web-research
/web-research
Pobiera strony internetowe przez browser39, zapisuje jako markdown i indeksuje lokalnie, a potem czyta selektywnie narzędziami Julie, aby oszczędzać tokeny.
- Autor
- anortham ↗
- Repozytorium
- julie ↗
- Licencja
- MIT
- Ostatnia zmiana repozytorium
- 02.10.2026
Ocena i bezpieczeństwo
- Pliki w paczce
- 1
- Skrypty
- 0
- Komentarze HTML
- 0
- Adresy URL w SKILL.md
- 1
Brak sygnałów. Repozytorium zmienione 02.10.2026, bez odwołań do wycofanych modeli i API.
MIT, otwarta: można używać i zmieniać, z zachowaniem informacji o autorze.
Co robi i kiedy
Pobiera strony internetowe przez browser39, zapisuje jako markdown i indeksuje lokalnie, a potem czyta selektywnie narzędziami Julie, aby oszczędzać tokeny.
Oryginalny opis (EN)
Fetch web pages and index them locally for token-efficient research. Use when you need to read documentation, articles, or any web content. Fetches via browser39, saves as markdown, indexes via Julie's filewatcher, then uses Julie tools for selective reading. Requires browser39 (binary release from GitHub).
Jak działa
- Prerequisites
- Workflow
- Tips
Podgląd SKILL.md
---
name: web-research
description: Fetch web pages and index them locally for token-efficient research. Use when you need to read documentation, articles, or any web content. Fetches via browser39, saves as markdown, indexes via Julie's filewatcher, then uses Julie tools for selective reading. Requires browser39 (binary release from GitHub).
user-invocable: true
arguments: "<url or research topic>"
allowed-tools: mcp__julie__fast_search, mcp__julie__get_symbols, mcp__julie__get_context, Bash, Write
---
# Web Research
Fetch web pages, index them locally, and read selectively using Julie's tools. This replaces dumping entire web pages into context (which wastes thousands of tokens) with a workflow that indexes the content and lets you pull out just the sections you need.
## Prerequisites
browser39 must be installed. Check with:
```bash
which browser39
```
If missing, tell the user to download the latest binary release from [browser39 releases](https://github.com/alejandroqh/browser39/releases) and add it to their PATH.
## Workflow
### Step 0 (Optional): Discover URLs from a Topic
If the user gave a research topic rather than a specific URL, find candidate pages first:
```bash
browser39 search "your research topic" --output json > /tmp/b39-search.json
```
Parse the `links[]` array and pick the URLs worth fetching. Skip this step when the user already provided URLs.
### Step 1: Fetch and Save
Determine the target file path from the URL. The directory structure mirrors the URL:
```
docs/web/
docs.rs/axum/latest.md
developer.mozilla.org/Web/API/Fetch_API.md
github.com/tokio-rs/tokio.md
```
Fetch the page and save directly to the target file. Never print full page content to stdout.
```bash
# Fetch (rm ensures no stale results — --output appends, not overwrites)
echo '{"id":"1","action":"fetch","v":1,"seq":1,"url":"THE_URL","options":{"selector":"article","strip_nav":true,"include_links":true}}' > /tmp/b39-cmd.jsonl
rm -f /tmp/b39-out.jsonl
browser39 batch /tmp/b39-cmd.jsonl --output /tmp/b39-out.jsonl
# Extract markdown directly to file (no stdout leak)
mkdir -p docs/web/TARGET_DOMAIN/TARGET_PATH_DIR
python3 -c "
import sys,json
with open('/tmp/b39-out.jsonl') as f:
for line in f:
d=json.loads(line)
md = d.get('markdown','')
if md:
with open('docs/web/TARGET_DOMAIN/TARGET_PATH.md','w') as out:
out.write(md)
print(f'Saved {len(md)} chars to docs/web/TARGET_DOMAIN/TARGET_PATH.md')
"
```
If the page content is empty or too short, retry without the `selector` option, or try `"main"`, `".content"`, or `"body"`.
The filewatcher automatically indexes the file within 1-2 seconds.
### Step 2: Explore the Content
See the page structure (table of contents):
```
get_symbols(file_path="docs/web/{domain}/{path}.md", mode="structure")
```
This returns section headings as a hierarchy, letting you see what's on the page without reading it.
### Step 3: Read Selectively
**Read a specific section** by name:
```
get_symbols(file_path="docs/web/{domain}/{path}.md", mode="minimal", target="Section Name")
```
**Search across all fetched pages:**
```
fast_search(query="your search terms", file_pattern="docs/web/**")
```
**Get token-budgeted context** for a concept:
```
get_context(query="concept or question", file_pattern="docs/web/**")
```
### Step 4: Follow Links (Optional)
The fetched markdown contains links. If you need more information, fetch additional pages by repeating Steps 1-2 with the linked URLs. The agent decides which links are worth following based on the research goal.
### Step 5: Clean Up
When research is complete, suggest removing fetched content:
```bash
# Remove a specific page
rm docs/web/{domain}/{path}.md
# Remove all fetched content
rm -rf docs/web/
```
The filewatcher automatically removes deleted files from the index.
## Tips
- **GitHub README pages**: Use `selector: "article"` to get just the README content
- **Documentation sites**: Try `selector: "main"` or `selector: ".content"` to skip navigation
- **Large pages**: Use browser39's `max_tokens` option to limit the fetch, then paginate with `offset`
- **Multiple pages at once**: Write multiple commands to the JSONL file (one per line) for batch fetching
- **Searching across fetched content**: `file_pattern="docs/web/**"` scopes any Julie tool to just the fetched web content
- **Empty results from get_symbols**: The filewatcher indexes within 1-2 seconds. If results are empty, the file may not have been written correctly; check the file exists and has content.
Wersje
| Repozytorium | Wynik | Licencja |
|---|---|---|
| Ta strona · anortham/julie | 100 | MIT |
| TerminalSkills/skills | 100 | Apache-2.0 |
| hughluang/agent-skills | 100 | MIT |
| langchain-ai/deepagents | 100 | MIT |
| jeremylongshore/tons-of-skills-marketplace | 100 | MIT |
| pollinations/pollinations | 100 | MIT |
| joeblackwaslike/agent-skills | 100 | MIT |
| langchain-ai/deepagentsjs | 100 | MIT |
| criptogus/agent-evolve-network | 100 | Apache-2.0 |
Migawka bazy z 13.11.2026 · wersja paczki 4179cfab006c (skrót SHA-256 całej paczki)
Pobrania w ostatnich 30 dniach: 0 · ranking
Zgłoś problem z tym skillem: szkodliwa treść, błędny opis albo prośba autora o usunięcie.
Podobne skille
- 100web-fetch
Pobiera treści z sieci jako czysty markdown, preferując odpowiedzi natywnie w markdown, a przy ich braku wyciąga elementy z HTML selektorami. - 100editing
Kieruje zmiany w istniejących plikach do narzędzia edit_file Julie, bez pętli Read i Edit. - 100wigolo
Zbiera informacje z sieci lokalnie: wyszukuje, pobiera strony jako markdown, przeszukuje serwisy, wyciąga dane strukturalne i monitoruje zmiany stron przez serwer MCP wigolo. - 100fetch
Pobiera strony internetowe w zwięzłej formie z wyodrębnioną treścią, bez surowego HTML, do doraźnego researchu. - 90uwae
Pobiera stronę internetową i zwraca ją jako oczyszczony tekst Markdown, gdy trzeba przeczytać link lub podsumować witrynę. - 100efficient-web-research
Opisuje protokół researchu internetowego oszczędzającego tokeny przy dostępie do adresów URL, repozytoriów GitHub i zapytań wyszukiwarki, by uniknąć pobierania całych stron.