker, and no required Git integration.
You need three things:
A computer that shows Online in Computers. Pair your own machine or create a cloud computer.
A project on that computer, or a clone URL the agent can fetch.
Shell and File access showing Allowed in that computer's Permissions list, on its Dashboard.
A computer you paired yourself starts with both of those blocked. Enable exactly the two the workflow needs, on the machine itself:
idapt-computer service policy enable remote-shell
idapt-computer service policy enable remote-filesEach command prints Applied., and a running daemon picks the change up
without a restart. Add remote-terminal only if you also want the interactive
Terminal tab. Cloud computers arrive with their managed policy
and need no such step.
Shell and file access let the agent read and change anything available to the operating-system user its commands run as, including credentials, SSH keys, and unrelated projects. Prefer a development account, a virtual machine, or a cloud computer over a machine holding unrelated sensitive data.
Open Skills, choose the curated idapt Code skill, and select Use in new chat.
In one message, name the computer, the project path, and the outcome you want.
Read the commands as they run, and answer the questions the skill asks before it changes anything.
Ask for the diff and the test output before you accept the work.
A first message that needs no follow-up questions looks like this:
On my workstation computer, in ~/work/checkout-api:
the payment-retry test fails. Find why, fix it, run the
package's tests, and show me the diff.The skill opens by confirming the computer, the operating-system user its
commands run as, the project and its Git state, and any AGENTS.md or
contributor instructions the repository ships. It treats changes already in
your worktree as yours: it will not reset, stash, or check out over them.
The Code with AI button on Computers and the Start coding button on the idapt Code page open the same chat with the skill already attached.
The skill stops at the outcome you asked for. Name the last step you want.
| Ask for | What you get |
|---|---|
| "Show me the diff and the test results" | Nothing is committed. You review first. |
| "Commit these changes" | A local commit on the current or a new branch. |
| "Push the branch" | A push through the remote and credentials already on the computer. |
| "Open a pull request" | A request opened with an installed gh or glab, with the URL reported back. |
If the provider CLI is missing, the agent still pushes the branch and gives you the URL to open the request by hand.
For a token that should not live on the computer, read Git credentials and Secrets.
Use a paired computer when the checkout, toolchain, SDKs, package caches, and Git configuration already exist on hardware you control.
Use a cloud computer when the task should not touch your everyday workstation: an unfamiliar repository, a long dependency install, or work that should keep running after you close the tab.
The workflow is identical on both. Setup lives in Computers and Cloud computers.
Ask the agent to start the project's development server and expose its port. The computer returns a temporary HTTPS URL in the access mode you choose, so a reviewer clicks a link instead of cloning a branch. Close the tunnel when the review ends. See Tunnels and public ports.
VS Code Server is optional and never required. The agent reads and edits through the computer's shell and filesystem; a browser editor is only for you.
What you see. The agent names the capability it needs, quotes the remediation the computer returned, and stops without retrying.
Why it happens. Pairing and authorization are separate. A paired computer refuses sensitive capabilities until you enable them on the machine, and the skill never changes daemon policy on your behalf.
What to do. Open the computer's Dashboard and
read the Permissions list. In the idapt desktop app, viewing
the machine it runs on, click Manage to open the native
permissions window. Otherwise run
idapt-computer service policy enable <capability> on the host. Enable only
the capability the agent named.
How to verify. That capability flips from Blocked to Allowed on the Dashboard. Ask the agent to retry the same step.
What not to do. Do not reach for full-control to unblock a
coding task: it also grants admin operations and desktop control, which this
workflow never uses. Full steps and recovery live in
Daemon troubleshooting.
Related articles
Computers
Pair your own machine or launch a cloud one, then let your AI run commands and work with files on it.
Cloud computers
Launch an on-demand cloud machine, understand what it costs per second, and hibernate it when you're done.
Git credentials and Secrets for idapt Code
Let idapt Code push and open pull requests with the computer's existing Git setup, or hand it one workspace Secret for a single command.
Use idapt Code safely
Pick the right machine, grant the two capabilities coding needs instead of full control, and know what an agent command can still reach.
Tunnels and public ports
Expose a port on one of your computers at an authenticated public URL, then close it again.
Skills
Reusable instruction bundles you invoke with a slash, attach to agents, and install from the Hub.
Was this helpful?