min-min
Auto-generated from project docs
North Star
Booklet Status
| Booklet | Status | Branch | Appetite | Deadline | Notes |
|---|---|---|---|---|---|
| B0-B7.1 | COMPLETE | N/A | — | — | OpenClaw era. Telegram, Discord, skills, autonomy, chat redesign |
| B8 | BLOCKED | booklet/min-min-b8 | 2 weeks | 2026-04-26 | Runtime migration: ZeroClaw live but webhook doesn't pass tools (INCIDENT-052). Decision needed: expand proxy / adopt nanobot / fork ZeroClaw |
| B9 | DONE | — | 1 week | 2026-04-21 | RESCOPED + COMPLETE Memory architecture (L0+L1+L2 injection + tool exec proxy). RESEARCH-250. |
| B9.5 | QUEUED | — | 3 days | — | Fix remaining broken capabilities (deferred from B9) |
| B9.6 | DECISION DONE, EXEC PENDING | — | 8 hrs exec | 2026-04-21 | RESEARCH-273 (2026-04-18): Expand minmin-proxy into full runtime (Option A). 11 deliverables A1-A11. Awaiting Mister go-ahead to execute. |
| B10 | QUEUED | — | 2 weeks | 2026-05-17 | Life Intelligence (morning briefing) |
| B11 | QUEUED | — | 2 weeks | 2026-05-31 | Client delivery automation |
Recent Decisions
| Date | Decision | Rationale |
|---|---|---|
| ------ | ---------- | ----------- |
| 2026-04-12 | Project kickstarted (rescope from OpenClaw workspace) | Never had proper project structure despite being core product |
| 2026-04-12 | Migrate OpenClaw → ZeroClaw | RESEARCH-226: 99% less RAM, 0 CVEs, native LiteLLM/Discord, migration tool exists |
| 2026-04-12 | Discord as primary channel (not Telegram) | Mister preference. Own full stack long-term (Chatii) |
| 2026-04-12 | OpenRouter as primary model (not Gemini) | Gemini too unreliable for agent tool calling. OpenRouter auto-top-up |
| 2026-04-12 | Monthly landscape review (QUICKREF-003) | AI agent space moving too fast to review annually |
| 2026-04-14 | Rescope B9: memory architecture replaces "fix broken things" | INCIDENT-052: tool-based memory fails with non-Claude models. RESEARCH-250: industry consensus is hybrid injection. Old B9 deliverables moved to B9.5 |
| 2026-04-14 | Switch default model: openrouter-auto → gemini-2.5-pro | OpenRouter auto-routing picked models bad at tool calling. Gemini Pro better but still doesn't call tools — confirms need for injection pattern |
| 2026-04-14 | Hybrid memory injection (L0+L1+L2) over tool-based memory | RESEARCH-250: every production agent system (MemPalace, Letta, Zep, GBrain) converged on this. Model-agnostic, 8 cents/day overhead |
| 2026-04-14 | Switch model: gemini-2.5-pro → gpt-4.1-mini | Gemini tool calling unreliable (documented Google forum issues, 5 LiteLLM bugs). GPT-4.1-mini: $0.40/1M input, 97-99% tool accuracy |
Source: /root/projects/min-min/