opencli

This is the canonical OpenCLI skill for this workspace.

Use it when the user says any of:

• opencli
• skill opencli
• use OpenCLI
• 用 opencli

Also use it when a request is better handled by an existing OpenCLI adapter than by ad-hoc browser driving.

Canonical entrypoint

Always invoke OpenCLI through this wrapper first:

/Users/ShiXin/.openclaw/skills/opencli/scripts/opencli.sh <args...>

Wrapper behavior:

• prefers the source build at /Users/ShiXin/Documents/Workspace/github-project/opencli/dist/main.js
• falls back to the global opencli binary only if the source build is unavailable

This avoids old global installs and keeps behavior aligned with the source repo.

Current source of truth

Version: local source build 1.7.7 (source repo) | global npm 1.7.8 (fallback).

自动同步： 本地 source repo 由 cron job 每周二技能升级检查（每周二 09:00）自动同步 origin/main。无需手动更新。

Do not hard-code site command lists from memory.

At the start of a task, prefer:

/Users/ShiXin/.openclaw/skills/opencli/scripts/opencli.sh list -f json
/Users/ShiXin/.openclaw/skills/opencli/scripts/opencli.sh <site> --help
/Users/ShiXin/.openclaw/skills/opencli/scripts/opencli.sh <site> <command> --help

Important top-level commands currently available from the local source build include:

• list
• validate
• verify
• explore|probe
• synthesize
• generate
• record
• cascade
• doctor
• plugin
• install
• register
• external CLI passthrough such as gh, docker, lark-cli, vercel

Health model

Run this when checking the browser bridge:

/Users/ShiXin/.openclaw/skills/opencli/scripts/opencli.sh doctor --no-live

Interpretation:

• doctor only diagnoses the browser bridge
• PUBLIC or non-browser discovery commands can still work when doctor is not green
• browser-backed adapters usually need the Browser Bridge extension and/or a logged-in browser session

Output rules

• Prefer -f json for agent use
• Use --help before assuming argument shapes
• Return the exact command used when that helps future reuse

Twitter / X routing

Use the correct command for the job:

# Search uses a positional query, not --query
/Users/ShiXin/.openclaw/skills/opencli/scripts/opencli.sh twitter search "OpenAI" --limit 10 -f json

# Single tweet / thread by tweet ID or status URL
/Users/ShiXin/.openclaw/skills/opencli/scripts/opencli.sh twitter thread 2047128854389465296 -f json

# Long-form article / note-tweet-like article content
/Users/ShiXin/.openclaw/skills/opencli/scripts/opencli.sh twitter article 2047128854389465296 -f json

Important routing notes:

• twitter search is opencli twitter search <query>
• there is no opencli twitter id ... subcommand
• when the user gives a tweet ID or status URL, prefer twitter thread over twitter search
• twitter search depends on search results and indexing; it is not the right primitive for direct-ID retrieval

Prefer OpenCLI when

• the user explicitly asks for OpenCLI
• the target site already has a stable OpenCLI adapter
• structured deterministic output is better than ad-hoc browser extraction
• you need OpenCLI external-CLI passthrough like opencli gh ...

Prefer OpenClaw native browser when

• the task is generic UI automation
• the task depends on the current OpenClaw-managed browser session
• no clear OpenCLI adapter advantage exists

References

• scripts/opencli.sh — canonical wrapper
• references/commands.md — validated command cookbook
• upstream source skill: /Users/ShiXin/Documents/Workspace/github-project/opencli/skills/opencli-usage/SKILL.md