Solo + Dev Mode Onboarding

This guide is for a solo developer who has chosen Dev mode: you work directly in your own coding-agent terminal while Creator Engine wraps that session with governance. If you prefer a less hands-on tour, see solo-ceo-onboarding.md instead.

You are in: Solo + Dev mode. You still type prompts to Claude Code or Codex yourself. CE's job is to start the governed session, keep the local state healthy, and hold privileged actions until the right human approval is present.

The day-one shape

The first run is deliberately small:

1. Prepare the host and repository.
2. Run ce onboard.
3. Use the governed agent pane that opens.
4. Return later with ce launch --backend host.

The detailed Frame -> Shape -> Build -> Review -> Ship loop happens inside that governed session. On day one, you do not need to memorize a command table before you can get value.

Prerequisites

Before onboarding, make sure these are true:

• A supported coding-agent CLI is installed and usable from your shell, such as Claude Code or Codex.

• The repository ignores CE's local Hermes state:

  .hermes/
  

• You have a CE install. If you have not installed yet, follow zero-to-governed-seat-quickstart.md first.

The .hermes/ ignore rule matters because CE refuses to create tracked local runtime state during onboarding. Add it before the first run, commit the ignore change through your normal project process if needed, then continue.

Phase 0 - First run: onboard

From the repository you want to use with CE, run:

ce onboard

ce onboard performs the first-run sequence: it verifies the local install, checks host readiness, initializes local CE state, prepares your path, and opens the first governed agent pane unless you explicitly opt out.

If you want to prepare the host and state without opening the pane yet, run:

ce onboard --no-launch

If onboarding refuses, read the refusal text and fix that specific prerequisite before trying again. Re-run support is still being tightened; do not assume ce onboard is a general recovery command for partially reused state.

Phase 1 - Daily entry point: launch

After onboarding has completed, return to the governed session with:

ce launch --backend host

ce launch --backend host opens or attaches the visible governed terminal launcher for your agent. It is the daily entry point after the repository has been onboarded; it is not the first command on a fresh host.

For a dry-run of the launcher plan without opening a pane, use:

ce launch --backend host --dry-run

Phase 2 - Work in the pane

Inside the governed pane, work the way you normally work with your coding agent:

1. Describe the change you want.
2. Answer clarifying questions.
3. Read the proposed work boundary and acceptance criteria.
4. Approve only the privileged actions you actually intend.
5. Review the resulting diff, tests, and pull request evidence before anything ships.

CE stays quiet until it has a reason to intervene. The important moment is when the agent reaches for something privileged, such as source-host mutation or credential access. CE refuses or holds that action outside the agent and tells you why.

For the in-product vocabulary tour, run:

ce guide

For a host/environment preflight, run:

ce doctor

Repo-connected install rail

ce onboard is the quick first-run rail for getting a governed local session. Connecting CE to a repository it can ship to is a separate, human-approved install rail. When you have the verified install artifacts and answers file, the shape is:

ce install \
  --spec <verified-spec> \
  --answers ce-install.answers.yaml \
  --plan

Then, only after the plan is acceptable:

ce install \
  --spec <verified-spec> \
  --answers ce-install.answers.yaml \
  --apply --non-interactive

The full version of that flow, including trust roots, anchors, and answer schema paths, lives in zero-to-governed-seat-quickstart.md.

Your irreducible gestures

CE is designed so that a small set of decisions stays with you:

In Dev mode you stay close to the shell and the agent pane. The governance layer is outside the agent, so the agent cannot reason around it.

Where to go next