Zero to shipped.

You will work across two surfaces. The web app is where you capture intel and review what your agent did, and your terminal or IDE is where the work happens. The naholo CLI works as a bridge between the web app and the local dev environment, for both humans and agents.

Prerequisites and setup

Lay the gear once per machine.

• A naholo account and a project. (Operations are created under a project.)
• macOS. (Linux and Windows support is coming later.)
• A local repository to make code changes in.
• Node.js v22 or newer, with npm or pnpm.
• Claude Code as your coding agent. Codex support is coming later.
  • If you use VS Code, the official Claude Code extension is highly recommended.
  • The Claude desktop app is not recommended. (The cycle requires reviewing a local file, OPERATION.md.)

Install the CLI, sign in, and wire the repository to a project.

npm install -g @naholo/cli
naholo login

# team setup: writes the project wiring into the repo so teammates inherit it
naholo init
# solo setup: keeps the same wiring under ~/.naholo/ and commits nothing
naholo covert init

Then install the plugin from your terminal. (You can update the plugin with the same command)

naholo claude-plugin install

Start (or restart) Claude Code so the plugin loads, then verify your setup with the doctor skill.

/naholo-doctor

After installing, if the skill does not show up in the VS Code Claude plugin, reload the window (press Cmd+P, then select Developer: Reload Window).

The doctor skill checks your CLI version, login, and project setup, and tells you how to fix anything that is off.

1. Create an operation and /infil

In the web app, create a new operation. Title it after the problem you want to solve or the idea you want to build, and jot down whatever comes to mind as logs or notes. Then run the skill below:

/infil <operationNumber>

Or skip the web app and run this in your agent:

/fob <title>
<first log (optional)>

/fob creates the operation server-side and chains /infil in one step, landing you in the same place.

2. /warno

Once the context is infilled, run the skill below:

/warno

Working from that context, it writes the solution's concept and architecture decisions into the WARNING ORDER. Review the Concept of Operations and the Constraints. If anything is unclear or you are not happy with it, you do not need another skill: just ask the agent or request changes in the same conversation. You can also edit the WARNING ORDER yourself.

That said, hold back on overly specific demands unless they are truly essential. They are not impossible to change, but they can get in the way of the OPERATION ORDER and the code changes.

3. /opord

Once the WARNING ORDER is complete, run the skill below:

/opord

It chops the WARNING ORDER prepared earlier into single-commit-sized tasks and writes them into the OPERATION ORDER. Tasks are described around module-level contracts, so module-level behavior changes are easy to review. Check the module names, the interface design, and the execution flow. If a task needs changes, again without another skill, ask the agent in the same conversation, or edit the file yourself.

There is no need to over-review here. Unlike the Constraints, these can still be adjusted after the code changes.

4. /splash

Once your review is done, run the skill below:

/splash

It runs the next unfinished task and writes its After-Action Report. If any change happens that was not described in the OPERATION ORDER, it is recorded under Deviations. Review the code changes and the After-Action Report.

If there is a minor issue with the code changes, without another skill you can ask the agent to fix it, or fix it yourself and then ask the agent to update the After-Action Report. If a larger change is needed, we recommend adding a task with /opord add a task....

If the code changes look good, move on to the next task. When it is fine, run /splash again; if it needs changes, adjust the task with /opord revise the next task... and then run /splash.

Post-skill phase: after /warno, /opord, or /splash runs, that skill's phase stays active until another skill is used. To edit the WARNING ORDER after /warno, the OPERATION ORDER after /opord, or the code after /splash, you do not need to re-run the skill: just ask directly in a prompt.

5. /exfil

Once every task is done, run the skill below:

/exfil

It registers all notes and agent usage data to the web app. When every task is finished, it also closes the operation on the web app.

Tip: Run /clear before /infiling a new operation to reset the context.

For detailed explanations of each skill and the extra skills like /recon and /chop, see the Workflow field manual.