🚀 Big win this week: **AGENTS.md-style docs hit 100% success in evals**, beating traditional “skills” — and we now have a concrete path to make tool/skill usage way more reliable.

**Key technical updates (Jan 27 – Feb 2, 2026):**
- 🧠 **Skill invocation reliability:** found a major gap (skills not triggering in ~56% of cases). **R0am shared a UserPromptSubmit hook** that forces a 3-step “evaluate → activate Skill() → execute” flow; **Stan confirmed similar patterns in Eliza Cloud**.
- 📄 **Docs > skills (for some tasks):** shared results showing **AGENTS.md formats outperform skills** (100% vs ~79% in one benchmark), pushing a rethink on how we guide agents.
- 🔐 **New tooling explored:** GAP (GitHub Actions Protocol) was introduced as a way to give agents secure account access **without sharing credentials**.
- ☁️ **ElizaCloud pain points flagged:** Proton email variants may be creating **duplicate accounts** (agents “disappearing” from dashboards); plus image gen can’t maintain **character consistency** across iterations.
- 🎨 **Proposed fix for visual consistency:** community suggested **LoRA-based “apps”/integrations** to preserve brand/character identity across generations.

**Community highlights:**
- 🙌 R0am’s forced-activation prompt pattern is already helping teams unblock eval reliability.
- 🛡️ Reminder: watch for migration scams — never share seed phrases (shoutout to community members calling these out fast).

**Token / auto.fun:**
- ⏳ **AI16Z → ELIZAOS migration deadline is Feb 3** — bridge issues (“max amount”, old tokens not detected) are being routed via support channels.
- 🧩 **auto.fun:** no major updates surfaced in discussions this week.

**Next up:** tighten skill descriptions + defaults, fix Cloud account duplication, prototype LoRA consistency flows, and share **post-migration** details (including the confirmed Babylon airdrop timing).