In short:
ak-orchestrateturns multi-part work into ordered jobs, runs each job as an independent session, and tracks and saves them so you can resume. An arbiter checks every result and writesreport.md, with diffs left for you to integrate. It does not merge diffs into your branch, and by default it uses no runtime other than the one you are running.
Disclosure: I maintain AgentKit and have an affiliate relationship with it: the AgentKit link in this post is an affiliate link; buying through it gets you 30% off.
Summary
Use it when the work has parts that can run on their own: a few scouts reading code in parallel, two jobs changing two modules, one review job. You do not need it for a simple code read: when splitting the work buys nothing, the skill tells you to do it directly in one session.
flowchart LR
A["Large work, several independent parts"] --> B["ak-orchestrate"]
B --> C["Sessions run in stages"]
C --> D["C3 arbiter checks"]
D --> E["report.md and diffs awaiting integration"]
B --> F["One direct session if not worth it"]
D --> G["fail or blocked: reason stated"]On every run, it answers 4 questions:
| Part | Question | Example |
|---|---|---|
| Runtime scope | Where may a job run? | By default a session of the current runtime; name Pi to get Pi, add --cross-runtime for every runtime |
| Capability tier and risk tier | How strong a model does the job need, and what controls? | A scout is C1/R0, an important implement is C3/R2, deploying or deleting data is R3 and needs your approval |
| Worktrees for parallel write jobs | What if two jobs change code at once? | Each job gets its own worktree and branch; jobs that touch a shared lockfile or migration run one after another |
| Arbiter | Who checks results before reporting done? | A fresh session on the strongest model compares against expected_output and runs checks; a missing artifact cannot pass |
The most important caveat: a session's report is only a claim. A job is accepted only when its artifacts and checks are verified, and within a prepared run a job never changes runtime or model.
Sample prompt for parallel research:
/ak-orchestrate Research 3 ways to cache the API of <project name> and compare them.
Outcome: one comparison report, every claim backed by evidence from the code.
Constraints: read only, do not change any file.
Done when: the arbiter cross-checks the 3 reports and drops claims without evidence.| Flag | What it is for | Since ak |
|---|---|---|
--resume | Continues a run that was interrupted | 1.0.0 |
--yes | Records your permission for the run | 1.0.0 |
--cross-runtime | Allows runtimes other than the current one | 3.0.0-beta.33 |
--internal | Old flag, retired | 1.1.0 |
Below is the detailed version: the run flow and the fallback ladder for failed jobs, 5 situations with prompts, each flag explained, the change history by ak release, and common pitfalls. The detailed version and the interactive page are for subscribers. Your agent can also search them through the portal's MCP.
Opens when your account has subscriber access
This guide is for subscribers
Once your account has access, the full piece, the interactive visual and the agent call all open.
- Sign in or create a modernEngineer accountReading access belongs to your account email.
- Choose a plan on the pricing page1, 3 or 12 months, paid by VietQR bank transfer. Access opens about a minute after the money arrives.
- Or redeem a Facebook subscriber codePeople supporting the page on Facebook get an activation code by message each month.
Already a subscriber? Sign in to read.
Use this piece in your agent
Get an API keyCopy the text below and paste it into Claude Code, Codex or whichever agent you use. The agent reads the piece and checks it against the repo you have open.
Read "ak-orchestrate: split work into sessions with a final arbiter" on modernEngineer and apply it to this repo.
1. Call the content_get tool of the modernengineer MCP server with locale "en", slug "ak-orchestrate". If the modernengineer MCP server is not connected, stop and walk me through connecting it from https://modernengineer.dev/en/mcp.
2. Summarize the main steps of the piece and point out which apply to this repo.
3. Propose concrete changes and ask me before editing any file.Connect MCP in Claude Code (once)
claude mcp add --transport http modernengineer \
https://api.modernengineer.dev/mcp \
--header "Authorization: Bearer <API key>"Reading the full piece over MCP needs an API key from a subscriber account or one that bought this guide. Without access, the agent only gets the opening.
- Version
- v1.0.0
- Updated
- 22/10/2026
- Licence
- All rights reserved
- Slug
- ak-orchestrate
