agent-forage
min9813/agent-forage
Claude Code skill marketplace: send an agent out to forage for knowledge
Instalação
npx convez-ai min9813/agent-forage1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
README
A tradução para o português deste README ainda está na fila.
🦊 agent-forage
A Claude Code plugin marketplace of skills that send an agent out to forage for knowledge — fetching, reading, and transcribing content that the built-in tools can't reach on their own.
The idea: give the agent ways to go and read a specific source it doesn't yet have — on demand, one item at a time. These are retrieval tools, not bulk scrapers or dataset/AI-training pipelines.
Plugins
| Plugin | What it does | External auth |
|---|---|---|
| x-content-get | Fetch a single x.com / twitter.com post — full text, author, media URLs, embedded video, and a whisper transcript of that video. | None (uses free public mirrors) |
Install
# 1. Add this marketplace
/plugin marketplace add min9813/agent-forage
# 2. Install a plugin
/plugin install x-content-get@agent-forage
Then just paste an X post URL and ask the agent to read / quote / download / transcribe it.
x-content-get
Scope: fetches a single public post you already have the URL for, so an agent can read, quote, or transcribe it. It is a one-at-a-time retrieval helper — not a crawler, a bulk harvester, or a tool for building datasets or training models.
Plain WebFetch on x.com usually fails (HTTP 402 / JS wall), and the official X API costs
$200+/month. This skill instead reads the post through free embed mirrors
(api.fxtwitter.com, falling back to api.vxtwitter.com) and downloads media via the direct
MP4 link. For video, X almost never ships caption tracks, so a real transcript is produced by
running openai-whisper on the audio.
CLI
XGET="$(...)/skills/x-content-get/scripts/x_get.py" # ${CLAUDE_PLUGIN_ROOT} at runtime
python "$XGET" "https://x.com/<user>/status/<id>" # text + media URLs
python "$XGET" <id> --json | jq -r '.tweet.text' # raw mirror JSON
python "$XGET" "<url>" --download-video ./out # download the video
python "$XGET" "<url>" --transcribe --whisper-model small # download + transcribe
python "$XGET" "<url>" --via jina # post + thread, as Markdown
Two fetch paths: the default mirror (fxtwitter/vxtwitter) gives verbatim text and a
direct MP4 link (best for citation, download, transcription of a single post), while
--via jina uses Jina Reader to render the page and capture
the surrounding thread/replies as Markdown (best for reading context; no MP4 download).
Set JINA_API_KEY to raise Jina's rate limit (optional). Both avoid hitting x.com
directly, which is why they slip past tools that block it.
Requirements
- python3 (stdlib only) — text, metadata, media URLs, and video download.
- uvx (optional) — only for
--transcribe; pullsopenai-whisperon demand. - ffmpeg (optional) — only if you extract audio manually (see
references/).
Scope, terms & disclaimer
- Public posts only, no login. The tool reads through public embed mirrors and uses no account, login, or cookies. It does not access anything behind a login wall.
- Retrieval, not harvesting. It is built to fetch one post you already have the URL for. It is not a crawler and does not enumerate, bulk-collect, or build datasets, and it is not intended to gather data for training or fine-tuning AI models.
- Your responsibility. You are responsible for your own use, including compliance with X's Terms of Service and the copyright and rights of post authors when you read, save, or redistribute their content. Provided as is, for personal and research use, with no warranty (see LICENSE).
- Third-party mirrors.
fxtwitter/vxtwitterare independent, unofficial services that can change or go offline; this project is not affiliated with them or with X Corp. The script tries both and surfaces errors when neither responds. - Need search instead? For searching across X or summarizing many posts (rather than fetching one known post), use a Grok-backed search skill, not this one.
License
MIT — see LICENSE.