wehub-resource-sync
124 lines
4.8 KiB
Markdown
124 lines
4.8 KiB
Markdown
# Skills
|
|
|
|
## Installing the Entry Skill
|
|
|
|
Tell your agent:
|
|
|
|
> Install the browser-act Skill from https://github.com/browser-act/skills/tree/main/browser-act
|
|
|
|
The agent automatically downloads the SKILL file into its Skills directory. Once installed, the agent becomes aware of BrowserAct and invokes it when appropriate.
|
|
|
|
## Two-Layer Architecture
|
|
|
|
```
|
|
┌─────────────────────────────────────────┐
|
|
│ Entry Skill (the installed file) │
|
|
│ Lightweight, stable, rarely changes │
|
|
│ Teaches the agent: "browser-act exists"│
|
|
├─────────────────────────────────────────┤
|
|
│ Runtime content (get-skills CLI) │
|
|
│ Environment-aware, version-matched │
|
|
│ Teaches the agent: "use it like this" │
|
|
└─────────────────────────────────────────┘
|
|
```
|
|
|
|
### Layer 1: Entry Skill
|
|
|
|
The file installed into the agent's Skills directory. It's intentionally minimal:
|
|
- Triggers agent awareness of BrowserAct
|
|
- Contains trigger words that activate it
|
|
- Points the agent to `get-skills` for actual instructions
|
|
|
|
The entry Skill rarely changes — it's the stable entry point.
|
|
|
|
### Layer 2: Runtime Content via `get-skills`
|
|
|
|
The agent's first command must always be:
|
|
|
|
```bash
|
|
browser-act get-skills core --skill-version <version>
|
|
```
|
|
|
|
It returns everything the agent needs for the current session:
|
|
- **Environment state** — CLI version, API Key status
|
|
- **Browser list** — All available browsers with their IDs, types, and descriptions
|
|
- **Active sessions** — Currently running sessions
|
|
- **Core commands** — Interaction reference
|
|
- **Directives** — Dynamic rules based on the current state
|
|
|
|
## Topics
|
|
|
|
| Command | Content | When to use |
|
|
|---------|---------|-------------|
|
|
| `get-skills core --skill-version <v>` | Core workflow, commands, environment state | Always first |
|
|
| `get-skills advanced` | Proxy, Profile, privacy mode, advanced patterns | When core isn't enough |
|
|
| `get-skills main` | Latest SKILL.md content for self-update | When a version mismatch is detected |
|
|
|
|
## Progressive Loading
|
|
|
|
- **Most tasks only need `get-skills core`** — covers 80% of use cases
|
|
- **Load `get-skills advanced` only when needed** — proxy config, Profile import, etc.
|
|
- Avoid pulling unrelated information all at once
|
|
|
|
## Version Compatibility
|
|
|
|
The `--skill-version` parameter lets the CLI detect incompatibilities:
|
|
|
|
```bash
|
|
# The Skill claims version 2.0.2
|
|
browser-act get-skills core --skill-version 2.0.2
|
|
```
|
|
|
|
If the CLI version is incompatible with the Skill version, the output includes upgrade guidance:
|
|
- CLI too old → `uv tool upgrade browser-act-cli`
|
|
- Skill too old → `browser-act get-skills main` for the latest Skill content
|
|
|
|
Version compatibility is automatic: the Skill file declares its version, the CLI checks compatibility. When incompatible, the upgrade guidance lands directly in the output and the agent runs it automatically.
|
|
|
|
## Dynamic Directives
|
|
|
|
`get-skills` output includes "Directives" — context-aware rules generated based on the current state:
|
|
|
|
| Directive | Trigger |
|
|
|-----------|---------|
|
|
| **Multi-browser directive** | When multiple browsers exist, provides selection guidance based on `desc` matching |
|
|
| **Session directive** | When existing sessions are detected, reminds the agent of ownership rules |
|
|
| **Authentication directive** | When the API Key is missing, steers away from stealth operations |
|
|
|
|
Directives are BrowserAct's mechanism for adapting guidance to the agent's current situation without hard-coding every scenario in the Skill.
|
|
|
|
## Workflow from the Agent's Perspective
|
|
|
|
```
|
|
1. User says "check the price on example.com"
|
|
2. The agent's entry Skill activates (trigger word match)
|
|
3. Agent runs: browser-act get-skills core --skill-version 2.0.2
|
|
4. Agent receives: environment state + browser list + commands + dynamic directives
|
|
5. Agent picks an appropriate browser (desc match)
|
|
6. Agent confirms with the user (confirmation gate)
|
|
7. Agent executes browser operations
|
|
8. Agent closes the session, updates the browser's desc
|
|
```
|
|
|
|
## Agent Compatibility
|
|
|
|
Works with any agent that can:
|
|
1. Read SKILL files from a Skills directory
|
|
2. Execute shell commands
|
|
3. Parse text output
|
|
|
|
Compatibility list:
|
|
- **Claude Code** — SKILL.md in `.claude/skills/`
|
|
- **GitHub Copilot** — Discovered via the Skills directory
|
|
- **Cursor** — Rules / Skills directory
|
|
- **Windsurf** — Agent Skills system
|
|
- **Google Gemini CLI** — AGENTS.md integration
|
|
- **OpenCode** — Skills system
|
|
- **Codex** — Skills integration
|
|
|
|
## Next Steps
|
|
|
|
- [Quick Start](quick-start.md) — Run your first automation
|
|
- [Commands](commands.md) — Complete command reference
|
|
- [Skill Forge](skill-forge.md) — Bake operations into reusable Skills
|