agent-reachlisted
Install: claude install-skill M4NUSH7/Niche-Claude-Code
# Agent Reach - internet research router
Source-acquisition layer for research: given a query or a URL, pick the right
free tool and run it. This skill never wraps or reimplements the underlying
tool - it calls `curl`, `gh`, `yt-dlp`, `feedparser`, or `mcporter` directly
and returns what they return.
**User-tier skill, not a harness component.** No pinned model, no hooks, no
project setup - it works in any repo or none.
**Platform: CLI edition.** Commands are POSIX (macOS / Linux / Git Bash on
Windows); where a native Windows shell differs, the alternative is given
inline.
<!-- PORT: for the Linux/Cowork edition - delete the "Windows notes" section,
the `powershell -File scripts/check_tools.ps1` line, and every `py -3`
alternative; `python3` is the only Python form there. -->
## agent-reach vs. deep-research
- **agent-reach** = breadth and speed. Fan out across platforms, pull raw
pages/threads/transcripts/repos, no synthesis.
- **deep-research** = depth and trust. Takes sources (including ones this
skill gathered), verifies adversarially, cites, writes the report.
For "research X": gather from 2-4 relevant platforms, then synthesize directly
or hand the combined material to deep-research if the user wants a rigorous,
cited writeup.
## Channels
Six channels. Only three need no setup at all, and two of the six can fail
while looking healthy - run the check (standing rule 1) before relying on one.
| Platform | Tool | Setup |
|----------|------|-------|
|