imperium-crawl

分类: 工具与效率 | 上传者: ceoimperiumprojects | 下载: 0 | 版本: v1.0（最新）
Comprehensive guide for AI agents using imperium-crawl's 26 CLI tools. Covers scraping, extraction, research, API discovery, skill building, and batch processing.
更新日志: Source: GitHub https://github.com/ceoimperiumprojects/imperium-crawl
目录结构

当前层级: SKILL/
📄 api-recon.md 7.7 KB
📄 build-skill.md 13.4 KB
📄 pipelines.md 6.4 KB
📄 recipes.md 10.5 KB
📄 research.md 5.8 KB
📄 site-intel.md 6.2 KB
📄 SKILL.md 17.7 KB
📄 smart-scrape.md 6.1 KB
📄 tool-reference.md 17.5 KB
SKILL.md

# imperium-crawl — Agent Skill Guide

Comprehensive guide for AI agents using imperium-crawl's 26 CLI tools. Covers scraping, extraction, research, API discovery, skill building, and batch processing.

**Progressive disclosure:** This file is the overview hub. Each skill and reference topic has a dedicated file with full details — read them when you need depth.

---

## File Map

| File | Lines | What's inside |
|------|-------|---------------|
| **SKILL.md** (this file) | ~500 | Overview hub — all tools, decision tree, skill summaries |
| [build-skill.md](build-skill.md) | ~356 | Full build-skill workflow: 6 steps, 9 combos, selector patterns, recipes |
| [smart-scrape.md](smart-scrape.md) | ~184 | Full smart-scrape: decision tree, stealth escalation, tool chains |
| [site-intel.md](site-intel.md) | ~206 | Full site-intel: 5-step workflow, report template, depth guidelines |
| [research.md](research.md) | ~199 | Full research: search → scrape → synthesize, depth guidelines |
| [api-recon.md](api-recon.md) | ~218 | Full API recon: discovery, categorization, WebSocket, report template |
| [tool-reference.md](tool-reference.md) | ~500 | All 26 tools — params, types, defaults, gotchas |
| [pipelines.md](pipelines.md) | ~310 | 10 pipeline patterns with full CLI examples |
| [recipes.md](recipes.md) | ~151 | 10 built-in recipes + custom skill JSON format |

---

## Table of Contents

1. [All 25 Tools](#all-25-tools)
2. [Master Decision Tree](#master-decision-tree)
3. [Tool Combinations — 10 Patterns](#tool-combinations--10-patterns)
4. [Smart Scrape](#smart-scrape)
5. [Build Skill](#build-skill)
6. [Site Intel](#site-intel)
7. [Research](#research)
8. [API Recon](#api-recon)
9. [CLI Gotchas](#cli-gotchas)
10. [Error Recovery](#error-recovery)
11. [Environment Variables](#environment-variables)

---

## All 25 Tools

Full parameter details per tool → [tool-reference.md](tool-reference.md)

**Naming convention:** CLI commands use hyphen → `create-skill`, `ai-extract`, `discover-apis` (code uses underscore: `create_skill`, `ai_extract`, `discover_apis`)

### Scraping Tools (6)

| Action | CLI Command | Key Params |
|--------|-------------|------------|
| Scrape page | `imperium-crawl scrape --url URL` | `include`, `stealth_level`, `format` |
| Crawl site | `imperium-crawl crawl --url URL` | `max_depth`, `max_pages`, `concurrency` |
| Map URLs | `imperium-crawl map --url URL` | `max_urls`, `include_sitemap` |
| CSS extract | `imperium-crawl extract --url URL --selectors '{}'` | `selectors`, `items_selector`, `llm_fallback` |
| Clean article | `imperium-crawl readability --url URL` | `format` |
| Screenshot | `imperium-crawl screenshot --url URL` | `full_page` |

### Search Tools (4) — Require `BRAVE_API_KEY`

| Action | CLI Command | Key Params |
|--------|-------------|------------|
| Web search | `imperium-crawl search --query "..."` | `query`, `count`, `freshness` |
| News search | `imperium-crawl news-search --query "..."` | `query`, `count`, `freshness` |
| Image search | `imperium-crawl image-search --query "..."` | `query`, `count` |
| Video search | `imperium-crawl video-search --query "..."` | `query`, `count`, `freshness` |

### AI Extraction (1) — Requires `LLM_API_KEY`

| Action | CLI Command | Key Params |
|--------|-------------|------------|
| AI extract | `imperium-crawl ai-extract --url URL --schema "..."` | `schema` (string/object/"auto"), `format` |

### Skill Tools (3)

| Action | CLI Command | Key Params |
|--------|-------------|------------|
| Create skill | `imperium-crawl create-skill --url URL --name NAME --description "..."` | `url`, `name`, `description` |
| Run skill | `imperium-crawl run-skill --name NAME` | `name`, `url` (override), `max_items` |
| List skills | `imperium-crawl list-skills` | *(none)* |

### API Discovery Tools (3)

| Action | CLI Command | Key Params |
|--------|-------------|------------|
| Discover APIs | `imperium-crawl discover-apis --url URL` | `wait_seconds`, `include_headers` |
| Query API | `imperium-crawl query-api --url URL` | `method`, `headers`, `body`, `params` |
| Monitor WS | `imperium-crawl monitor-websocket --url URL` | `duration_seconds`, `max_messages` |

### Interaction (2)

| Action | CLI Command | Key Params |
|--------|-------------|------------|
| Interact | `imperium-crawl interact --url URL --actions '[...]'` | `actions`, `session_id`, `return_snapshot`, `action_policy_path`, `allowed_domains`, `device` |
| Snapshot | `imperium-crawl snapshot --url URL` | `session_id`, `return_screenshot`, `selector` |

### Batch Processing (4)

| Action | CLI Command | Key Params |
|--------|-------------|------------|
| Batch scrape | `imperium-crawl batch-scrape --urls "url1,url2"` | `urls`, `extraction_schema`, `concurrency` |
| List jobs | `imperium-crawl list-jobs` | *(none)* |
| Job status | `imperium-crawl job-status --job-id ID` | `job_id` |
| Delete job | `imperium-crawl delete-job --job-id ID` | `job_id` |

---

## Master Decision Tree

```
User has a web data task
│
├─ "Read this article / get the text"
│  └─ readability → if fails → scrape (markdown)
│  → Full guide: smart-scrape.md
│
├─ "Scrape this page / get content"
│  └─ scrape (markdown + metadata) → offer extract if structured needed
│  → Full guide: smart-scrape.md
│
├─ "Extract specific data (products, prices, listings)"
│  ├─ Know CSS selectors? → extract (fast, deterministic)
│  │  └─ Empty results? → enable llm_fallback: true
│  └─ Don't know structure? → ai_extract with "auto" or natural language
│  → Full guide: smart-scrape.md
│
├─ "Create a reusable scraper / build a skill"
│  └─ inspect → extract/ai_extract → create_skill → run_skill → verify
│  → Full guide: build-skill.md
│
├─ "Analyze this website / site audit"
│  └─ map → scrape homepage → crawl → screenshot → compile report
│  → Full guide: site-intel.md
│
├─ "Research a topic / find information"
│  └─ search (+ news_search) → select sources → readability → synthesize
│  → Full guide: research.md
│
├─ "Find APIs / reverse engineer"
│  └─ discover_apis → categorize → query_api → monitor_websocket
│  → Full guide: api-recon.md
│
├─ "Need page structure / interactive elements"
│  └─ snapshot → analyze ARIA refs → interact with ref targeting
│
├─ "Page behind login"
│  └─ interact (session_id) → login → then scrape/extract
│     └─ Or use chrome_profile for existing browser session
│
├─ "Suspect hidden API (SPA, dynamic)"
│  └─ discover_apis → query_api (10x faster than HTML scraping)
│  → Full guide: api-recon.md
│
├─ "Bulk URLs (10+)"
│  └─ batch_scrape (parallel, resumable, soft-fail)
│
└─ "Full site harvest"
   └─ map → batch_scrape (parallel fetch)
   → Full guide: site-intel.md
```

---

## Tool Combinations — 10 Patterns

Full CLI examples for each pattern → [pipelines.md](pipelines.md)

### 1. Inspect → Extract → Skill (Standard Path)
**When:** Known page with repeating elements
```
scrape(include: structured_data) → extract(selectors) → create_skill → run_skill (verify)
```

### 2. AI Auto-Discover → Skill (Unknown Structure)
**When:** Unknown page, let LLM figure it out
```
ai_extract(schema: "auto") → create_skill → run_skill (verify)
```

### 3. API-First → Query (Bypass HTML, 10x Faster)
**When:** SPA, trading platforms, hidden JSON APIs
```
discover_apis(wait: 10) → query_api(endpoint) → create_skill (optional)
```

### 4. Login → Session → Extract (Authenticated)
**When:** Content behind login wall
```
interact(login actions, session_id) → scrape/extract(protected page)
```
Alternative: `chrome_profile` param for existing browser cookies.

### 5. Search → Batch → AI Extract (Research Pipeline)
**When:** Research, competitor analysis, bulk collection
```
search(query) → batch_scrape(urls, extraction_schema) → job_status
```

### 6. Map → Batch (Full Site Harvest)
**When:** Site audit, content migration, full backup
```
map(max_urls: 200) → batch_scrape(urls, concurrency: 5)
```

### 7. WebSocket → API Reverse Engineering
**When:** Trading, chat, live dashboards
```
monitor_websocket(duration: 30) → discover_apis → query_api (REST fallback)
```

### 8. Screenshot → Debug → Refine (Visual Feedback)
**When:** Selectors not working
```
screenshot → extract(selectors) → [empty?] → screenshot → refine → extract
```

### 9. Skill Iteration Loop (Feedback)
**When:** ALWAYS — every skill needs validation
```
create_skill → run_skill → [bad?] → extract (manual) → create_skill (overwrite) → run_skill
```

### 10. Snapshot → Interact (Ref Targeting)
**When:** Need precise element targeting without fragile CSS selectors
```
snapshot(url) → analyze ARIA refs → interact(url, actions: [{ref: "N"}]) → snapshot(url) to verify
```

---

## Smart Scrape

*Use when: "scrape a website", "get content from URL", "read this article", "grab the text", "get product info"*

→ **Full guide: [smart-scrape.md](smart-scrape.md)**

### Quick Decision

| User intent | Tool path |
|-------------|-----------|
| Read article / get text | `readability` → if empty → `scrape` (markdown) |
| Get product info / extract prices | `scrape` (structured_data) → `extract` (CSS) → `llm_fallback` if empty |
| Unknown page structure | `ai_extract` (schema: "auto") |
| General scrape | `scrape` (markdown + metadata + links) |
| Visual capture | `screenshot` (full_page) |

### Stealth Escalation

| Situation | Action |
|-----------|--------|
| Simple static site | Default (auto-escalation handles it) |
| Known anti-bot (Cloudflare, Amazon, LinkedIn) | `stealth_level: 3` directly |
| JavaScript SPA (React, Angular, Vue) | `stealth_level: 3` (needs browser) |
| Empty/blocked results | `stealth_level: 3` + `proxy` |

### Key Tool Chains

- **readability → scrape** escalation
- **extract + llm_fallback**: CSS first, LLM if empty — best of both worlds
- **ai_extract → extract**: AI discovers structure → CSS for fast repeat runs
- **screenshot debug loop**: visual → extract → refine → extract

---

## Build Skill

*Use when: "create a scraper", "build an extractor", "make a skill", "save this pattern", "automate this scraping"*

→ **Full guide: [build-skill.md](build-skill.md)**

### Workflow Summary — 6 Steps

1. **Gather Requirements** — What data? Single/multiple pages? Login needed? Recurring?
2. **Page Inspection** — `scrape` with `include: [structured_data, links, metadata]`
3. **Pattern Detection** — CSS (`extract`) vs AI (`ai_extract`) vs Hybrid (`llm_fallback`)
4. **Create Skill** — `create_skill` with `url`, `name`, `description`
5. **Verify — MANDATORY** — `run_skill` on original + different page
6. **Educate User** — How to run, override URL, limit items, storage location

### Common Selectors

| Site Type | items_selector | Typical Fields |
|-----------|---------------|----------------|
| E-commerce | `.product-card`, `.item`, `[data-product]` | name, price, image, url, rating |
| News/Blog | `article`, `.post`, `.story` | title, date, author, excerpt, url |
| Job Board | `.job-listing`, `.vacancy` | title, company, location, salary, url |
| Directory | `.listing`, `.result` | name, address, phone, website, category |
| Table Data | `table tbody tr` | cell values by `td:nth-child(N)` |

### Built-in Recipes

Check before building custom skills → [recipes.md](recipes.md):
`hn-top-stories`, `github-trending`, `job-listings-greenhouse`, `ecommerce-product`, `product-reviews`, `crypto-websocket`, `news-article-reader`, `reddit-posts`, `seo-page-audit`, `social-media-mentions`

Run: `run_skill(name: "recipe-name")` / `imperium-crawl run-skill --name "recipe-name"`

---

## Site Intel

*Use when: "analyze this website", "map this site", "site audit", "what is this website about"*

→ **Full guide: [site-intel.md](site-intel.md)**

### Workflow Summary

1. **Site Mapping** — `map(url, max_urls: 100, include_sitemap: true)` → group URLs by section
2. **Homepage Deep-Dive** — `scrape(url, include: [structured_data, metadata, links])` → identity, tech stack
3. **Content Crawl** — `crawl(url, max_depth: 2, max_pages: 10)` → content types, quality
4. **Visual Capture** — `screenshot(url)` → design, layout
5. **Compile Report** — Overview, Structure, Content, Technology, Scraping Recommendations

### Depth Guidelines

| Request | map max_urls | crawl max_pages |
|---------|-------------|-----------------|
| Quick overview | 50 | 5 |
| Standard analysis | 100 | 10 |
| Deep audit | 200-500 | 20-30 |

### Bonus Steps

- **Batch harvest**: `map` → `batch_scrape(urls, concurrency: 5)` → full site data
- **API discovery**: `discover_apis(url, wait_seconds: 10)` → find hidden JSON endpoints
- **SEO extract**: `extract(url, selectors: {title, h1, meta_desc})` per page

---

## Research

*Use when: "research a topic", "find information about X", "deep dive into Z", "gather intel"*

→ **Full guide: [research.md](research.md)**

### Workflow Summary

1. **Decompose** — Core question, angles, recency needs, depth
2. **Formulate queries** — 2-3 targeted queries (primary, comparative, expert)
3. **Search** — `search(query, count: 10)` + optionally `news_search(query, freshness: "pw")`
4. **Select sources** — 3-5 URLs: authoritative, content-rich, diverse, recent
5. **Deep scrape** — `readability` first → fallback `scrape`. Or `batch_scrape` for 5+ sources
6. **Synthesize report** — Key Findings, Detailed Analysis, Sources with summaries

### Freshness Values

`pd` (past day), `pw` (past week), `pm` (past month), `py` (past year). Requires `BRAVE_API_KEY`.

### Depth Guidelines

| Request | Sources | Approach |
|---------|---------|----------|
| Quick | 1-2 | Single search, top results |
| Standard | 3-5 | 2 searches, curated |
| Deep dive | 5-8 | 3 searches + news |
| Competitive | 4-6 per competitor | Targeted per entity |

---

## API Recon

*Use when: "find APIs on a website", "discover hidden endpoints", "reverse engineer an API", "monitor WebSocket"*

→ **Full guide: [api-recon.md](api-recon.md)**

### Workflow Summary

1. **Discovery** — `discover_apis(url, wait_seconds: 8, include_headers: true)`
2. **Categorize** — By type (REST/GraphQL/WS), origin (1st/3rd party), auth (none/cookie/bearer/key/CSRF)
3. **Investigate** — `query_api` for GET endpoints, GraphQL introspection, pagination testing
4. **WebSocket** — `monitor_websocket(url, duration: 15-60)` → message format, types, auth
5. **Report** — Summary, API Inventory table, Detailed Analysis, Recommendations

### wait_seconds Tuning

| Site type | wait_seconds |
|-----------|-------------|
| Simple static | 5 |
| Standard web app | 8 |
| Heavy SPA | 12-15 |
| Infinite scroll | 15-20 |

### Key Tool Chains

- **discover → query → monitor** — full recon chain
- **interact (login) → discover** — authenticated API recon
- **websocket → query_api** — REST fallback (easier to automate)

---

## CLI Gotchas

**Boolean flags** — no value needed:
```bash
imperium-crawl screenshot --url URL --full-page     # correct
imperium-crawl screenshot --url URL --full-page true # wrong
```

**JSON params** — single quotes around JSON:
```bash
imperium-crawl extract --url URL --selectors '{"title":"h1","price":".cost"}'
```

**Output formats:**
```bash
--output-format json|csv|jsonl|markdown
--pretty          # human-readable JSON
--output file.json # write to file
```

**Pipe-safe:** Spinner → stderr, data → stdout:
```bash
imperium-crawl scrape --url URL --output-format json | jq '.title'
```

**Include:** `--include structured_data,links,metadata` (comma-separated, no spaces)

**Actions JSON:**
```bash
imperium-crawl interact --url URL --actions '[{"type":"click","selector":"button"},{"type":"wait","duration":2000}]'
```

**Batch URLs:** `--urls "url1,url2,url3"` (comma-separated in quotes)

**Freshness:** `--freshness pd` (pd=day, pw=week, pm=month, py=year)

---

## Error Recovery

| Error | Cause | Fix |
|-------|-------|-----|
| Empty content | JS-rendered or blocked | `stealth_level: 3` or `llm_fallback: true` |
| 403/Blocked | Anti-bot protection | `stealth_level: 3` + `proxy` |
| Timeout | Slow page / heavy JS | Increase `timeout`; try `readability` |
| Readability garbage | Non-article page | Fallback to `scrape` (markdown) |
| CAPTCHA | Bot detection | Auto-solved if `TWOCAPTCHA_API_KEY` set + Level 3 |
| Login required | Protected content | `interact` with session_id, or `chrome_profile` |
| Skill works, run_skill fails | Selectors changed | Re-inspect, recreate skill |
| `LLM_API_KEY` missing | ai_extract needs it | Run `imperium-crawl setup` |
| `BRAVE_API_KEY` missing | Search tools need it | Run `imperium-crawl setup` |
| CLI JSON parse error | Bad JSON | Validate JSON, single quotes, escape inner quotes |
| Batch job stalled | Network issues | Resume with same `job_id` (completed URLs skipped) |
| No APIs captured | Site needs interaction | Increase `wait_seconds` to 15-20 |
| All third-party APIs | Server-side rendering | Try HTML scraping instead |
| WebSocket refused | Needs auth cookies | `interact` login first, or `chrome_profile` |

---

## Environment Variables

| Variable | Required for | Purpose |
|----------|-------------|---------|
| `BRAVE_API_KEY` | search, news/image/video search | Brave Search API |
| `LLM_API_KEY` | ai_extract, llm_fallback, batch extraction | Anthropic or OpenAI |
| `LLM_PROVIDER` | — | `anthropic` (default), `openai`, `minimax` |
| `TWOCAPTCHA_API_KEY` | Auto CAPTCHA solving | 2Captcha API |
| `PROXY_URL` | — | Single proxy URL |
| `PROXY_URLS` | — | Comma-separated proxy rotation |
| `CHROME_PROFILE_PATH` | — | Chrome user data dir for authenticated sessions |
| `RESPECT_ROBOTS` | — | Honor robots.txt (default: `true`) |

Quick setup: `imperium-crawl setup` (interactive wizard)
登录后下载/点赞/收藏 ❤ 3 | ★ 0
imperium-crawl

目录结构

SKILL.md

举报内容

提示
