homerail-install-ops

Files (2)

• SKILL.md
• openai.yaml · agents/

SKILL.md (first lines)

# HomeRail Install Ops

Before acting, apply `homerail-shared` rules. This pre-publication flow uses a local source checkout or local Git clone, not npm publication.

## Local Skill Install

Install HomeRail skills by linking the checkout's `skills/homerail-*`
directories into the target agent skill directory. Prefer symlinks over copies
so skill updates follow `git pull`.

If the source tree is already present, use `skills/README.md` as the source of
truth for Codex and Claude Code install targets. On macOS/Linux:

```bash
repo=/path/to/HomeRail

for root in "${CODEX_HOME:-$HOME/.codex}/skills" "$HOME/.claude/skills"; do
  mkdir -p "$root"
  for skill in "$repo"/skills/homerail-*; do
    [ -d "$skill" ] || continue
    name="$(basename "$skill")"
    dst="$root/$name"
    if [ -L "$dst" ]; then
      rm "$dst"
    elif [ -e "$dst" ]; then
      echo "Refusing to replace existing non-symlink skill: $dst" >&2
      continue
    fi
    ln -s "$skill" "$dst"
  done
done
```

If the user gives a local Git address, clone it first, then install links from
the checkout:

```bash
git clone <local-homerail-git-url> HomeRail
cd HomeRail
```

On Windows, use PowerShell directory symlinks when Developer Mode or
Administrator permissions allow them. If symlinks are blocked, use a junction
as the local fallback; repeat the junction for every `skills/homerail-*`
directory that should be installed:

```powershell
cmd /c mklink /J "%USERPROFILE%\.claude\skills\homerail-install-ops" "C:\path\to\HomeRail\skills\homerail-install-ops"
```

After installing or updating skills, ask the user to restart the AI agent session if their agent host only loads skills on startup.

After restart, route user requests to the root skills this way:

- Use `homerail-install-ops` for first install, update, service startup,
  provider configuration, and smoke verification. Codex invocation:
  `$homerail-install-ops`; Claude Code invocation: `/homerail-install-ops`.
- Use `homerail-dag-ops` for the full DAG lifecycle: pattern selection, workflow
  authoring, execution, background supervision, decisions and output inspection.
  Its references cover advanced modes; no separate supervision skill is needed.
- Use `homerail-cli` for command syntax, flags, and evidence commands.
- Use `homerail-shared` as background rules for all HomeRail operations.

## Deploy From Local Source

Run from the source root:

```bash
npm run install:all
npm run build
npm run typecheck
npm test
npm run ci
```

Prepare a local data root:

```bash
export HOMERAIL_HOME="${HOMERAIL_HOME:-$HOME/.homerail}"
```

Configure local runtime settings and credentials through the CLI:

```bash
node homerail_cli/dist/cli.js config
```

Start Manager and Node together:

```bash
node homerail_cli/dist/cli.js start
node homerail_cli/dist/cli.js start --rebuild-worker-image
node homerail_cli/dist/cli.js start --ui
node homerail_cli/dist/cli.js doctor
```

`hr start` uses the shared `HOMERAIL_HOME`, starts Manager if needed, applies
stored model config, builds `homerail-worker:latest` when missing, and starts a
Node with Docker capability when no configured Node is connected.
Use `--rebuild-worker-image` after Worker runtime changes so provisioned DAG
workers run the current checkout.
Use `--ui` or `node homerail_cli/dist/cli.js ui start` to start the decoupled
Agent UI. HTTPS is the default endpoint, `https://localhost:19192`, and HTTP is
kept as a fallback on `http://localhost:19193`.
If provider credentials are not configured yet, `hr start` should still bring
up Manager and Node; `hr doctor` reports the missing model or Manager Agent
runtime configuration and the agent should run `hr config` /
`hr model configure`.

Use `node homerail_cli/dist/cli.js runtime status|logs|stop` for service
inspection and cleanup. Use `node homerail_cli/dist/cli.js ui status|logs|stop`
for UI-only inspection and cleanup.

`hr doctor` is the first environment probe an agent needs for normal operation.
Do not supplement it with `ps`, `lsof`, `curl /health`, or source reading unless
the CLI output is insufficient for a concrete failure. It reports,
per line: manager reachability + URL, runtime state (connected nodes/workers,
active runs, node capabilities), node count, any active LLM model setting, and
the Manager Agent runtime/harness readiness.

From xiaotianfotos/homerail, MIT.

More from xiaotianfotos/homerail