🔌 How Planning Chat Accesses the Substrate
Claude's web_fetch tool can only follow URLs that:
- Are explicitly given in user messages, OR
- Appear in pages already fetched
This means:
- ✅ Hub-linked pages: always accessible
- ✅ Pages linked from accessible pages: accessible via chain
- ❌ Files in folders without
index.html: invisible
- ❌ Files not linked from anywhere: invisible
The fix (deployed 2026-05-13):
- Every folder now has an
index.html (35 generated)
- Master /sitemap.html lists every file (~220 entries)
- Key files (coordination, etc.) linked directly from AI Hub
robots.txt at root scopes crawler access (bets public, personal private)
For SameI Tier 1: API access bypasses this entirely. This is one reason the API path matters — file-system access is the bottleneck of the planning-chat workflow.