| name | decisions-agent-reach |
| description | Use with DecisionsAI when a ticket or workflow needs internet research — social posts, articles, video subtitles, GitHub, RSS, or competitor scans. Routes through Agent Reach upstream tools; run agent-reach doctor first. |
Decisions + Agent Reach
I am fetching external sources so the user can decide or ship with evidence — not inventing citations.
Agent Reach installs and routes upstream CLIs. Decisions adds workflow discipline: scope, evidence packets, and harness reporting.
When to invoke
- Ticket asks to research, compare competitors, read tweets/Reddit threads, summarize a YouTube/Bilibili video
- User pastes a URL from Twitter, GitHub, a blog, or RSS
- Content workflows (
content-engine, article-writing) need primary sources
- Not for: writing the final doc (use content skills), UI mockups (use
decisions-ui-ideation), or posting/commenting on social platforms
Workflow
- Run
agent-reach doctor --json — use each platform's active_backend.
- Fetch in parallel when useful (Exa + Twitter + Reddit + web via Jina).
- Save raw extracts under
/tmp/ or cite URLs — never write fetch artifacts into the project repo unless the ticket asks for committed research notes.
- In the Decisions result packet include:
Sources: bullet list with URLs
Method: platform + backend used
Summary: what matters for the ticket decision
Install / repair
If agent-reach is missing:
Install Agent Reach: https://raw.githubusercontent.com/Panniantong/agent-reach/main/docs/install.md
Or from Decisions reference clone (after bin/setup.py):
pip install -e ../reference/agent-reach
agent-reach install --env=auto
Safe mode on shared servers: agent-reach install --env=auto --safe
Skill pairing
| Task | Also use |
|---|
| UI informed by trends | decisions-ui-ideation, decisions-design-references |
| Article from research | content-engine, article-writing |
| Verify a shipped page | browser-qa, decisions-playwright |
| Minimal implementation | ponytail |
Full routing table
Read the projected agent-reach skill (references/*.md) — it is the source of truth for per-platform commands and retry chains.
Reference repo (local): ../reference/agent-reach from DecisionsAI parent folder.