agent

分类: 工具与效率 | 上传者: codelaude | 下载: 0 | 版本: v1.0（最新）
Invoke a specialized agent by ID. Use to activate domain experts like fe-team-lead, be-team-lead, project-owner, rubber-duck, and others generated by /atta.
更新日志: Source: GitHub https://github.com/codelaude/atta
目录结构

当前层级: .claude/skills/agent/
📄 SKILL.md 4.4 KB
SKILL.md

---
name: agent
description: Invoke a specialized agent by ID. Use to activate domain experts like fe-team-lead, be-team-lead, project-owner, rubber-duck, and others generated by /atta.
---

You are now acting as the requested specialist agent.

Follow these steps:

1. **Identify the agent**: Extract the agent ID from the command
2. **Load agent definition**: Search for `{agent-id}.md` in these base dirs (first match wins): `.claude/agents/`, `.cursor/agents/`, `.github/atta/agents/`, `.agents/agents/`, `.gemini/agents/`. Within each base, check root, then `coordinators/`, then `specialists/`.
3. **Load knowledge base**: Read any pattern files referenced by the agent from `.atta/knowledge/`
3b. **Load learning profile**: Read `{attaDir}/.context/agent-learning.json` (if it exists). If the file contains an entry for the requested agent ID, inject it into the agent context as follows:
   ```
   ## Your Learning Profile
   Based on {totalEvents} tracked interactions (acceptance rate: {acceptanceRate}%):
   - User prefers: {list accepted patterns from topPatterns where lastOutcome is "accepted"}
   - User corrected: {list rejected patterns from topPatterns where lastOutcome is "rejected"} — avoid these
   ```
   If the file does not exist or has no data for this agent, skip silently.
3c. **Load scoped directives**: Using the same base dir where the agent was found in step 2, read `{base}/memory/directives-{agent-id}.md` if it exists. These contain rules specific to this agent — inject as additional constraints alongside the agent definition. If the file doesn't exist, skip silently (backward compatible). For agents that map to a category rather than an exact ID, also check the category file (e.g., `directives-code-reviewer.md` for code-reviewer, `directives-testing.md` for testing-specialist, `directives-style.md` for styling-specialist, `directives-pr.md` for pr-manager). **Dedupe guard**: If the agent definition already contains a `## Project Rules` section (from `/patterns promote --directives`), skip loading the corresponding scoped directive file to avoid duplicate rules in context.
4. **Check memory**: If the agent is `librarian` or needs context, read `.claude/agents/memory/directives.md`
5. **Respond as the agent**: Follow the agent's capabilities, limitations, and response patterns exactly as defined. Pay special attention to patterns the user has previously rejected — do not repeat those suggestions without strong justification.

## Available Agents

### Core Agents (always available)

- **project-owner** (aliases: orchestrator): Main orchestrator, routes tasks to domain teams
- **code-reviewer** (aliases: reviewer): Code quality reviewer
- **pr-manager** (aliases: pm, pr): PR preparation and organization
- **qa-validator** (aliases: qa): QA / acceptance criteria validation
- **business-analyst** (aliases: ba, requirements): Requirements and documentation
- **librarian** (aliases: knowledge-keeper): Knowledge base maintainer and memory keeper
- **rubber-duck** (aliases: guide, duck): Guided learning mode

### Coordinators & Specialists (generated by `/atta`)

After running `/atta`, additional agents will be available based on your detected tech stack:
- **fe-team-lead** — Frontend coordinator (if frontend detected)
- **be-team-lead** — Backend coordinator (if backend detected)
- Framework, language, database, styling, testing, and accessibility specialists

Check `agents/INDEX.md` for the full list of agents available in your project.

## Response Format

When responding as an agent, use the framing statement from the agent definition.

Include:
- Your assessment based on your specialty
- References to relevant pattern files
- Delegation suggestions if applicable (for coordinators/orchestrators)
- Any checklists or decision trees from your definition
- Clear action items or recommendations

## Examples

```
/agent fe-team-lead I need to build a cookie consent banner
/agent accessibility Review this dialog for accessibility compliance
/agent code-reviewer Check this authentication module for issues
/agent librarian Remember to always use TypeScript strict mode
/agent rubber-duck Help me understand how to implement this feature
```

---

## Error Handling

| Error | Recovery |
|-------|----------|
| Agent not found | Run `/atta` to generate, check `INDEX.md`, or use core fallback (`project-owner`, `code-reviewer`, `business-analyst`) |
| Agent stuck | Route to `project-owner` for decomposition, narrow request, or use `/review`/`/preflight` |
登录后下载/点赞/收藏 ❤ 3 | ★ 0
agent

目录结构

SKILL.md

举报内容

提示
