runweft-kit

Original：🇨🇳 Chinese

Translated

Install, upgrade, and uninstall Runweft; connect Hub identities and spaces; register machines and Codex/Claude runtimes; manage Hub Agents, repositories, and personal selections; delegate, resume, stop, and retrieve code and attachments through MCP; and diagnose connection and execution issues. Use when the user mentions Runweft integration, collaborative Agents, the Hub backend, or task maintenance; do not use it to modify native model credentials or configure other products.

4installs

Sourceitgo-lyq/runweft-kit

Added on2026-10-14

NPX Install

npx skill4agent add itgo-lyq/runweft-kit runweft-kit

SKILL.md Content (Chinese)

View Translation Comparison →

Runweft Kit

Enable users to complete operations through conversation. You are responsible for the CLI and verification, while users only handle identity storage, space selection, and connection confirmation in their own browsers. Reuse existing installations and connections; do not rebuild the environment from scratch.

Choose the path first

Installing, making the first connection, or adding an execution machine: read onboarding.md.
Configuring machines/runtimes, Agents, repositories, and personal selections: read configuration.md.
Delegating, waiting, resuming, stopping, approving, and handling attachments and code: read delegation.md.
Troubleshooting, upgrading, uninstalling, and managing current attachments and artifact retention: read maintenance.md.

For a complete uninstall, first verify the version and preview the operation according to maintenance.md, and do not open business data. For other operations, first run

runweft --version

,

runweft --json config show

, and

runweft hub connect status

to determine the actual version, data root, Hub, and saved space. If the program is not installed, install it first. Use

--help

to check for version differences; do not invent commands. The global CLI options

--home <directory>

,

--space <UUID>

,

--identity-file <file>

, and

--json

must be placed before the subcommand. Replace all angle-bracket placeholders in the examples with actual returned values.

Boundaries that must not be omitted

Delegation requires a connection to an explicit Hub.
```
mode local
```
only hosts a local Hub; it is not an offline Agent/YAML mode. A local Hub also uses an independent identity and space permissions.
Do not request, read, or print identity backups, machine credentials, or native model credentials. Credentials must not be placed in prompts, logs, URLs, or command arguments. Web→CLI uses the encrypted
```
hub connect begin/complete
```
handoff. Only display secret-free verification codes or confirmation links. If an identity expires, ask the user to select the original identity; do not automatically create a new one.
Modify global installations/MCP, resource sharing, services, or data only within the scope authorized by the user. Preserve other MCP integrations, hooks, native configurations, and existing sessions. Do not modify model login, permission modes, or trust settings on execution machines without authorization.
When the user explicitly requests a complete Runweft uninstall, first preview the scope with
```
runweft --home <directory> uninstall
```
, then directly execute
```
runweft --home <directory> uninstall --yes
```
. This removes all data under that home directory, the current program, Runweft's own MCP/hook, and the confirmed Skill without creating a backup, while preserving other software configurations, model credentials, native sessions, and user projects.
Machines, runtimes, and Agents are three distinct objects. An Agent is a definition in the Hub, not the current primary host. Use full UUIDs; do not guess identifiers from names or create repository YAML. Administrative governance does not grant permission to read other users' private content.
Acceptance by
```
spawn
```
does not mean execution succeeded. For an unknown outcome, first query the original request/task and retain the original request ID and parameters. Do not blindly resend the request, clear history, or switch identities to bypass an uncertain state.
Managed sub-agents disable Runweft MCP and reject recursive CLI/MCP delegation; do not bypass these restrictions. After the primary host exits, do not wake it in the background or restore an old parent session.
Do not automatically push or merge artifacts, delete worktrees, or release the only copy of a result. Sharing a runtime is a native account execution capability, not an OS sandbox.

Deliver to the user

Briefly report the actual Hub, space, resource IDs, completed operations, and real verification results, and provide a secret-free

/hub

link. Clearly distinguish “tool discovered / MCP handshake / native invocation performed / model completed / artifact usable”; do not treat one as evidence of another. When an external condition is missing, identify the exact failure stage and required conditions. Do not return local checks to the user when those checks can still be completed automatically.