Self-manage NextClaw runtime via CLI guide. For install/start/status/doctor/service/channels/config/agents/projects/sessions/apps and App data/cron/remote/update operations, and for discovering local HTTP/API/webhook addresses.
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
A direct command skips the review prompt. Inspect the source before running it.
Self-manage NextClaw runtime via CLI guide. For install/start/status/doctor/service/channels/config/agents/projects/sessions/apps and App data/cron/remote/update operations, and for discovering local HTTP/API/webhook addresses.
Use this skill whenever the user asks to manage NextClaw itself (version, service status, diagnostics, channels, config, agents, projects, sessions, apps and App data, cron, remote, update, installed skills, marketplace skills, local HTTP/API/webhook addresses).
Source of Truth
Always use the built-in NextClaw self-management guide as the operation guide.
Read the built-in guide path provided by the system prompt first.
If the packaged guide is unavailable in a repo source checkout, use repo docs path docs/USAGE.md.
Never treat workspace USAGE.md snapshots or copied artifacts as the source of truth.
If both guide paths are unavailable, use nextclaw --help and nextclaw <subcommand> --help as fallback and tell the user that guide file is missing.
Routing Rules
Treat NextClaw self-management as a product-management intent, not a generic "create/install/publish" intent.
Read the built-in self-management guide before opening unrelated generic skills.
Example: "create a new Agent" maps to NextClaw agent management, not skill-creator.
Stable Execution Rules
Map version lookup directly to nextclaw --version; do not substitute status for version queries.
Prefer machine-readable output: use --json when available.
Before calling local HTTP APIs or /webhook, run nextclaw status --json and read endpoints.uiUrl / endpoints.apiUrl; do not guess the service port.
For webhook payload details, read the focused guide linked from the self-management guide only when you need to implement or debug a webhook caller.
Execute only commands documented in the self-management guide or CLI help; do not invent commands or config paths.
In desktop-installed runtimes, still use the same nextclaw ... commands. The desktop launcher provides a managed command surface to AI command tools; do not ask the user to install the NPM CLI just to run self-management commands.
Keep installed skills and marketplace catalog as two different domains:
local installed: nextclaw skills installed|info
marketplace catalog and installed marketplace lifecycle: nextclaw marketplace skills search|info|recommend|install|update
After mutating operations, close the loop with:
nextclaw status --json
and nextclaw doctor --json when needed
Be explicit about restart semantics after changes. When a restart is required, ask the user to run nextclaw restart in an external terminal; do not invoke a restart from the active agent session.
nextclaw gateway starts a foreground gateway. It has no start, status, restart, or stop subcommands.
After modifying a running Service App, run nextclaw app restart <app-id> --json before validating through the live product UI or panel-to-service action calls.
Treat App code and managed App data as separate lifecycles. Uninstall/removal keeps managed data by default; delete it only when the user explicitly chooses the destructive data-removal option.
For App data discovery, run nextclaw app data list --json against the running host. Treat the returned id, appId, lifecycle, usage breakdown, and instance path as authoritative.
Independently delete only an entry whose latest list result says and . Use , where the data id comes from that list result and the confirmation exactly matches its .
Minimal Self-Management Flow
Understand user intent and map to one concrete CLI action.
Read the relevant section in the self-management guide.
Execute the documented command with safe parameters.
Verify with status/doctor.
Report outcome + next action (if any).
Release Notes / Changelog Lookup
When user asks "what changed in version X", follow:
references/release-notes-changelog.md
Do not claim details without a traceable source path.
High-frequency Intents
Version lookup: nextclaw --version
Service health: nextclaw status --json / nextclaw doctor --json
Local HTTP/API/webhook addresses: nextclaw status --json and read endpoints.uiUrl / endpoints.apiUrl
Lifecycle: nextclaw start|restart|stop
Service App live runtime: nextclaw app restart <app-id> --json
nextclaw app data delete <data-id> --confirm <app-id> --json
appId
Never synthesize App data ids, delete an active entry through the retained-data command, or substitute direct recursive filesystem deletion for the managed API. After deletion, run nextclaw app data list --json again and verify the exact entry is absent.
Reset a Service App development instance only when explicitly requested, using nextclaw app dev <service-app-dir> --reset-data --confirm <app-id> --json; the confirmation must match the manifest id.
App data removal never authorizes deletion of documents or directories separately granted to the App outside its managed instance.
For channel discovery before messaging, use nextclaw channels list --json and treat returned channels[].id values as authoritative.
For cron notifications, do not add delivery flags to the cron command. Put the notification intent in the scheduled message and let the scheduled agent call the message tool with an explicit channel and recipient.
For Agent creation/update/removal, treat nextclaw agents list|new|update|remove --json as the default path and follow the Agent management section in the self-management guide.
For runtime context injection, use nextclaw agents runtime config <runtime-id> --json to inspect and --inject-nextclaw-context <true|false> to update. Run nextclaw restart in an external terminal after changing it.
For project creation and discovery, use nextclaw projects list|templates|create --json; do not synthesize placeholder sessions or edit the project registry file directly.
For session naming and project binding, use nextclaw sessions rename|set-project|clear-project --json; do not edit session journal metadata directly.
Do not edit config.json or agents.list directly for normal Agent CRUD; only do that when the user explicitly wants a manual recovery path.
When creating an Agent, prefer an explicit non-text avatar and avoid text-based styles such as DiceBear initials.
