@binaryworks1024/ai-task-hub-connector

Local bootstrap and command wrapper for AI Task Hub connector installs.

Product Position

• This connector is the officially supported host integration layer for the current built-in hosts openclaw, mobileclaw, codex, and claude.
• Future hosts should reuse the same connector/continuity contract instead of defining a new identity flow.
• The CLI accepts --host local-command or any lowercase host slug. For a future host, use the same slug for --host and --entry-host unless you intentionally need a different mapping.
• Account authorization is conditional, not a first-use prerequisite.
• When trial points are available, users may invoke AI capabilities before connecting an account.
• The connector preserves and reuses the same entry_user_key from the first call through post-authorization follow-up requests.
• entry_user_key is a continuity handle, not a user credential.

Install

npm install -g @binaryworks1024/ai-task-hub-connector

For one-shot local host setup, npx is the preferred entrypoint because it avoids a separate global install step:

npx -y @binaryworks1024/ai-task-hub-connector@latest bootstrap --host mobileclaw --check
npx -y @binaryworks1024/ai-task-hub-connector@latest bootstrap --host codex --check

Bootstrap

ai-task-hub-connector bootstrap --host openclaw
ai-task-hub-connector bootstrap --host mobileclaw
ai-task-hub-connector bootstrap --host codex
ai-task-hub-connector bootstrap --host claude
ai-task-hub-connector bootstrap --host futureclaw --entry-host futureclaw

--check runs the post-bootstrap doctor checks immediately, which is the recommended local flow for mobileclaw and codex:

npx -y @binaryworks1024/ai-task-hub-connector@latest bootstrap --host mobileclaw --check
npx -y @binaryworks1024/ai-task-hub-connector@latest bootstrap --host codex --check

bootstrap writes:

• connector install manifest
• bridge context store
• launcher script
• host connector config

Default host config locations:

• ~/.openclaw/connectors/ai-task-hub-connector.json
• ~/.mobileclaw/connectors/ai-task-hub-connector.json
• ~/.codex/connectors/ai-task-hub-connector.json
• ~/.claude/connectors/ai-task-hub-connector.json
• ~/.<host>/connectors/ai-task-hub-connector.json for future non-local hosts

Common Commands

ai-task-hub-connector connect --host mobileclaw
ai-task-hub-connector status --host mobileclaw
ai-task-hub-connector run --host mobileclaw --entry-host mobileclaw --capability image-generation --prompt "Create a polished portrait"
ai-task-hub-connector run --host mobileclaw --entry-host mobileclaw --capability tts_report --prompt "Read this text aloud" --confirm-write true
ai-task-hub-connector execute --host mobileclaw --entry-host mobileclaw --capability image-generation --prompt "Create a polished portrait"
ai-task-hub-connector logout --host mobileclaw

run is the preferred high-level entrypoint when you want the connector to return the result surface first and keep lower-level orchestration out of the host/agent path. In the current V1 implementation, run is the default result-first entrypoint for capabilities that fit instant_result or aggregated_short_wait, such as tts_low_cost, tts_report, and image-generation. asset_delivery and long_running capabilities, such as segmentation, matting, cutout, and larger async tasks, still keep their formal result surface on the existing presentation or explicit follow-up path. execute remains the raw compatibility path when you intentionally need lower-level action control or explicit follow-up handling. Use --confirm-write true for write-capable actions that require explicit confirmation, and --verbose true when you need raw transport details such as poll history.

Non-goals:

• Do not treat execute as the preferred default when run already covers the result-first path.
• Do not create a host-specific continuity model outside connector/runtime.
• Do not teach every capability as execute -> poll -> presentation; follow the published delivery_mode.
• Do not expose connector/raw transport details to end users when the connector can already return the result surface directly.

connect is for browser authorization and connection status. It does not replace connector-managed continuity. The same connector install should keep reusing the same entry_user_key before and after browser authorization.

Homepage

• https://gateway.binaryworks.app

Package Pair

This CLI publishes together with @binaryworks1024/ai-task-hub-connector-runtime.

Repository:

• https://github.com/chinasilva/skill_hub_gateway_mvp