For coding agents

Delegate from your agent. Get a pull request back.

Your DevIntern worker already turns tickets, schedules, and alerts into pull requests across every repository you register. Your coding agent can use it too: agree on a fix in Claude Code or Codex, and the agent queues it for the worker instead of doing it inline or opening a ticket.

Install the agent skills Set up a worker

agent session · handing off

devintern worker add --repo backend \
  --title "Fix checkout retry handling" \
  --body-file brief.md \
  --idempotency-key checkout-retry-followup --json

{
  "id": 42,
  "taskKey": "local-fix-checkout-retry-handling",
  "repo": "backend",
  "status": "pending",
  "deduplicated": false
}

How it works

From chat to pull request in four steps

01

Agree on the work

You and your agent settle what should change in an interactive session: the behavior, the files involved, and how to verify it.
02

Pick the repository

The agent reads your workspace and matches the target by its git remote, not by a folder name. Only repositories registered with the worker can receive work.
03

Queue a self-contained brief

The agent writes a brief with acceptance criteria, scope, and verification commands, then queues it with one command. The worker never sees the chat, so the brief carries the context.
04

Get a pull request back

The worker runs the task in a fresh checkout with your configured coding agent and pipeline, opens a pull request, and addresses review comments on the same branch.

One worker, every repo

A software factory your agents can plug into

A workspace registers your organization's repositories once. One worker then takes work from every source below, runs each task in a disposable checkout, and returns reviewed pull requests. Agents are one more input, not a separate system.

Trackers

Ready tickets from Jira, Linear, GitHub Issues, GitLab, Azure DevOps, Asana, Trello, or Markdown files.

Schedules

Recurring prompts such as weekly dependency updates or repairing a failing main branch.

Monitors

Qualifying unresolved Sentry errors become repo-routed fixes.

Your agents

Follow-ups handed off from Claude Code, Codex, or any agent that can run a shell command.

Set up a multi-repo workspace →

Install

Teach your agent when and how to delegate

The skills cover setting up a worker, changing workspace policy, and dispatching tasks. They give your agent instructions only: you still connect accounts and run the worker yourself. Agents without skill support can call devintern worker add directly.

Rolling it out to a team? Commit the DevIntern marketplace to your repository's .claude/settings.json, and Claude Code prompts everyone who opens the repo to install it.

Agent Skills guide Delegation guide

Claude Code plugin

/plugin marketplace add getdevintern/devintern
/plugin install devintern@devintern

Codex and other agents (Skills CLI)

bunx skills add getdevintern/devintern \
  --skill devintern-setup devintern-workspace \
          devintern-delegate devintern-report \
  --agent codex   # or --agent claude-code

Guardrails

Only registered repositories

Agents target repositories listed in the workspace. Unknown or ambiguous destinations are reported, not guessed.

Retries do not duplicate work

An idempotency key reuses the pending or running task instead of queueing a second copy when an agent retries.

The same review bar

Delegated work goes through the same checkout, pipeline, and pull request flow as tracker tickets. Humans still decide what merges.

Your machines and keys

The worker and coding agent run on your hardware with your model provider. DevIntern services never receive code, diffs, or task briefs.

When not to delegate

A one-line change you can finish in the current session.
Work the user has not agreed to yet, or that cannot be specified up front.
A repository that is not registered with the worker, or a machine with no worker set up.
A team that wants a hosted service with no machine to run.

FAQ

Which coding agents can delegate to DevIntern?

Any agent that can run a shell command can call devintern worker add. DevIntern ships Agent Skills for Claude Code and Codex that teach the agent when to delegate, how to pick the repository, and how to write the brief.

Do I need a task tracker to delegate work?

No. Delegated tasks are queued directly on the worker for a named repository. They are recorded in the worker's run history and dashboard, and they still open a normal pull request.

Does the worker see my conversation with the agent?

No. The worker only receives the title and brief the agent submits, so the agent includes the context, acceptance criteria, and verification steps in that brief. Uncommitted local edits are not carried over either.

What does delegation cost?

Interactive use of DevIntern is free. Delegated tasks run unattended on the worker, which needs the 14-day Worker Pilot or a Supporter, Team, or Business license.

How does an agent know whether DevIntern fits?

DevIntern publishes an llms.txt index with fit criteria, delegation steps, and links to every doc as raw markdown, so agents can check the details before suggesting or using it.

Reading this as an agent? Start at /llms.txt.

Give your agent a worker to hand off to

Interactive use is free. Unattended runs, including delegated tasks, start with a 14-day Worker Pilot and no credit card.

Install the agent skills See pricing