Codex CLI Orchestration for Founder-Led Teams
Mrrlin adds goal context, task state, handoffs, evidence, and review gates around Codex CLI work.
Chat tools vs Mrrlin
The execution layer beats another blank chat window.
Chat tools make you re-send context and manage the work. Mrrlin keeps project memory and spends tokens deliberately.
Your agents, under one harness
One coding agent gives you output. Mrrlin gives you a harness.
Codex, Claude Code, and other agents can write. Mrrlin keeps the brief, routes each task, challenges weak work, and brings you in only for decisions that need you.
Read the workflow transcript
Without a harness, a person carries context between prompts, checks each output, follows up, and decides what happens next. With Mrrlin, one goal becomes a visible plan: work is routed, reviewed, retried when weak, and brought back only for the decisions that need a human.
Why it matters
Codex CLI does the work. The session forgets it happened.
Mrrlin turns Codex CLI sessions into managed execution runs with context, artifacts, next decisions, and reviewable outcomes.
Workflow
Turning a Codex CLI session into a run with a record.
Frame the task for Codex
Good Codex CLI orchestration starts before Codex runs. Write the goal as a task with acceptance checks and notes about the files involved, or let the Director draft it from one sentence. Codex reads the task over MCP, so instructions arrive from the task itself rather than from your clipboard.
Run it on your subscription
Codex CLI is launched by the bridge on your own machine, under your own Codex subscription, inside an isolated worktree. The run record captures the branch, the base commit, the worktree and the provider, so nobody has to ask later where the change was made.
Keep evidence as it happens
The plan, checkpoints, command logs, test results and a summary of the diff are stored on the run while Codex works, and a produced URL is added once a preview exists. The status field shows whether the run is running, in self-review, at pull request or failed.
Decide what happens next
A reviewer from outside Codex's provider reads the result and either passes it or returns it with notes. Whatever the review turns up that is out of scope becomes a new task in the same plan instead of a comment nobody tracks.
Use cases
Where it fits.
Marketing-site changes
Ask Codex to add a changelog page to the marketing site and link it from the footer. The run records the build and a preview address, the reviewer checks the page against the task's acceptance checks, and publishing waits in your inbox until you approve it.
Internal tool updates
Add a CSV export to the support tickets screen of an internal admin tool. The task can run at the auto autonomy level because nobody outside the team sees it, but the change still gets its own worktree, test results and a separate review before it counts as done.
API implementation
Before Codex writes a paginated endpoint for order history, the spec is reviewed for architecture, security and scope until the reviewers agree on auth, limits and error shapes. Codex then implements against that spec, and its test results and diff summary land on the run as artifacts.
Build and route verification
After Codex restructures the site navigation, a verification step runs the build and checks every route and redirect. A broken redirect returns the run to Codex with the error it produced, and nothing is marked done until the build is green and every address responds.
Comparison
A Codex session on its own, and the same work as a tracked run.
Keep reading
More on coordinating coding agents.
Start from the AI coding agent orchestration overview, or go deeper with the guides next to this one.
FAQ
Before you start.
Does Codex CLI orchestration in Mrrlin bill us per token?
No. Codex CLI runs on your own Codex subscription, on your own machine, so the work is not metered as API calls, and Mrrlin neither stores your model credential nor resells usage. The subscription's own limits still apply, and the Autopilot daily job budget and concurrency cap exist so unattended runs stay within the usage you intend.
Can Codex read and update Mrrlin tasks directly?
Yes. The Mrrlin MCP server, published on npm as @mrrlin-dev/mcp, lets Codex read and update tasks, plans, the wiki, execution runs, artifacts and the inbox. That is what makes Codex CLI orchestration work without you as the relay: a Codex session can pick up its task, record evidence and raise a question on its own.
If a Codex run dies halfway through, is the work lost?
No. The run's status becomes failed, and its error is recorded as an artifact next to the checkpoints it reached. You can resume it, and the new attempt picks up with just the changes since the failure, not a replay of everything before it. Under Autopilot, a task that keeps failing stops after repeated attempts instead of looping through the night.
Orchestrate Claude Code & Codex CLI without terminal lockups.
Stop juggling disconnected CLI windows and resolving manual merge collisions. Mrrlin coordinates your agents in isolated git worktrees, collects passing test artifacts, and opens verified pull requests.
No credit card · No migration · One goal