Repository navigation
Conversation
Contributor
|
Important Review skippedAuto reviews are disabled on this repository. Please check the settings in the CodeRabbit UI or the
 |
![github-advanced-security[bot]](https://avatars.githubusercontent.com/in/57789?s=60&v=4){kind=small}
elsand
force-pushed
the
feat/agentctl-setup
branch
5 times, most recently
from
October 29, 2026 08:48
9a0f001 to
d65e6db
Compare
elsand
force-pushed
the
feat/agentctl-setup
branch
from
October 29, 2026 09:00
d65e6db to
62e61ca
Compare
elsand
force-pushed
the
feat/agentctl-setup
branch
from
October 29, 2026 12:26
62e61ca to
73e2747
Compare
`agentctl setup` will sign people in to Claude Code, Codex and GitHub without those tools on the host. It runs each sign-in in a sandbox of the base image, relays its terminal, and opens the sign-in page in the host's browser. This adds the library side that it builds on. - `sandbox::start_terminal_execution` starts a terminal Execution that the caller relays itself, so it can scan the output as well as show it. - `harness::sandbox_sign_in` and `harness::github::sign_in` describe each sign-in: the program to run, the file it leaves its credential in, and how to read the credential once it exits. - Claude's sign-in wraps its token at the terminal width, ending a row with a newline or with cursor movements. The token parser reads it off the screen the output draws, so it joins the rows the token fills either way. - `harness::sign_in_links` finds the sign-in pages in output as it arrives. It accepts only those pages, so a program cannot send the browser elsewhere. A link still arriving is left out until it ends. - `authentication.v1.status` reports which harness logins the daemon holds. The protocol version stays the same, because a client already requires the daemon's exact build. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The SSH configuration agentd writes reaches each agent through a ProxyCommand that runs agentctl. It named no control-plane home, so for an agent on another home than the default, such as one chosen with `--home`, `ssh agentctl-<name>`, `agentctl ssh` and editors connecting over SSH asked the default home's agentd and found no such agent. The ProxyCommand now passes `--home`, quoted for the shell OpenSSH runs it with, and the forwarding command `agentctl vnc-info` prints does the same. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
`agentctl setup` will offer standard agents to start from. This adds the first, dotnet-backend-developer, with the catalog that describes the standard agents to the setup wizard and the KI-hub directory site. The agent references a published image, so its host needs no Docker. The image adds the .NET SDKs 8.0 and 10.0 as Ubuntu packages, and the containers feature: Podman with the `docker` command, Docker Compose and a Docker API socket that clients find without configuration. Ubuntu 26.04 carries only .NET 10, so .NET 8 comes from the backports archive of Canonical's .NET team, whose signing key sits beside the Dockerfile. CI builds it on the base image from the same commit, which is why its FROM names the base tag without a digest, and publishes it from main. It shares one workflow with the self-dev images, which differed from it only in the images they name. The manifest has no tmpfs at /tmp, since the base now keeps /tmp on the sandbox disk, and GITHUB_TOKEN is optional so the agent starts without GitHub access. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Getting a first agent took a hand-written manifest, the harnesses and `gh` installed on the host to sign in, a `.env` of tokens, and SSH wired up for an editor. `agentctl setup` does all of it in one wizard that a newcomer can answer with defaults and that every flag can answer for a script. The installers start it after a first managed install, unless `AGENT_SETUP=0`. An upgrade has agents already, and a standalone install is for images and CI jobs, so neither starts it. - Sources: a standard agent from the catalog, which the build embeds in agentctl so it works offline and matches its manifest schema; a team's agent on disk or on GitHub, fetched into a temporary checkout; or an agent set up before in ~/agents, offered with what fits its state. A manifest that mounts its own repository, by a relative source such as `../..`, mounts a checkout on this machine there, asked for when the manifest was fetched. - Each agent gets a folder of its own, ~/agents/<name>, holding its manifest, the files it reads and its `.env`. Replacing one asks first, and removes the old folder and agent before the new one takes its place, so no definition ends up mixed. - Sign-ins run in a short-lived sandbox, so the host needs no harness or `gh`, and the agent only ever sees placeholders. The sandbox has no clipboard, so the host copies GitHub's one-time code to its own. A harness login agentd holds and a GitHub account another agent uses are offered for reuse. - A standard agent whose image cannot be pulled, as before it is published, can be built here from the Dockerfile agentctl carries. - The review offers the manifest in an editor, for everything the questions leave at its recommended value, such as an exact size, the model or hosts the agent may not reach. - Questions take the arrow keys, Esc goes back by replaying the answers before it, and setup ends by printing the command that repeats it without questions. - Leftovers of a setup that ends early, its staging folder and sign-in agent, are removed by the next run; `get` and the TUI hide sign-in agents. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
elsand
force-pushed
the
feat/agentctl-setup
branch
from
October 29, 2026 18:49
73e2747 to
30f4e84
Compare
Author
|
/agentctl-experimental |
|
Experimental agentctl Install it on macOS or Linux, with platform=macos-aarch64
gh run download 40936043351 --repo digdir/digdir-agents --name "agent-${platform}" --dir agentctl-experimental
curl -fsSL https://raw.githubusercontent.com/digdir/digdir-agents/30f4e843eeabf24719e442d99670c2b528a7478a/agentctl/install.sh \
| AGENT_VERSION=v0.1.0-preview.13.experimental.pr359.4 AGENT_LOCAL_ARCHIVE="${PWD}/agentctl-experimental/agent-${platform}.tar.gz" shOn Windows, with $Platform = 'windows-x86_64'
gh run download 40936043351 --repo digdir/digdir-agents --name "agent-$Platform" --dir agentctl-experimental
$env:AGENT_VERSION = 'v0.1.0-preview.13.experimental.pr359.4'
$env:AGENT_LOCAL_ARCHIVE = "$PWD\agentctl-experimental\agent-$Platform.tar.gz"
irm https://raw.githubusercontent.com/digdir/digdir-agents/30f4e843eeabf24719e442d99670c2b528a7478a/agentctl/install.ps1 | iex
Remove-Item Env:AGENT_VERSION, Env:AGENT_LOCAL_ARCHIVEWithout the GitHub CLI, download
|
This was referenced Oct 29, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
agentctl setuptakes you from an installed agentctl to a running agent, connected to your editor if you like. A newcomer answers a few questions with defaults. Every question also has a flag, and setup ends by printing the command that repeats the same setup without questions. This is an alternative to the onboarding stack in #309–#320.agents/standard/, which the build embeds in agentctl. A team's agent from a path or a GitHub address, fetched into a temporary checkout. When its manifest mounts its own repository, as Altinn'sworktreevariant does, setup mounts a checkout on this machine there. Or an agent set up before, which setup offers first with what fits its state.ghon the host. Claude Code, Codex and GitHub sign in inside a short-lived sandbox of the base image. agentd keeps the credential, and the agent sees only placeholders. Logins agentctl holds and GitHub accounts other agents use are offered for reuse.~/agents/<name>/holds the manifest, the files it reads and its.env. Replacing a definition asks first and swaps the whole folder.dotnet-backend-developer: .NET 8 and 10, Podman,dockerand Docker Compose. When its image cannot be pulled, setup offers to build it here from the Dockerfile agentctl carries.vnc-infocommands now name the control-plane home, so they reach agents on a home other than the default.The four commits review one at a time: the sign-in primitives, the SSH and VNC fix, the standard agent, and setup.
Try it
Comment
/agentctl-experimentalfor an installable build and install it as the reply says. You have agentctl already, so the installer does not start setup. Run it on a home of its own, so your agents stay as they are:agentctl --home ~/.agent-try setup--from https://github.com/Altinn/altinn-studio//agents/minimal.Clean up with
agentctl --home ~/.agent-try delete agent/<name>andrm -rf ~/agents/<name>, since the folder lives in~/agentswhichever home you use.🤖 Generated with Claude Code