A survey of 52 hackathon projects integrating Supermemory found that
the patterns most frequently broken by AI codegen — wrong auth header,
deprecated containerTags plural, legacy /v3/search, and missing
containerTag scoping — were being actively recommended by these two
docs.
vibe-coding.mdx (the prompt users copy-paste into Claude/Cursor):
- Added a CANONICAL API SURFACE + DO NOT USE block before STEP 1 so
the agent sees the rules before any code example
- All x-supermemory-api-key → Authorization: Bearer
- containerTags: [userId] → containerTag: userId
- Repaired ~10 broken code fences (STEPs 3-8 were malformed; the
USER-ONLY APP example was empty)
- Note that rerank/rewriteQuery/filters are v4-search-only
- Note that client.memories.update/delete/forget are real but most
apps don't need them — memories are auto-extracted from documents
install.md (same prompt, separate doc):
- All 8 x-supermemory-api-key occurrences → Authorization: Bearer
with Content-Type where needed
Files in apps/docs/{memory-router,ai-sdk/infinite-chat,
model-enhancement,supermemory-mcp}/ are intentionally untouched —
those endpoints route off x-sm-user-id / x-sm-project so the headers
are required there.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
