-
Notifications
You must be signed in to change notification settings - Fork 2
Description
Bản tin Hệ sinh thái OpenClaw 2026-03-21
Issues: 500 | PRs: 500 | Dự án: 12 | Thời gian tạo: 2026-03-21 01:53 UTC
- OpenClaw
- NanoBot
- Zeroclaw
- PicoClaw
- NanoClaw
- IronClaw
- LobsterAI
- TinyClaw
- Moltis
- CoPaw
- ZeptoClaw
- EasyClaw
Phân tích sâu OpenClaw
Bản tin dự án OpenClaw — 2026-03-21
- Tổng quan hôm nay
- Hoạt động cộng đồng rất cao: 500 issue cập nhật và 500 PR cập nhật trong 24 giờ qua (mở/hoạt động: 393 issues; PR mở: 254, đóng/merge: 246).
- Không có phát hành mới. Dòng chính (main) đang nhận nhiều PR tính năng và sửa lỗi liên quan tới gateway, provider và trải nghiệm UI.
- Tâm điểm hiện tại là ổn định gateway/WebSocket, xác thực provider (Anthropic / Google / Ollama / OpenAI paths) và cải thiện UX (i18n, Control UI, cron/announce).
- Phát hành phiên bản
- Không có phiên bản mới trong ngày — mục này bỏ qua.
- Tiến độ dự án (PRs đã merge/đóng / hoạt động đáng chú ý)
- Trong 24 giờ có ~246 PR được đóng/merge; danh sách nổi bật (một số đã đóng/merge hoặc đang review):
- Fix iOS WebSocket ping crash: #51371 — fix(ios): prevent CheckedContinuation double-resume crash in WebSocket ping openclaw/openclaw#51371
- Cải thiện chọn model trong UI (preserve qualified refs) — đóng: #51367 — fix(ui): preserve qualified model refs in picker openclaw/openclaw#51367
- iOS: cải thiện QR pairing flow (device-pair) — #51359 — iOS: improve QR pairing flow openclaw/openclaw#51359
- Gateway stability bundle (OOM, failover 404, hook crash, handshake timeout) — #51362 (open, trọng tâm ổn định gateway) — fix: gateway stability — OOM, failover 404, hook crash, handshake timeout openclaw/openclaw#51362
- Microsoft Foundry provider (Entra ID auth) — #50378 — feat: Add Microsoft Foundry provider with Entra ID authentication openclaw/openclaw#50378
- Docker self-build fix sau migration extension — #48523 — docker: fix self-build breakage after extension migration openclaw/openclaw#48523
- Cron: clear stale running markers theo per-job timeout — #51024 — fix(cron): clear stale running markers using per-job timeouts openclaw/openclaw#51024
- Ollama model options & tool-call recovery — #51356 — fix(agents): accept model
optionsin config and safely recover Ollama tool calls (#50894) openclaw/openclaw#51356
- Ghi chú: nhiều PR liên quan đến providers, gateway và channel plugins đang mở — tập trung vào giảm số lỗi hồi quy của các release gần đây.
- Chủ đề nóng trong cộng đồng (issues/PRs có hoạt động nhiều nhất)
- Internationalization (i18n) & Localization Support — #3460 (106 bình luận) — Internationalization (i18n) & Localization Support openclaw/openclaw#3460
Phân tích: nhu cầu mạnh mẽ từ cộng đồng để hỗ trợ đa ngôn ngữ; hiện tại team báo không đủ băng thông — điều này gây áp lực cho UX toàn cầu và tài liệu. - Anthropic setup-token auth 401 Invalid bearer token — #23538 (23 bình luận) — [Bug] Anthropic setup-token auth returns 401 Invalid bearer token on 2026.2.21-2 (isolated profile repro) openclaw/openclaw#23538
Phân tích: lỗi xác thực runtime gây hỏng agent calls; tác động lớn với người dùng Anthropic. - Gateway Heartbeat timer stops after 1-2 triggers (introduced in v2026.3.8) — #45772 (20 bình luận) — Gateway Heartbeat timer stops after 1-2 triggers (introduced in v2026.3.8) openclaw/openclaw#45772
Phân tích: tính năng heartbeat không reschedule → giảm tính khả dụng của channel recovery; ảnh hưởng trực tiếp tới độ ổn định hoạt động. - macOS Gateway never becomes ready (Setup Wizard stuck) — #6156 (15 bình luận) — [Bug]: macOS Gateway never becomes ready (Setup Wizard stuck on Retry) openclaw/openclaw#6156
Phân tích: blocker trải nghiệm onboarding trên macOS (Apple Silicon) — nhiều người dùng bị chặn ngay bước cài đặt. - google-vertex ADC auth broken — #49191 (13 bình luận) — google-vertex ADC auth broken: "<authenticated>" sentinel passed as literal API key → 401 openclaw/openclaw#49191
Phân tích: ADC (Application Default Credentials) bị xử lý sai, gây 401 và ảnh hưởng tới khách hàng Google Vertex.
- Lỗi & ổn định — xếp hạng theo mức độ nghiêm trọng (cao → trung → thấp)
- Cao (production-impact / blocker)
- Gateway heartbeat timer stops (#45772) — Gateway Heartbeat timer stops after 1-2 triggers (introduced in v2026.3.8) openclaw/openclaw#45772
Ghi chú: ảnh hưởng trực tiếp tới recoverability; PR #51362 liệt kê nhiều bản vá stability liên quan (xem phần PRs). - WebSocket/CLI handshake & gateway closed errors (#47265, #46892 closed, #49510) — WebSocket CLI connection fails after upgrade to 2026.3.13 — handshake timeout on loopback openclaw/openclaw#47265 và Gateway WS handshake timeout (3s) too aggressive — causes spurious 'gateway closed (1000)' on busy event loops openclaw/openclaw#46892 và [Bug]: Error: Gateway closed (1000): openclaw/openclaw#49510
Ghi chú: CLI/WS bị lỗi sau upgrade → tác vụ vận hành bị gián đoạn; PR #51362 tuyên bố khắc phục handshake timeout. - Anthropic setup-token auth 401 (#23538) — [Bug] Anthropic setup-token auth returns 401 Invalid bearer token on 2026.2.21-2 (isolated profile repro) openclaw/openclaw#23538
Ghi chú: runtime token invalid → agent calls thất bại; chưa thấy PR fix. - macOS Gateway never becomes ready (#6156) — [Bug]: macOS Gateway never becomes ready (Setup Wizard stuck on Retry) openclaw/openclaw#6156
Ghi chú: onboarding blocker trên macOS; chưa thấy PR liên quan trong danh sách hiện tại.
- Gateway heartbeat timer stops (#45772) — Gateway Heartbeat timer stops after 1-2 triggers (introduced in v2026.3.8) openclaw/openclaw#45772
- Trung bình
- google-vertex ADC auth broken (#49191) — google-vertex ADC auth broken: "<authenticated>" sentinel passed as literal API key → 401 openclaw/openclaw#49191
- LLM request timeout ignores configured timeouts (#46049) — [Bug]: LLM request timeout ignores configured timeout settings (agent/channel/tool timeouts not respected) openclaw/openclaw#46049
- Switching to smaller-context model mid-session causes hard crash (#44303) — Switching to a smaller-context model mid-session causes hard crash instead of graceful truncation openclaw/openclaw#44303
- image tool ignores current imageModel / stale model state (#45221) — [Bug]: image tool ignores current imageModel and uses stale runtime image/fallback state openclaw/openclaw#45221
Ghi chú: PRs gần đây về model refs / Ollama (#51367, #51356, #49512) có liên quan và có thể giúp giảm tần suất.
- Thấp (UX / non-blocking / docs)
- UI glitches: context limit icon scaling (#44906) — [Bug]: UI Bug: Context limit warning icon scales abnormally and blocks chat view in v2026.3.12 openclaw/openclaw#44906
- STT mic blocked by Permissions-Policy header (#51085) — STT mic button broken: Permissions-Policy header blocks microphone access openclaw/openclaw#51085 — PR/doc fixes liên quan đang được mở.
- Yêu cầu tính năng & tín hiệu lộ trình
- Internationalization / Localization support (i18n) — #3460 — Internationalization (i18n) & Localization Support openclaw/openclaw#3460
- Nhu cầu mạnh, có khả năng trở thành mục ưu tiên khi có contributors/community volunteers.
- External Memory Provider API for zero-downtime context compaction — #49233 — Feature Request: External Memory Provider API for Zero-Downtime Context Compaction openclaw/openclaw#49233
- Yêu cầu một API mở cho memory offload — sẽ giảm blackout time trong compaction và cải thiện SLA agent.
- Community Skill Development & ClawHub (ecosystem cho skills) — #50090 — Community Skill Development & ClawHub openclaw/openclaw#50090
- Dấu hiệu rõ rệt: người dùng muốn hệ sinh thái skill giống marketplace — có thể là trọng tâm lộ trình cộng đồng.
- Microsoft Foundry provider PR (#50378) và OpenAI GPT-5.4 mini/nano discovery request (#50265) cho thấy nhu cầu mở rộng hỗ trợ providers và cập nhật model catalog:
- Tóm tắt phản hồi người dùng
- Điểm đau chính:
- Ổn định gateway & WebSocket: nhiều báo cáo gateway crash, handshake timeout, và gateway-closed (1000) khiến CLI và channels không đáng tin cậy.
- Xác thực provider: Anthropic/Google ADC/OAuth race/refresh vấn đề dẫn tới 401/refresh_token_reused và tác động trực tiếp đến khả năng gọi API.
- Onboarding/UX: macOS gateway setup stuck, mic/STT bị header chặn, và Control UI hiển thị refs sai gây nhầm lẫn.
- Tooling: exec/web_fetch failures và image tool stale state làm giảm khả năng sử dụng agent cho tác vụ thực tế.
- Mức độ hài lòng: cộng đồng tích cực đóng góp PR/issue nhưng bức xúc tập trung vào regressions sau các release gần đây; mong muốn cải thiện ổn định trước khi thúc đẩy tính năng lớn.
- Theo dõi tồn đọng (cần chú ý của maintainers)
- i18n & localization discussion (long-running, nhiều comments) — #3460 — Internationalization (i18n) & Localization Support openclaw/openclaw#3460
- Đề xuất: triage thành epic + community-run localization sprints hoặc docs-first approach.
- Anthropic auth 401 (runtime) — #23538 — [Bug] Anthropic setup-token auth returns 401 Invalid bearer token on 2026.2.21-2 (isolated profile repro) openclaw/openclaw#23538
- Cần investigation reproduce/trace và regression bisect; ưu tiên cao.
- Gateway heartbeat timer (#45772) — Gateway Heartbeat timer stops after 1-2 triggers (introduced in v2026.3.8) openclaw/openclaw#45772
- Cần test kế hoạch reschedule / thêm metrics & e2e scenario; PR #51362 có thể liên quan: fix: gateway stability — OOM, failover 404, hook crash, handshake timeout openclaw/openclaw#51362
- macOS Gateway never ready (onboarding) — #6156 — [Bug]: macOS Gateway never becomes ready (Setup Wizard stuck on Retry) openclaw/openclaw#6156
- Cần triage logs, thêm fallback/backoff trong setup wizard.
- google-vertex ADC auth (#49191) — google-vertex ADC auth broken: "<authenticated>" sentinel passed as literal API key → 401 openclaw/openclaw#49191
- Cần patch để không nhầm sentinel "" làm API key; priority cao cho Google users.
- External Memory Provider API proposal (#49233) — Feature Request: External Memory Provider API for Zero-Downtime Context Compaction openclaw/openclaw#49233
- Nếu mục tiêu sản phẩm là giảm blackout time, nên đưa vào roadmap như alpha design.
Kết luận & khuyến nghị ngắn
- Ưu tiên ngắn hạn: (1) ổn định gateway & WebSocket/CLI (triage + cherry-pick PR #51362 và test), (2) investigate/auth fixes cho Anthropic & Google ADC (401s), (3) fix onboarding blockers trên macOS.
- Trung hạn: chuẩn hóa provider/model discovery (Ollama/OpenAI fixes có PR đang review), cải thiện cron/announce reliability (PR #51024 đã mở), và đưa i18n thành epic cộng đồng.
- Cộng đồng sôi động nhưng cần tập trung hơn vào triage và regression-testing để giảm số issue hồi quy xuất hiện sau release.
Tài nguyên & liên kết nhanh (chỉ chọn mục đã nêu)
- #3460 i18n: Internationalization (i18n) & Localization Support openclaw/openclaw#3460
- #23538 Anthropic auth: [Bug] Anthropic setup-token auth returns 401 Invalid bearer token on 2026.2.21-2 (isolated profile repro) openclaw/openclaw#23538
- #45772 Gateway heartbeat: Gateway Heartbeat timer stops after 1-2 triggers (introduced in v2026.3.8) openclaw/openclaw#45772
- #6156 macOS Gateway setup: [Bug]: macOS Gateway never becomes ready (Setup Wizard stuck on Retry) openclaw/openclaw#6156
- #49191 google-vertex ADC auth: google-vertex ADC auth broken: "<authenticated>" sentinel passed as literal API key → 401 openclaw/openclaw#49191
- PR #51362 gateway stability: fix: gateway stability — OOM, failover 404, hook crash, handshake timeout openclaw/openclaw#51362
- PR #51371 iOS WebSocket ping fix: fix(ios): prevent CheckedContinuation double-resume crash in WebSocket ping openclaw/openclaw#51371
- PR #51367 preserve qualified model refs (UI): fix(ui): preserve qualified model refs in picker openclaw/openclaw#51367
- PR #51356 Ollama model options: fix(agents): accept model
optionsin config and safely recover Ollama tool calls (#50894) openclaw/openclaw#51356 - PR #51024 cron stale-run fix: fix(cron): clear stale running markers using per-job timeouts openclaw/openclaw#51024
Nếu muốn, tôi có thể:
- chuẩn bị danh sách PR ưu tiên đề xuất merge theo impact / easiness of review; hoặc
- soạn checklist regression test (gateway, auth, model discovery, cron) để maintainers chạy CI/manual.
So sánh hệ sinh thái chéo
- Tổng quan hệ sinh thái
- Hệ sinh thái mã nguồn mở cho AI agent/trợ lý cá nhân đang rất sôi động và phân mảnh: vài dự án quy mô lớn (OpenClaw, IronClaw) tập trung vào gateway, provider ecosystem và khả năng tích hợp enterprise; nhiều dự án trung bình (NanoBot, PicoClaw, NanoClaw, LobsterAI, CoPaw) tập trung vào kênh nhắn tin, mô hình local và UX; một số nhỏ/đặc thù (TinyClaw, Moltis, ZeptoClaw, EasyClaw) nhắm vào on‑ramp, embedded, hoặc thảo luận kiến trúc.
- Điểm chung: nhu cầu mạnh về độ ổn định gateway/WebSocket, tương thích provider (auth/token), observability/tracing, và hỗ trợ mô hình local / hot-reload; đồng thời nhiều cộng đồng đang mở rộng tích hợp providers (Gemini/Bedrock/Foundry/Ollama).
- So sánh mức độ hoạt động
- Dưới đây là bảng tóm tắt ngắn (số liệu lấy từ bản tin 2026-03-21):
| Project | Issues (24h) | PRs (24h) | Release (gần nhất) | Sức khỏe (tổng quan) |
|---|---|---|---|---|
| OpenClaw | ~500 (393 open active) | ~500 (254 open, 246 merged/closed) | No release (main active) | Rất năng động; nhiều regressions → Active but risky |
| IronClaw | 39 | 50 | v0.21.0 (2026-03-20) | Năng động; có CRITICAL security/migration issues → High‑risk active |
| CoPaw | ~50 (mix issues/PRs) | ~23 merged/closed | v0.1.0.post1 → bump | Năng động; nhiều runtime compaction issues |
| NanoBot | 38 (20 open) | 66 (37 open) | No release | Nhiều PR, focus local models & channels; active |
| PicoClaw | 23 | 61 (32 open) | nightly v0.2.3‑nightly | Active; nightly builds, channel/provider fixes |
| LobsterAI | 33 | 30 (19 open) | 2026.3.20 | Active but runtime/gateway instability reported |
| NanoClaw (qwibitai) | 7 | 25 (23 open) | No release | Active feature expansion (multi‑tenant, channels); DB migration risk |
| Moltis | 3 | 5 | No release | Moderate activity; onboarding/installer issues |
| TinyClaw | 0 | 5 | v0.0.16 | Low volume, release focused on onboarding; stable |
| ZeptoClaw | 2 | 0 | No release | Low activity; design/infra discussion stage |
| EasyClaw | 1 | 0 | v1.7.3 | Very low activity; small userbase |
(Ghi chú: “Issues/PRs (24h)” là số cập nhật/tracking trong 24 giờ theo từng bản tin.)
- Vị thế của OpenClaw
- Ưu thế: OpenClaw là điểm tập trung lớn nhất về cộng đồng và đóng góp (hàng trăm issue/PR trong 24h), có kiến trúc gateway/channel/provider rất mở, nhiều plugin provider (Anthropic, Google, Ollama, Microsoft Foundry…) và tính năng UX/Control UI phong phú — điều này tạo lợi thế mạng lưới (ecosystem effect) so với dự án nhỏ.
- Khác biệt kỹ thuật: OpenClaw đặt trọng tâm vào gateway/WebSocket, provider abstraction và hệ sinh thái skills/plugins (ClawHub ý tưởng), trong khi nhiều dự án khác chuyên vào một miền (ví dụ NanoBot: local models & channels; IronClaw: workflows/approvals/WASM tools).
- Quy mô cộng đồng: lớn nhất trong danh sách → nhanh test & nhiều PR nhưng cũng dễ sinh regressions; cần quy trình triage/release chặt hơn so với các dự án nhỏ có vòng đóng gói thận trọng.
- Hướng kỹ thuật chung (yêu cầu lặp lại, kèm dự án)
- Gateway / WebSocket stability: OpenClaw (heartbeat, handshake), LobsterAI (gateway restart), PicoClaw (cron/agent stability) — cần robust heartbeat, reschedule, better metrics and e2e tests for channels.
- Provider auth & token handling: OpenClaw (Anthropic/Google ADC 401), IronClaw (Google OAuth blocked), NanoBot/PicoClaw (OpenRouter/Ollama auth issues) — yêu cầu: refresh/token race handling, clearer error surfaces, retry/backoff.
- Local LLM & Ollama compatibility: NanoBot (Ollama qwen2.5 tool fails), OpenClaw (Ollama model options), PicoClaw (local model connectors) — cần better config examples, more CI for local endpoints, and robust tool-call recovery.
- Observability / tracing & eval harness: NanoBot (LLM TRACE), IronClaw (debug panel), CoPaw/NanoBot (eval harness) — cần standardized traces, request/response logs, deterministic eval harness for regressions.
- Cron/agent reliability & memory compaction: OpenClaw (cron stale markers), CoPaw (context compaction cancels tasks), PicoClaw (cron silent discard) — cần safe compaction APIs, per-job timeouts, fail-safe for long‑running tasks.
- Hot-reload & model switching safety: LobsterAI (IM hot-reload), PicoClaw, OpenClaw (model refs) — need atomic swap semantics and state consistency to avoid stale calls/double outputs.
- Packaging/installer & migration safety: IronClaw (migration checksum), Moltis (Windows installer), NanoClaw (pre-flight credential checks) — need migration scripts, fail-fast preflight, and clear upgrade notes.
- Phân tích khác biệt hóa (trọng tâm, người dùng, kiến trúc)
- OpenClaw: trọng tâm ecosystem & gateway; user: operators/teams muốn multi‑provider, plugin skills; kiến trúc: gateway + provider plugin, heavy community contributions.
- IronClaw: enterprise workflows, approvals, WASM tools; user: doanh nghiệp/teams dùng routines/approval flows; kiến trúc: workspace-driven, focus on security/race handling.
- NanoBot: local-first, multi-channel integration (Feishu/WhatsApp/Telegram) và tracing; user: deployers muốn local models / channel bots.
- PicoClaw: edge/embedded + channel crypto (Matrix E2EE), focus trên device/SELinux/aarch64; user: edge/IoT/embedded deployments.
- LobsterAI: runtime tuned for IM integrations and OpenClaw runtime tweaks; user: IM-centric deployments (Feishu, QQ) with Windows focus.
- NanoClaw (qwibitai): multi‑tenant, API usage tracking, persistent graph memory; user: orgs needing tenancy & billing.
- CoPaw: agent memory compaction, model selection and stop controls; user: developers who run long‑running agents and need memory management.
- TinyClaw / Moltis / ZeptoClaw / EasyClaw: niches—onboarding simplicity, provider additions (Gemini), infra/test harness or small CLI apps respectively.
- Động lực & độ trưởng thành của cộng đồng
- Tier A — High velocity & large community: OpenClaw (very high PR/issue volume; needs stronger triage), IronClaw (active, release cadence, but CRITICAL CI findings), CoPaw (fast iterations, many fixes).
- Tier B — Active, feature-driven: NanoBot, PicoClaw, LobsterAI, NanoClaw (qwibitai) — đều có nhiều PRs, kênh tích hợp, đang mở rộng tính năng; thường gặp lỗi tích hợp/compatibility.
- Tier C — Moderate/steady or niche: Moltis, TinyClaw — có releases/PRs nhưng quy mô nhỏ hơn; tập trung vào UX/on‑ramp.
- Tier D — Low activity / design stage: ZeptoClaw, EasyClaw — ít PR, tập trung thảo luận/bug triage.
- Ghi chú: dự án lớn lặp nhanh (OpenClaw) nhưng dễ trả giá bằng regressions; dự án vừa/một số nhỏ giữ vòng release chậm hơn nhưng ít regressions.
- Tín hiệu xu hướng & khuyến nghị cho nhà phát triển AI agent
- Xu hướng rút ra:
- Đa dạng hoá providers (Gemini/Bedrock/Foundry/Ollama) → cần provider abstraction và model catalog management.
- Hỗ trợ local LLMs và recovery for tool-calls (Ollama, LiteLLM) → local-first deployments đang tăng.
- Observability, deterministic eval & trace là yêu cầu thiết yếu để debug và CI.
- Memory offload / external memory provider APIs để giảm blackout time khi compacting context.
- Hot-reload model semantics, safe swap, và preflight validation (credentials/migrations) để giảm downtime.
- Multimodal (TTS/ASR, WhatsApp multimodal) và channel stability (long connections, dedup) là cần thiết cho sản phẩm hoàn chỉnh.
- Khuyến nghị hành động cho teams/devs:
- Ưu tiên: robust provider auth (refresh/race-free), gateway/WS e2e tests, and observability hooks (trace IDs per request).
- Thiết kế: provider adapter + capability discovery + atomic model swap API để hỗ trợ hot-reload an toàn.
- CI/QA: bổ sung deterministic eval harness + fixture runner cho tools (giảm regressions khi thay provider/LLM).
- Ops: preflight credential checks, migration scripts & backups, and per-job timeouts/clear stale markers for cron.
- Cộng đồng: nếu duy trì dự án lớn (như OpenClaw), cần đầu tư team triage & release managers để giảm regressions khi tốc độ merge cao.
Kết luận ngắn
- Hệ sinh thái đang phát triển nhanh, có hai trục chính: (1) mở rộng provider/ecosystem và gateway tại các dự án lớn; (2) hỗ trợ local LLMs, channel stability và observability tại các dự án trung bình. Nhà phát triển/ra quyết định kỹ thuật nên đầu tư vào ba nền tảng kỹ thuật chung: provider auth & discovery, gateway/WebSocket reliability (với e2e tests), và observability/eval harness — những đầu tư này sẽ giảm chi phí vận hành và tăng khả năng tích hợp cho hầu hết các dự án trong hệ sinh thái.
Báo cáo các dự án cùng nhóm
NanoBot — HKUDS/nanobot
NanoBot — Bản tin dự án (2026-03-21)
- Tổng quan hôm nay
- Hoạt động cộng đồng cao: trong 24 giờ qua có 38 issue được cập nhật (20 mở/đang hoạt động, 18 đã đóng) và 66 PR được cập nhật (37 mở, 29 đã merge/đóng). Điều này cho thấy nhịp độ phát triển và hỗ trợ người dùng vẫn rất sôi động.
- Không có bản phát hành mới hôm nay. Nhóm đóng/merge nhiều PR tiện ích/fix; nhiều đề xuất tính năng và vấn đề về kênh tích hợp/local model vẫn đang thảo luận.
- Xu hướng chính: cải thiện độ ổn định của kênh (Feishu/WhatsApp/DingTalk/Telegram), hỗ trợ mô hình local (Ollama, LiteLLM...), và thêm cơ chế quan sát/đánh giá (tracing / eval harness).
- Phát hành phiên bản
- Không có phiên bản mới trong ngày (Không có release được ghi nhận).
- Tiến độ dự án (PR đã merge/đóng hôm nay & điểm nổi bật)
- Một số PR đã đóng/merge trong vòng 24h:
- Add Rust nanobot MVP — PR #2313 (CLOSED) — thêm crate Rust MVP.
- docs: add github copilot oauth setup — PR #2310 (CLOSED) — tài liệu OAuth Copilot; loại bỏ trường cấu hình không cần thiết.
- fix: add openssh-client and use HTTPS for GitHub in Docker build — PR #1911 (CLOSED).
- feat(cron): add run history tracking — PR #1838 (CLOSED).
- feat(tools): enhance ExecTool with enable flag and deny_patterns — PR #1824 (CLOSED).
- Enable web search provider and support tavily — PR #2012 (CLOSED).
- feat(channels): add base64 media upload for QQ and WeCom — PR #2306 (CLOSED).
- Add files via upload — PR #2299 (CLOSED).
- Các luồng chức năng đang được đẩy mạnh (mở/PR to-nightly):
- Hệ thống hook / event-driven hooks + skill disable — PR #1934 (OPEN).
- Multi-provider web search + fallback — PR #2113 (OPEN).
- Agent evaluation harness (deterministic eval) — PR #2283 (OPEN).
- Feishu topic/thread reply support — PR #2307 (OPEN).
- Cron: system_event execution — PR #2311 (OPEN).
- Nhiều PR tập trung cải thiện kênh (Flask so_reuse patch cho WeCom — PR #2308 open) và giảm log spam Telegram — PR #2272 (OPEN).
- Chủ đề nóng trong cộng đồng (issues/PR hoạt động nhiều)
- 🌐 AI 官方内容追踪报告 2026-03-19 duanyytop/agents-radar#215 [CLOSED] 飞书无法建立长连接 — 飞书无法建立长连接 HKUDS/nanobot#215 (24 bình luận)
Vấn đề: kết nối dài (long-poll / long-connect) với Feishu; nhiều người gặp khó khi cấu hình app id/app secret. Nhu cầu: hướng dẫn cấu hình/khả năng tự phục hồi kết nối. - 📈 AI Open Source Trends 2026-03-19 duanyytop/agents-radar#218 [CLOSED] feat(whatsapp): 支持"发给自己"功能 — feat(whatsapp): 支持"发给自己"功能 - 删除 fromMe 消息过滤 HKUDS/nanobot#218 (23 bình luận)
Vấn đề: lọc tin từ chính người dùng (fromMe) gây khó khăn khi người dùng muốn "gửi cho chính mình" để test. Nhu cầu: tùy chọn bật/tắt lọc. - 📊 Bản tin hàng ngày công cụ AI CLI 2026-03-23 #75 [CLOSED] How Do I content the Ollama — How Do I content the Ollama HKUDS/nanobot#75 (19 bình luận)
Vấn đề: cấu hình / tích hợp Ollama và local models. Nhu cầu: ví dụ cấu hình, xử lý token/endpoint local. - #1922 [OPEN] I've created [nanobot-webui] — I've created [nanobot-webui], A self-hosted web management panel for nanobot HKUDS/nanobot#1922 (6 bình luận, +6 👍)
Yêu cầu: web UI self-hosted để quản lý provider/channel/skills — tín hiệu rõ ràng về nhu cầu UX quản trị. - #2154 [OPEN] 建议支持LLM TRACE — 建议支持LLM TRACE (轨迹可观测),方便运行期间问题的溯源解决,最好能是接口方式,支持多种方式的trace HKUDS/nanobot#2154 (5 bình luận)
Yêu cầu: tracing/observability cho luồng LLM (để truy vết lỗi runtime).
- Lỗi & Ổn định — báo cáo hôm nay, xếp hạng theo mức độ ảnh hưởng
- Mức độ nghiêm trọng: Cao
- #2293 [OPEN] Ollama local deployment of qwen2.5:14b – conversations silently die when executing tasks — Ollama local deployment of qwen2.5:14b – conversations silently die when executing tasks HKUDS/nanobot#2293 (1 bình luận)
Tác động: agent dừng khi gọi tool với mô hình local — làm gián đoạn khả năng thực thi tác vụ. Chưa có fix. - #2185 [OPEN] Upgrade nanobot from 0.1.4 to 0.1.4post5 breaks gemini-3-flash-preview — Upgrade nanobot from 0.1.4 to 0.1.4post5 breaks the usage of gemini-3-flash-preview HKUDS/nanobot#2185 (3 bình luận)
Tác động: hồi quy tương thích provider/model — cần workaround / fix nhanh.
- #2293 [OPEN] Ollama local deployment of qwen2.5:14b – conversations silently die when executing tasks — Ollama local deployment of qwen2.5:14b – conversations silently die when executing tasks HKUDS/nanobot#2293 (1 bình luận)
- Mức độ nghiêm trọng: Trung bình
- #2235 [OPEN] Telegram responses show twice — Telegram responses show twice HKUDS/nanobot#2235 (3 bình luận, +6 👍)
Tác động: trải nghiệm người dùng bị nhiễu; có PR để giảm log spam (#2272) nhưng chưa giải quyết hoàn toàn. - #2288 [OPEN] Intermitten Openrouter API Exception — Intermitten Openrouter API Exception HKUDS/nanobot#2288 (1 bình luận)
Tác động: gọi LLM thất thường gây lỗi JSON từ OpenRouter/LiteLLM. - #2298 [OPEN] Breaking endless tool calling loops — Breaking endless tool calling loops HKUDS/nanobot#2298 (2 bình luận)
Tác động: vòng lặp gọi tool vô tận — ảnh hưởng đến mô hình nhỏ/local; cần giới hạn/backs-off logic.
- #2235 [OPEN] Telegram responses show twice — Telegram responses show twice HKUDS/nanobot#2235 (3 bình luận, +6 👍)
- Mức độ nghiêm trọng: Thấp
- #1418 [OPEN] DingTalk channel: Agent cannot see sender name — DingTalk channel: Agent cannot see sender name, only staffId HKUDS/nanobot#1418 (1 bình luận)
- #1864 [OPEN] DingTalk 用户侧不支持上传文件 — DingTalk用户侧不支持上传文件 HKUDS/nanobot#1864 (3 bình luận)
- Ghi chú về PR sửa lỗi / đang chờ: PR #2308 (Fix Flask port reuse for WeCom) đang mở; PR #2272 giảm log spam Telegram; PR #1824 đã thêm deny_patterns cho ExecTool (đã merge) giúp an toàn lệnh.
- Yêu cầu tính năng & tín hiệu lộ trình
- Observability / Trace: #2154 yêu cầu mở rộng tracing LLM — nhiều người muốn interface pluggable cho trace/record. (link: 建议支持LLM TRACE (轨迹可观测),方便运行期间问题的溯源解决,最好能是接口方式,支持多种方式的trace HKUDS/nanobot#2154)
- Đánh giá agent / CI-safe eval: PR #2283 (agent evaluation harness) — nếu merge sẽ cung cấp khung đo lường hành vi agent (link: feat(eval): add agent evaluation harness with built-in scenarios HKUDS/nanobot#2283).
- Web UI quản lý: Issue #1922 (nanobot-webui) có phản hồi tích cực — khả năng tích hợp GUI quản trị là tín hiệu rõ ràng (link: I've created [nanobot-webui], A self-hosted web management panel for nanobot HKUDS/nanobot#1922).
- Multimodal native agent & base64 media upload: PR #2304 (multimodal sensory) và PR #2306 (base64 media upload, đã merge) chỉ ra roadmap hướng tới xử lý media tốt hơn cho kênh.
- Tool safety / loop prevention: #2298 — mong muốn giới hạn vòng lặp và logic remediate.
- Tóm tắt phản hồi người dùng
- Điểm đau chính:
- Tính phức tạp cấu hình provider/local model (Ollama, LiteLLM, OpenRouter) gây lỗi runtime — nhiều issue hướng đến ví dụ cấu hình và tương thích. (vd. 📊 Bản tin hàng ngày công cụ AI CLI 2026-03-23 #75, #2293, #2288)
- Kênh nhắn tin (Feishu/WeCom/WhatsApp/DingTalk/Telegram) có nhiều tình huống cạnh: long connection failures (🌐 AI 官方内容追踪报告 2026-03-19 duanyytop/agents-radar#215), file upload lỗi (#909, #1864), trả lời đôi trên Telegram (#2235). Người dùng mong muốn tích hợp ổn định, retry, và cấu hình đơn giản.
- An toàn tool và vòng lặp vô hạn: người dùng báo agent gọi tool lặp lại khi dùng mô hình nhỏ/local — mong muốn backoff / max retries / detection. (#2298)
- Mức độ hài lòng: hỗn hợp — người dùng đánh giá cao tính năng mới (onboard wizard #2018/#2019 được đón nhận), nhưng bức xúc với các lỗi tích hợp và mô hình local.
- Theo dõi tồn đọng (cần chú ý từ maintainers)
- Issues quan trọng chưa xử lý / chờ phản hồi:
- #2293 Ollama local tool execution silent fail — Ollama local deployment of qwen2.5:14b – conversations silently die when executing tasks HKUDS/nanobot#2293
- #2298 Breaking endless tool calling loops — Breaking endless tool calling loops HKUDS/nanobot#2298
- #2154 LLM TRACE request — 建议支持LLM TRACE (轨迹可观测),方便运行期间问题的溯源解决,最好能是接口方式,支持多种方式的trace HKUDS/nanobot#2154
- #2235 Telegram responses show twice — Telegram responses show twice HKUDS/nanobot#2235
- #1922 nanobot-webui (integration/collab) — I've created [nanobot-webui], A self-hosted web management panel for nanobot HKUDS/nanobot#1922
- #2250 onboard CLI -c/--config option bug — nanobot onboard with (--config or -c) says no such option HKUDS/nanobot#2250
- PRs cần review / prioritise:
- PR #2283 feat(eval): agent evaluation harness — feat(eval): add agent evaluation harness with built-in scenarios HKUDS/nanobot#2283
- PR #1934 feat(hooks): event-driven hook system — feat(hooks): event-driven hook system with skill disable support (#1932) HKUDS/nanobot#1934
- PR #2113 feat(web_search): multiple search providers fallback — feat(web_search): support multiple search providers with automatic fallback HKUDS/nanobot#2113
- PR #2304 feat(agent): multimodal sensory capabilities — feat(agent): implement native multimodal autonomous sensory capabilities HKUDS/nanobot#2304
- PR #2307 feat(feishu): thread reply support — feat(feishu): add thread reply support for topic group messages HKUDS/nanobot#2307
- PR #2311 feat(cron): system_event execution — feat(cron): implement system_event execution for cron jobs HKUDS/nanobot#2311
- PR #2272 fix(telegram) log throttling — fix(telegram): log network errors as warnings without stacktrace HKUDS/nanobot#2272
- Gợi ý ưu tiên cho maintainers:
- Ưu tiên fix các vấn đề làm gián đoạn execution (Ollama/tool-exec, endless loops, provider regressions).
- Tăng cường hướng dẫn cấu hình cho local models và kênh để giảm lượng issue lặp lại.
- Xem xét roadmap cho tracing + eval harness để cải thiện khả năng debug và CI/qa.
Tài nguyên tham chiếu nhanh (vài link tiêu biểu)
- Issue 🌐 AI 官方内容追踪报告 2026-03-19 duanyytop/agents-radar#215 (Feishu long connection): 飞书无法建立长连接 HKUDS/nanobot#215
- Issue 📈 AI Open Source Trends 2026-03-19 duanyytop/agents-radar#218 (WhatsApp "send to self"): feat(whatsapp): 支持"发给自己"功能 - 删除 fromMe 消息过滤 HKUDS/nanobot#218
- Issue #1922 (nanobot-webui): I've created [nanobot-webui], A self-hosted web management panel for nanobot HKUDS/nanobot#1922
- Issue #2293 (Ollama local tool fail): Ollama local deployment of qwen2.5:14b – conversations silently die when executing tasks HKUDS/nanobot#2293
- Issue #2298 (endless tool loops): Breaking endless tool calling loops HKUDS/nanobot#2298
- PR #2283 (eval harness): feat(eval): add agent evaluation harness with built-in scenarios HKUDS/nanobot#2283
- PR #1934 (hooks): feat(hooks): event-driven hook system with skill disable support (#1932) HKUDS/nanobot#1934
Kết luận ngắn: dự án đang rất năng động, cộng đồng góp ý nhiều về tính năng quan sát/đánh giá, cải tiến kênh và hỗ trợ mô hình local. Ưu tiên kỹ thuật hiện tại nên là ổn định hoá execution path (tool calls / local LLM), giảm hồi quy provider, và hoàn thiện observability để giảm thời gian debug cho người dùng.
Zeroclaw — zeroclaw-labs/zeroclaw
PicoClaw — sipeed/picoclaw
PicoClaw — Bản tin dự án (2026-03-21)
- Tổng quan hôm nay
- Hoạt động cộng đồng rất cao: 24 giờ qua có 23 issue hoạt động (20 mở/đang thảo luận, 3 đóng) và 61 PR được cập nhật (32 mở, 29 merge/đóng).
- Dòng phát triển chính đang tập trung vào độ ổn định của cron/agent, tương thích provider (OpenAI-compat/Gemini) và tăng cường kênh (Matrix/WhatsApp/QQ).
- Có một nightly build tự động mới (v0.2.3-nightly.20260321.100720bb); đây là bản build thử, có thể không ổn định — dùng với thận trọng.
- Nhìn chung: hoạt động phát triển/PR mạnh, nhiều sửa lỗi quan trọng đang chờ review/merge — sức khỏe dự án tích cực nhưng vẫn còn một số vấn đề ổn định cần ưu tiên.
- Phát hành phiên bản
- Phiên bản mới (nightly): v0.2.3-nightly.20260321.100720bb (automated nightly build).
Lưu ý: đây là bản build tự động và có thể không ổn định; người dùng triển khai vào môi trường sản xuất nên đợi bản release chính thức.
Full changelog: sipeed/picoclaw@v0.2.3...main
- Tiến độ dự án (PRs merge/đóng & nổi bật hôm nay)
- Số PR được merge/đóng trong 24 giờ: ~29 (tổng 61 PR được cập nhật). Một số PR đã đóng/merge:
- #1818 — fix(agent): separate empty-response and tool-limit fallbacks (CLOSED). Giải quyết thông báo fallback sai khi model trả về empty response. fix(agent): separate empty-response and tool-limit fallbacks sipeed/picoclaw#1818
- #1837 — docs: fix inaccuracies, add translations, and expand channel docs (CLOSED). Cải thiện tài liệu, dịch thuật và doc các channel. docs: fix inaccuracies, add translations, and expand channel docs sipeed/picoclaw#1837
- #1819 — docs: fix inaccuracies and add missing translations (CLOSED). docs: fix inaccuracies and add missing translations sipeed/picoclaw#1819
- #1687 — feat: upload local QQ attachments before sending (CLOSED). Cải thiện gửi media trên QQ. feat: upload local QQ attachments before sending sipeed/picoclaw#1687
- #1835 — sửa lỗi DNS trên Android (CLOSED). 修改android下的dns问题 sipeed/picoclaw#1835
- PR mở đáng chú ý đang chờ review hoặc tiếp tục phát triển:
- #1841 — feat(matrix): support encrypted messages with E2EE (OPEN). Thêm cấu hình crypto và giải mã event
m.room.encrypted. feat(matrix): support encrypted messages with E2EE sipeed/picoclaw#1841 - #1843 — Feat/whatsapp native multimodal (OPEN). Bổ sung hỗ trợ đa phương tiện cho WhatsApp. Feat/whatsapp native multimodal sipeed/picoclaw#1843
- #1844 — fix(agent) scope steering (OPEN). Scoped Steering để tránh inject message chéo giữa session. fix(agent) scope steering sipeed/picoclaw#1844
- #1847 — fix(providers): honour request_timeout for CLI providers (OPEN). Bảo vệ subprocess khỏi chạy vô hạn. fix(providers): honour request_timeout for CLI providers with clear timeout errors and fallback sipeed/picoclaw#1847
- #1846 — feat: add new compose file for local build and selinux enabled (OPEN). Docker/SELinux friendly compose. feat: add new compose file for local build and selinux enabled sipeed/picoclaw#1846
- #1841 — feat(matrix): support encrypted messages with E2EE (OPEN). Thêm cấu hình crypto và giải mã event
- Chủ đề nóng trong cộng đồng (những issue/PR nhiều thảo luận nhất)
- #1648 — [Feature] Adding TTS and ASR Support to PicoClaw (16 bình luận). Người dùng muốn kiến trúc audio (TTS/ASR) để có tương tác giọng nói; đã có PR liên quan (#1642) nhưng chưa tích hợp vào gateway. Link: [Feature] Adding TTS and ASR Support to PicoClaw sipeed/picoclaw#1648
Phân tích: nhu cầu rõ ràng cho voice I/O — ảnh hưởng lớn đến trải nghiệm người dùng cuối, đặc biệt trên thiết bị biên. - 📈 Bản tin xu hướng AI mã nguồn mở 2026-03-12 #28 — Feat Request: LM Studio Easy Connect (9 bình luận). Yêu cầu giao tiếp đơn giản với LM Studio (đặc biệt trên Android). Link: Feat Request: LM Studio Easy Connect sipeed/picoclaw#28
Phân tích: nhiều người muốn kết nối dễ dàng tới hệ sinh thái LLM cục bộ; rào cản kỹ thuật hiện chưa thân thiện cho người dùng phổ thông. - #806 — [Feature]: Add webUI support (6 bình luận). Đề xuất phát triển Web UI để giảm rào cản cho người không chuyên. Link: [Feature]: Add webUI support (Refactoring now) sipeed/picoclaw#806
Phân tích: Web UI là tính năng “on-ramp” quan trọng để tăng lượng người dùng không-CLI. - Các thread liên quan đến cron/agent/llm failures (ví dụ #629, #1058) và config/provider (ví dụ #1491, #1790) cũng thu hút sự chú ý kỹ thuật.
- Lỗi & Ổn định — xếp hạng theo mức độ nghiêm trọng
- Mức độ nghiêm trọng: Cao
- #629 — [BUG] Didn't retry if LLM call failed. Agent task treo nếu backend trả lỗi (impact: tác vụ dài/cron bị kẹt). Link: [BUG] Didn't retry if meet LLM call failed sipeed/picoclaw#629
- #1790 — [BUG] openrouter free not working (model ID không hợp lệ gây 400). Ảnh hưởng trải nghiệm provider OpenRouter. Link: [BUG] openrouter free not working sipeed/picoclaw#1790
- #1491 — [BUG] Picoclaw fails to load config.json, "no API key configured for model: glm-4.7". Cản trở cấu hình provider. Link: [BUG]
Picoclaw fails to load config.json, persistently errors with "no API key configured for model: glm-4.7"sipeed/picoclaw#1491 - #1058 — [BUG] Cron deliver=false tasks silently discard LLM response — messages never reach channel. Tác vụ cron xử lý nhưng không giao kết quả cho người dùng. Link: [BUG] Cron deliver=false tasks silently discard LLM response — messages never reach channel sipeed/picoclaw#1058
- Tiến triển / PR liên quan: #1839 và #1842 (PRs mở bởi @securityguy) nhằm sửa lỗi cron (route cron jobs -> correct agent, reload store on external file change / only save when state changes). fix(cron): route cron jobs to correct agent and publish response to channel sipeed/picoclaw#1839 và fix(cron): reload store on external file change; only save when state changes sipeed/picoclaw#1842
- Mức độ nghiêm trọng: Trung bình
- #1763 — aarch64 .deb not install (cài .deb thất bại trên aarch64). Link: [BUG] aarch64 .deb not install sipeed/picoclaw#1763
- #1825 — Container keep running while the app stopped on SIGINT/SIGTERM. Vấn đề quản lý container/restart policy. Link: [BUG] Container keep running while the app stopped on receiving SIGINT or SIGTERM sipeed/picoclaw#1825
- #1824 — Agent run heartbeat example behaves unexpectedly. Link: [BUG] Agent run heartbeat from example in HEARTBEAT.md sipeed/picoclaw#1824
- PR #1847 đang sửa request_timeout cho CLI providers (giải quyết subprocess chạy vô hạn). fix(providers): honour request_timeout for CLI providers with clear timeout errors and fallback sipeed/picoclaw#1847
- Mức độ nghiêm trọng: Thấp
- Các lỗi UI/UX, docs, i18n, small channel features (ví dụ #1737 doc missing port, #1830 docs i18n).
- Yêu cầu tính năng & tín hiệu lộ trình
- Tính năng được đề nghị nhiều / có sức hút cao:
- TTS/ASR audio support — #1648 (voice I/O). [Feature] Adding TTS and ASR Support to PicoClaw sipeed/picoclaw#1648
- Web UI để tiếp cận người dùng phổ thông — #806. [Feature]: Add webUI support (Refactoring now) sipeed/picoclaw#806
- OpenAI API-compatible channel (HTTP API) để dễ tích hợp — #1738. [Feature] 添加OpenAI API 格式的channel支持 sipeed/picoclaw#1738
- Event-driven Hooks System (trigger-based automation) — #1796. Feature: Event-driven Hooks System sipeed/picoclaw#1796
- Matrix E2EE và WhatsApp multimodal (PRs #1841, #1843) đang tiến hành.
- Dự đoán tính năng sắp được đưa vào release tiếp theo:
- Các sửa lỗi cron/agent và cải thiện provider compatibility có khả năng ưu tiên (PR #1839, #1842, #1847, #1460, #1342).
- Hỗ trợ kênh mã hóa Matrix (#1841) và WhatsApp multimodal (#1843) có xác suất vào bản release nếu được hoàn thiện và review nhanh.
- Tóm tắt phản hồi người dùng
- Điểm đau chính:
- Khó cấu hình provider / API key mismatch (ví dụ lỗi với glm-4.7, OpenRouter model IDs). (#1491, #1790)
- Tác vụ nền (cron/agent) đôi khi không giao kết quả hoặc treo do lỗi provider / không retry — ảnh hưởng đến tính tin cậy cho workflows tự động. (#629, #1058)
- Thiếu Web UI khiến người dùng non-CLI gặp rào cản, nhiều đề nghị xây Web UI. (#806)
- Docker / SELinux / packaging (aarch64 .deb) gây khó khăn cho deploy trên môi trường khác nhau. (#1763, #1846)
- Cảm nhận: cộng đồng năng động, nhiều đóng góp về docs và i18n (đã có PR cải thiện tài liệu lớn), nhưng người dùng mong muốn cải thiện độ ổn định, trải nghiệm cấu hình và các kênh giao tiếp.
- Theo dõi tồn đọng (cần maintainer chú ý)
- Issues/PRs lâu cần review / quyết định:
- #1648 (TTS/ASR) — 16 bình luận, có PR liên quan chưa tích hợp: [Feature] Adding TTS and ASR Support to PicoClaw sipeed/picoclaw#1648
- 📈 Bản tin xu hướng AI mã nguồn mở 2026-03-12 #28 (LM Studio Easy Connect) — mở từ 2026-02-11, nhiều người quan tâm: Feat Request: LM Studio Easy Connect sipeed/picoclaw#28
- #806 (WebUI) — đề xuất roadmap quan trọng, cần chỉ đạo/triage: [Feature]: Add webUI support (Refactoring now) sipeed/picoclaw#806
- PRs quan trọng chờ review: #1841 (Matrix E2EE), #1843 (WhatsApp multimodal), #1847 (CLI providers timeout), #1842/#1839 (cron fixes), #1844 (scoped steering).
- #1841: feat(matrix): support encrypted messages with E2EE sipeed/picoclaw#1841
- #1843: Feat/whatsapp native multimodal sipeed/picoclaw#1843
- #1847: fix(providers): honour request_timeout for CLI providers with clear timeout errors and fallback sipeed/picoclaw#1847
- #1842: fix(cron): reload store on external file change; only save when state changes sipeed/picoclaw#1842
- #1839: fix(cron): route cron jobs to correct agent and publish response to channel sipeed/picoclaw#1839
- #1844: fix(agent) scope steering sipeed/picoclaw#1844
- Đề xuất hành động nhanh cho maintainers:
- Ưu tiên review/merge các PR sửa lỗi cron/agent và provider timeout (giảm số vấn đề nghiêm trọng).
- Triager các feature requests lớn (TTS/ASR, WebUI) để quyết định scope/roadmap.
- Tăng cường hướng dẫn cài đặt cho aarch64 / SELinux / docker-compose mới để giảm friction cho deployers.
Kết luận ngắn gọn
- PicoClaw đang trong giai đoạn phát triển năng động: nhiều PR chất lượng (docs, provider fixes, channel features) và các issue tập trung vào hai trục chính: (1) độ ổn định/độ tin cậy của agent & cron, (2) trải nghiệm tích hợp kênh và UI. Nếu maintainer ưu tiên xử lý một nhóm PR hot (cron fixes, provider timeouts, Matrix E2EE), trải nghiệm người dùng sẽ cải thiện rõ rệt trong bản release tiếp theo.
Tài nguyên tham chiếu nhanh
- Nightly build / changelog: sipeed/picoclaw@v0.2.3...main
- Issue TTS/ASR: [Feature] Adding TTS and ASR Support to PicoClaw sipeed/picoclaw#1648
- Feature LM Studio: Feat Request: LM Studio Easy Connect sipeed/picoclaw#28
- WebUI: [Feature]: Add webUI support (Refactoring now) sipeed/picoclaw#806
- Cron bug (silently discard): [BUG] Cron deliver=false tasks silently discard LLM response — messages never reach channel sipeed/picoclaw#1058
- Cron PRs: fix(cron): route cron jobs to correct agent and publish response to channel sipeed/picoclaw#1839 , fix(cron): reload store on external file change; only save when state changes sipeed/picoclaw#1842
- Matrix E2EE PR: feat(matrix): support encrypted messages with E2EE sipeed/picoclaw#1841
- WhatsApp multimodal PR: Feat/whatsapp native multimodal sipeed/picoclaw#1843
Nếu bạn muốn, tôi có thể soạn mẫu câu trả lời/ghi chú cho maintainers để thúc đẩy review những PR ưu tiên.
NanoClaw — qwibitai/nanoclaw
-
Tổng quan hôm nay
NanoClaw đang rất hoạt động: trong 24 giờ qua có 7 issue được cập nhật (4 còn mở, 3 đã đóng) và 25 PR được cập nhật (23 mở, 2 đã đóng/merged). Không có phát hành mới. Hoạt động tập trung vào mở rộng kênh (Signal, White Noise, Twilio WhatsApp, Proton), tính năng quản lý mô hình/run-time và tích hợp bộ nhớ bền vững; đồng thời có các vấn đề ổn định khởi động và di trú dữ liệu cần chú ý. Nhìn chung sức khỏe dự án là tích cực về tính năng nhưng có vài điểm ổn định/DB cần ưu tiên sửa chữa. -
Phát hành phiên bản
Không có phiên bản mới trong ngày — phần này bỏ qua. -
Tiến độ dự án (PRs đã merge/đóng hôm nay & tính năng nổi bật)
- PR đóng/merged trong 24h: 2 PR được đóng (chi tiết: #1298 — web channel, dashboard, multi-tenant — đóng: feat: Add web channel, dashboard, and multi-tenant support qwibitai/nanoclaw#1298; #1203 — honour ANTHROPIC_BASE_URL path — đóng: fix: honour ANTHROPIC_BASE_URL path for third-party API endpoints qwibitai/nanoclaw#1203).
- Tính năng được đẩy mạnh (PRs lớn đang chờ review):
- Theo dõi sử dụng API: #1111 (api usage tracking) — adds usage DB và phân loại usage (feat: add API usage tracking qwibitai/nanoclaw#1111).
- Kênh mới: Signal (#1057, feat(skill): add Signal messenger channel qwibitai/nanoclaw#1057), White Noise (#1059, feat(skill): add White Noise channel (MLS + Nostr) qwibitai/nanoclaw#1059), Twilio WhatsApp (#1294, feat: add Twilio WhatsApp channel qwibitai/nanoclaw#1294), Proton Mail via Bridge (#1117, feat(skill/proton): complete Proton suite — 36 MCP tools (Mail, Pass, Drive, Calendar, VPN) qwibitai/nanoclaw#1117).
- Nâng cấp trải nghiệm agent: per-group persona từ mô tả nhóm WhatsApp (#1292, skill: add-group-persona — per-group agent personas from WhatsApp group description qwibitai/nanoclaw#1292), runtime /model command (#1205, feat: add /model command for runtime model switching qwibitai/nanoclaw#1205), CLI skill
claw(#1296, skill: claw — run NanoClaw agents from the command line qwibitai/nanoclaw#1296). - Memory & infra: persistent graph memory (mem0 + Qdrant + Neo4j) mà không cần container mới (#1256, feat: add /add-mem0-graph skill — persistent memory with Qdrant + Neo4j + local embeddings qwibitai/nanoclaw#1256).
- Ổn định khởi động: PR để thêm pre‑flight credential validation vào docker-entrypoint.sh (#1290) nhằm khắc phục trạng thái khởi tạo không hoàn chỉnh (fix for #1289) (fix: add pre-flight credential validation to docker-entrypoint.sh qwibitai/nanoclaw#1290).
- Chủ đề nóng trong cộng đồng (issues/PRs hoạt động nhất)
- API usage tracking (PR #1111): mong muốn theo dõi token/cost/model per-invocation để kiểm soát chi phí và báo cáo — tín hiệu rằng tổ chức hoá billing/usage là nhu cầu lớn (feat: add API usage tracking qwibitai/nanoclaw#1111).
- Web channel + dashboard + multi-tenant (PR #1298, đã đóng): thay đổi lớn về mặt mặt điều khiển và multi-tenant; nếu được merge sẽ thay đổi cách triển khai — cần rà soát bảo mật và race conditions (feat: Add web channel, dashboard, and multi-tenant support qwibitai/nanoclaw#1298).
- Per-group persona từ mô tả WhatsApp (issue #1291, PR #1292): người dùng muốn cấu hình persona động qua UI WhatsApp, là tính năng nâng cao UX cho multi-group deployments (issue: feat: per-group agent personas from WhatsApp group description qwibitai/nanoclaw#1291, PR: skill: add-group-persona — per-group agent personas from WhatsApp group description qwibitai/nanoclaw#1292).
- Startup / pre-flight credential validation (issue #1289, PR #1290): người dùng báo rằng khởi chạy container thiếu credential để lại trạng thái dữ liệu hỏng — cần fail-fast (issue: feat: add pre-flight credential validation on startup qwibitai/nanoclaw#1289, PR: fix: add pre-flight credential validation to docker-entrypoint.sh qwibitai/nanoclaw#1290).
- Lỗi & Ổn định (xếp hạng theo mức độ ảnh hưởng)
- Cao: DB migration đánh dấu sai Telegram DMs là nhóm — có nguy cơ phá vỡ phân loại cuộc trò chuyện sau backfill (#1272) (open) — liên kết: Bug: DB migration incorrectly marks Telegram direct chats as is_group=1 qwibitai/nanoclaw#1272. Ưu tiên sửa vì ảnh hưởng dữ liệu dài hạn.
- Trung bình-cao: IC01 relay nhận lệnh từ phòng IC00 do dùng chung tài khoản operator — có thể gây phản hồi nhầm phòng (#1284) — đã được đóng (BUG: IC01 relay processes commands from IC00 rooms via shared operator account qwibitai/nanoclaw#1284). Kiểm tra xem fix đã được triển khai trên production.
- Trung bình: Khởi động thiếu credential tạo state bán phần — báo cáo #1289; có PR sửa kèm (#1290) đang chờ review (issue: feat: add pre-flight credential validation on startup qwibitai/nanoclaw#1289, PR: fix: add pre-flight credential validation to docker-entrypoint.sh qwibitai/nanoclaw#1290).
- Thấp: Tài liệu Linux support mâu thuẫn (doc vs website) — cần cập nhật docs (#1075) (Mismatch in Linux Support Documentation qwibitai/nanoclaw#1075).
Ghi chú: một số issue nhỏ đã được đóng hôm nay (#1288 đọc reaction đã đóng, #1291 đóng) — kiểm tra changelog/commits để xác định liệu đã merge hay chỉ đóng.
- Yêu cầu tính năng & tín hiệu lộ trình
- Quan trọng: theo dõi usage / billing (PR #1111) — khả năng cao sẽ được đưa vào roadmap để hỗ trợ tổ chức hoá chi phí.
- Kênh & tích hợp chính thức: Twilio WhatsApp (#1294), Signal (#1057), Proton Bridge (#1117) cho thấy hướng đi hỗ trợ kênh “official API” và private-messaging.
- UX quản trị: dashboard/web multi-tenant (#1298) và runtime model switch (/model #1205) đều là tín hiệu rõ ràng cho nhu cầu quản lý multi-group, multi-model.
- Agent behaviour: per-group persona (#1292) và emoji reaction read (#1288) phản ánh nhu cầu cá nhân hoá và tương tác nhạy bén hơn trong nhóm.
Dự đoán: bản tiếp theo sẽ ưu tiên ổn định khởi động/DB migration, thêm theo dõi usage, và một số kênh chính thức (Twilio/Signal).
- Tóm tắt phản hồi người dùng
- Điểm đau chính: khởi chạy khi thiếu cấu hình làm rối dữ liệu (mức độ không hài lòng cao với trải nghiệm dev/ops) — issue #1289.
- Trường hợp sử dụng nổi bật: triển khai nhiều nhóm/kênh, cần phân quyền/tenancy, theo dõi chi phí cho từng agent/group. Người dùng muốn quản lý mô hình runtime và persona per-group để dùng trong nhiều bối cảnh (family vs work).
- Phản hồi tích cực: cộng đồng đóng góp nhiều skill/kênh (Signal, Proton, White Noise, OpenMail, A2A) — cho thấy người dùng thích khả năng mở rộng và tích hợp với nhiều hệ sinh thái.
- Theo dõi tồn đọng (cần chú ý của bảo trì)
- Issue DB migration Telegram is_group bug (#1272) — ưu tiên cao: Bug: DB migration incorrectly marks Telegram direct chats as is_group=1 qwibitai/nanoclaw#1272.
- PRs lớn chờ review / merge: #1111 (api usage tracking) feat: add API usage tracking qwibitai/nanoclaw#1111; #1057 (Signal) feat(skill): add Signal messenger channel qwibitai/nanoclaw#1057; #1117 (Proton) feat(skill/proton): complete Proton suite — 36 MCP tools (Mail, Pass, Drive, Calendar, VPN) qwibitai/nanoclaw#1117; #1256 (mem0 graph memory) feat: add /add-mem0-graph skill — persistent memory with Qdrant + Neo4j + local embeddings qwibitai/nanoclaw#1256; #1294 (Twilio WhatsApp) feat: add Twilio WhatsApp channel qwibitai/nanoclaw#1294; #1293 (IPC media support) feat: add IPC media attachment support with container path translation qwibitai/nanoclaw#1293.
- Issue tài liệu Linux mismatch (#1075) — dễ xử lý, cải thiện trải nghiệm: Mismatch in Linux Support Documentation qwibitai/nanoclaw#1075.
- Kiểm tra trạng thái fix cho relay-command-ngược (#1284) — confirm đã triển khai: BUG: IC01 relay processes commands from IC00 rooms via shared operator account qwibitai/nanoclaw#1284.
Đề xuất hành động: review & prioritise PR #1290 (pre-flight validation) và #1272 (DB migration), sau đó đẩy review cho #1111 và các kênh mới, đồng thời chạy một checklist regression trước khi merge PRs lớn (multi-tenant, web channel).
Kết luận ngắn gọn: hoạt động đóng góp cao với nhiều PR mở triển vọng mở rộng kênh và tính năng quản lý, nhưng cần ưu tiên hai vấn đề ổn định: lỗi migration Telegram và hành vi khởi động khi thiếu credential để tránh hỏng dữ liệu. Các maintainer nên tập trung review PR bảo mật/ổn định trước khi merge các PR tính năng lớn.
IronClaw — nearai/ironclaw
Tổng quan hôm nay
- Hoạt động cộng đồng rất cao: 39 issue và 50 PR có cập nhật trong 24 giờ qua, nhiều thay đổi lớn đang được thảo luận và vài PR/issue có độ rủi ro bảo mật/độ ưu tiên cao.
- Phiên bản v0.21.0 vừa phát hành (2026-03-20) mang theo vài cải tiến quan trọng (embedding cache, structured fallback deliverables, webhook relay), nhưng đồng thời xuất hiện một số lỗi và vấn đề tương thích cần theo dõi.
- Dòng công việc hiện ưu tiên: ổn định luồng phê duyệt (approval threads), cải thiện trải nghiệm web/UX, và hoàn thiện tải/kiểm tra schema cho WASM tools — nhiều PR lớn (XL) đang chờ review.
- Tổng kết: sức khỏe dự án năng động nhưng có vài cảnh báo bảo mật/khả năng hỏng dữ liệu cần xử lý sớm.
Phát hành phiên bản
- v0.21.0 — 2026-03-20. Những thay đổi chính (từ ghi chú phát hành):
- Thêm structured fallback deliverables cho job bị fail / bị kẹt (PR 🌐 AI 官方内容追踪报告 2026-03-21 duanyytop/agents-radar#236).
Link PR: feat: structured fallback deliverables for failed/stuck jobs nearai/ironclaw#236 - Thêm LRU embedding cache để tăng tốc tìm kiếm workspace (PR #1423).
Link PR: feat: LRU embedding cache for workspace search nearai/ironclaw#1423 - Hỗ trợ nhận relay events qua webhook callbacks (PR #1254, ghi chú bị cắt trong nguồn).
Link PR: feat: receive relay events via webhook callbacks nearai/ironclaw#1254
- Thêm structured fallback deliverables cho job bị fail / bị kẹt (PR 🌐 AI 官方内容追踪报告 2026-03-21 duanyytop/agents-radar#236).
- Lưu ý di chuyển / tương thích:
- Ghi chú phát hành không nêu breaking changes rõ ràng, nhưng có báo cáo migration/database liên quan (xem Issue #1328) — khuyến nghị kiểm tra bản sao lưu DB trước khi nâng cấp.
- Link release tag: (xem trang releases của repo để tải/vérify).
Tiến độ dự án (PRs đã merge/đóng / thay đổi đáng chú ý)
- PRs đóng / merged gần đây:
- feat(webhooks): public webhook trigger endpoint for routines — PR #736 (CLOSED).
feat(webhooks): add public webhook trigger endpoint for routines nearai/ironclaw#736 - Conversation history loading & batching — PR #465 (CLOSED).
feat(agent): add conversation history loading and message batching nearai/ironclaw#465 - WASM tool schema fallback extraction — PR #343 (CLOSED).
fix(wasm): extract tool schema/description with fallback nearai/ironclaw#343 - Docker build fix (COPY build.rs) — PR #332 (CLOSED).
[fix] Dockerfile was missing build.rs in COPY call, Cargo.toml test for … nearai/ironclaw#332 - CLI logs v2 — PR #527 (CLOSED).
Feat(cli) logs v2 nearai/ironclaw#527 - Bootstrap .env cleanup (remove redundant LLM config) — PR #1448 (CLOSED).
fix(setup): remove redundant LLM config and API keys from bootstrap .env nearai/ironclaw#1448
- feat(webhooks): public webhook trigger endpoint for routines — PR #736 (CLOSED).
- PRs mở, có tác động lớn đang chờ review:
- #1397 — Toàn diện: coercion & validation cho oneOf/anyOf/allOf schemas (quan trọng cho WASM tools).
fix: parameter coercion and validation for oneOf/anyOf/allOf schemas nearai/ironclaw#1397 - #1277 — UX overhaul lớn (web design system, onboarding, REPL polish).
feat(ux): complete UX overhaul — design system, onboarding, web polish nearai/ironclaw#1277 - #1470 / #1471 — Routines: notification normalization & bounded retry cho lỗi tạm thời.
fix(routines): normalize notification summaries with truncation and metadata nearai/ironclaw#1470 | fix(routines): add bounded retry for transient lightweight failures nearai/ironclaw#1471 - #1495 / #1497 — Sửa lỗi liên quan cross-channel approval và giảm lock contention (liên quan các cảnh báo staging CI).
fix(agent): prevent cross-channel approval hijacking nearai/ironclaw#1495 | fix(agent): reduce approval thread lock contention nearai/ironclaw#1497
- #1397 — Toàn diện: coercion & validation cho oneOf/anyOf/allOf schemas (quan trọng cho WASM tools).
Chủ đề nóng trong cộng đồng
- Google OAuth bị Google chặn trên Google Suite tools — Issue #902 (ảnh hưởng trải nghiệm tích hợp Google, nhiều báo cáo về “This app is blocked”).
Google OAuth blocked across Google Suite tools + request optional gws fallback provider nearai/ironclaw#902- Nhu cầu: chẩn đoán scope/sensitive-scope, fallback provider cho GWS, hướng dẫn cấu hình OAuth.
- Migration checksum mismatch khi nâng cấp lên v0.19.0 — Issue #1328 (nguy cơ upgrade thất bại trên DB hiện hữu).
Upgrade to v0.19.0 fails: V6__routines migration checksum mismatch on existing PostgreSQL databases nearai/ironclaw#1328- Nhu cầu: hướng dẫn migration fix/rollback, script khôi phục checksum, thông báo upgrade-safe.
- Cảnh báo bảo mật/độ ổn định được staging CI phát hiện (approval hijacking, TOCTOU, lock contention) — Issues #1485, #1486, #1489 (auto-created).
[CRITICAL] Cross-channel approval thread hijacking - authorization bypass nearai/ironclaw#1485 | [CRITICAL] TOCTOU race condition in approval thread resolution nearai/ironclaw#1486 | [MEDIUM] Lock contention in approval thread resolution nearai/ironclaw#1489- Nhu cầu: review bảo mật nhanh, harden channel-scoped approvals, race fixes.
- Lỗi tương tác với Gemini (function call thiếu thought_signature) báo cáo trên v0.21.0 — Issue #1510 (mới).
BUG: Gemini: function call is missing a thought_signature in functionCall parts. nearai/ironclaw#1510
Lỗi & Ổn định (xếp theo mức độ nghiêm trọng)
- CRITICAL
- Cross-channel approval thread hijacking — Issue #1485 (CRITICAL, confidence 100%) (staging CI). Requires immediate attention.
[CRITICAL] Cross-channel approval thread hijacking - authorization bypass nearai/ironclaw#1485 - TOCTOU race condition in approval thread resolution — Issue #1486 (CRITICAL).
[CRITICAL] TOCTOU race condition in approval thread resolution nearai/ironclaw#1486
- Cross-channel approval thread hijacking — Issue #1485 (CRITICAL, confidence 100%) (staging CI). Requires immediate attention.
- HIGH
- Migration checksum mismatch (V6__routines) causing upgrade failure — Issue #1328. Affects upgrades from older releases.
Upgrade to v0.19.0 fails: V6__routines migration checksum mismatch on existing PostgreSQL databases nearai/ironclaw#1328 - Google OAuth blocked across Google Suite tools — Issue #902 (impacts many users using Google tools).
Google OAuth blocked across Google Suite tools + request optional gws fallback provider nearai/ironclaw#902 - Incomplete fallback logic for non-existent approval threads — Issue #1487 (staging CI).
[HIGH] Incomplete fallback logic for non-existent approval threads nearai/ironclaw#1487
- Migration checksum mismatch (V6__routines) causing upgrade failure — Issue #1328. Affects upgrades from older releases.
- MEDIUM
- Lock contention in approval thread resolution — Issue #1489 (PR #1497 aims to reduce contention).
[MEDIUM] Lock contention in approval thread resolution nearai/ironclaw#1489 | fix(agent): reduce approval thread lock contention nearai/ironclaw#1497 - Gemini provider 400 on function call (Issue #1510) — affects specific backend/model combos; needs debug.
BUG: Gemini: function call is missing a thought_signature in functionCall parts. nearai/ironclaw#1510
- Lock contention in approval thread resolution — Issue #1489 (PR #1497 aims to reduce contention).
- Ghi chú sửa lỗi có sẵn:
- #1495 (fix) đã gửi để ngăn cross-channel approval hijacking: fix(agent): prevent cross-channel approval hijacking nearai/ironclaw#1495
- #1497 giảm lock contention: fix(agent): reduce approval thread lock contention nearai/ironclaw#1497
- Routines retry/notification PRs (#1470/#1471) giúp giảm thất bại tạm thời và cải thiện thông báo: fix(routines): normalize notification summaries with truncation and metadata nearai/ironclaw#1470 | fix(routines): add bounded retry for transient lightweight failures nearai/ironclaw#1471
Yêu cầu tính năng & tín hiệu lộ trình
- Yêu cầu đang được thảo luận / có khả năng vào bản tiếp theo:
- Làm cho workspace trở thành nguồn dữ liệu chính cho skills (thay filesystem) — Issue #1504 (XL, thay đổi kiến trúc).
feat(skills,workspace): make workspace the source of truth for skill storage, replacing filesystem nearai/ironclaw#1504 - Auto-export workspace .md files ra đĩa để phục hồi khi DB mất dữ liệu — Issue #1278.
feat: auto-export workspace .md files to disk for DB-resilient recovery nearai/ironclaw#1278 - Thêm ACP (Agent Client Protocol) job mode để ủy quyền cho coding agent khác — Issue #1506.
feat: Add ACP (Agent Client Protocol) job mode for delegating to any compatible coding agent nearai/ironclaw#1506 - Hỗ trợ embedding provider AWS Bedrock — Issue #1501.
feat: Add AWS Bedrock as an embedding provider nearai/ironclaw#1501 - Debug/inspector panel cho web gateway, verbose SSE và endpoints debug — Issues #1493 / #1492.
feat(web): add debug inspector panel to web gateway chat UI nearai/ironclaw#1493 | feat(web): add debug inspection endpoints and verbose SSE mode for the web gateway nearai/ironclaw#1492
- Làm cho workspace trở thành nguồn dữ liệu chính cho skills (thay filesystem) — Issue #1504 (XL, thay đổi kiến trúc).
- Dự đoán: trong vài phiên bản tới, đội sẽ ưu tiên:
- Khắc phục các vấn đề bảo mật/phê duyệt (những issue do CI nêu) trước khi đưa những tính năng lộ trình lớn vào release ổn định.
- Tích hợp cải tiến về workspace/embeddings (LRU cache đã xuất hiện trong v0.21.0) và tiếp tục mở rộng providers (Bedrock).
Tóm tắt phản hồi người dùng
- Điểm đau lặp lại:
- Google integration/OAuth không ổn định — người dùng không thể kết nối Google tools hoặc bị link auth sai (Issues #902, #1500, #1502, #1503).
Google OAuth blocked across Google Suite tools + request optional gws fallback provider nearai/ironclaw#902 | Google authentication fails connecting to the second google tool nearai/ironclaw#1500 - Thiếu tài liệu hướng dẫn cấu hình & sử dụng (Issue #1174) làm chậm quá trình onboarding.
Where is the documentation? nearai/ironclaw#1174 - Rủi ro mất dữ liệu workspace khi DB reset — người dùng đề xuất export/backup tự động (Issue #1278).
feat: auto-export workspace .md files to disk for DB-resilient recovery nearai/ironclaw#1278
- Google integration/OAuth không ổn định — người dùng không thể kết nối Google tools hoặc bị link auth sai (Issues #902, #1500, #1502, #1503).
- Phản hồi tích cực:
- Cải tiến UX/web và caching embeddings được nhắc đến như bước tiến hữu ích (PRs/Release notes).
- Người dùng chú ý tới tính ổn định của routines và khả năng retry/notifications (PRs #1470/#1471 được hoan nghênh).
Theo dõi tồn đọng (cần chú ý của bảo trì)
- Vấn đề bảo mật/độ nghiêm trọng cao do CI nêu — cần review & backport nếu cần:
- #1485 Cross-channel approval hijacking — [CRITICAL] Cross-channel approval thread hijacking - authorization bypass nearai/ironclaw#1485
- #1486 TOCTOU race in approval resolution — [CRITICAL] TOCTOU race condition in approval thread resolution nearai/ironclaw#1486
- #1487 Fallback logic for non-existent approval threads — [HIGH] Incomplete fallback logic for non-existent approval threads nearai/ironclaw#1487
- Nâng cấp DB thất bại:
- #1328 Migration checksum mismatch (V6__routines) — Upgrade to v0.19.0 fails: V6__routines migration checksum mismatch on existing PostgreSQL databases nearai/ironclaw#1328
- Tính năng / PR lớn cần review kỹ để tránh regressions:
- PR #1397 (parameter coercion & schema validation) — có phạm vi rộng, làm thay đổi hành vi tools/WASM: fix: parameter coercion and validation for oneOf/anyOf/allOf schemas nearai/ironclaw#1397
- PR #1277 UX overhaul (XL) — cần review phân mảnh UI và tương thích: feat(ux): complete UX overhaul — design system, onboarding, web polish nearai/ironclaw#1277
- PRs import history (#1498, #1505) — thao tác dữ liệu nhạy cảm, cần kiểm tra migration/import safety: Import claude convo history nearai/ironclaw#1498 | Import OpenAI history 151 nearai/ironclaw#1505
- Các issue người dùng cần câu trả lời / triage:
- #902 Google OAuth blocked — Google OAuth blocked across Google Suite tools + request optional gws fallback provider nearai/ironclaw#902
- #1174 Where is the documentation? — Where is the documentation? nearai/ironclaw#1174
- #1510 Gemini function call error (mới) — BUG: Gemini: function call is missing a thought_signature in functionCall parts. nearai/ironclaw#1510
Kết luận ngắn
- Dự án rất năng động với nhiều PR/issue lớn; v0.21.0 mang đến tính năng hữu ích nhưng cũng lộ ra các điểm yếu về tích hợp (Google), migration DB và một số rủi ro bảo mật/phê duyệt do staging CI phát hiện. Ưu tiên hiện tại nên là: (1) xử lý các issue bảo mật/approval cao cấp, (2) sửa các lỗi tích hợp Google & provider (Gemini/Bedrock), (3) review cẩn trọng PR lớn trước khi merge để tránh regressions.
Nếu muốn, tôi có thể:
- Tổng hợp đề xuất patch/action cho từng issue CRITICAL (checklist + người/nhóm cần tham gia), hoặc
- Soạn draft comment/triage template để maintainers dễ phân loại và ưu tiên những issue/PR kia.
LobsterAI — netease-youdao/LobsterAI
Tổng quan hôm nay
- Hoạt động cộng đồng cao: trong 24 giờ qua có 33 issue (mở/hoạt động 32, đóng 1) và 30 PR (mở 19, merge/đóng 11), nhiều báo cáo liên quan đến độ ổn định của gateway OpenClaw và trải nghiệm IM.
- Có phát hành phiên bản nhỏ 2026.3.20 với một sửa lỗi liên quan tới runtime OpenClaw.
- Dòng PR hiện tập trung vào ổn định gateway (Windows), sửa memory/IO blocking (SQLite), và làm cho cấu hình/im channel có thể hot-reload. Tổng quan cho thấy dự án đang ở giai đoạn khẩn trương khắc phục sự cố vận hành sau khi tung các bản 3.17–3.19.
Phát hành phiên bản (mới nhất)
- Phiên bản: 2026.3.20 (phát hành 2026-03-20)
- Nội dung chính: fix: stop stubbing playwright-core and pdfjs-dist in OpenClaw runtime (bằng PR #548, tác giả @winsan-zhang).
- Lưu ý di chuyển / breaking changes: changelog ngắn, không thấy thay đổi phá vỡ API lớn; nhưng do liên quan OpenClaw runtime, người dùng có gateway tùy chỉnh nên thử nghiệm lại các tích hợp có dùng playwright/pdfjs.
- Full changelog: netease-youdao/LobsterAI@2026.3.19...2026.3.20
Tiến độ dự án (PRs đã merge/đóng hôm nay)
- Theo dữ liệu, 11 PR đã được merge/đóng trong 24h; một số PR đáng chú ý đã đóng/merge:
- #593 Liuzhq/skills safety — thêm scan an toàn khi cài skill (CLOSED) https://github.com/netease-youdao/LobsterAI/pull/593
- #591 remove plaintext secrets from openclaw.json — thay plaintext bằng biến môi trường (CLOSED) https://github.com/netease-youdao/LobsterAI/pull/591
- #569 logger: daily rotation, 80 MB limit, 7-day retention (CLOSED) https://github.com/netease-youdao/LobsterAI/pull/569
- #565 suppress timeout hint when task is manually stopped (CLOSED) https://github.com/netease-youdao/LobsterAI/pull/565
- #558 one-click Feishu bot creation via QR device flow (CLOSED) https://github.com/netease-youdao/LobsterAI/pull/558
- #556 POPO WebSocket long connection support (CLOSED) https://github.com/netease-youdao/LobsterAI/pull/556
- PR đang tích cực review / chuẩn bị merge:
- #599 fix: sửa báo cáo kết nối model lỗi giả (open) https://github.com/netease-youdao/LobsterAI/pull/599
- #597 fix(im): hot-reload model cho IM channels (open) https://github.com/netease-youdao/LobsterAI/pull/597
- #596 fix(security): sanitize workspace path (open) https://github.com/netease-youdao/LobsterAI/pull/596
- #573 perf(sqlite): chuyển SqliteStore sang async IO (open) https://github.com/netease-youdao/LobsterAI/pull/573
- #576 fix(openclaw): cải thiện khởi động gateway trên Windows (open) https://github.com/netease-youdao/LobsterAI/pull/576
Chủ đề nóng trong cộng đồng
- OpenClaw gateway liên tục restart / khởi động thất bại:
- #540 open claw 频繁重启 (issue) — báo cáo restart liên tục trên 3.19 https://github.com/netease-youdao/LobsterAI/issues/540
- #572 openclaw gateway 拉不起来 (issue) — gateway không thể khởi động https://github.com/netease-youdao/LobsterAI/issues/572
- Nhu cầu: chẩn đoán startup, log rõ ràng, và độ ổn định hàng loạt cho Windows/Linux.
- Tin nhắn tự gửi / trùng lặp khi nhận từ IM bot:
- #594 从飞书机器人发送时,客户端频繁自动发送无关消息 — nghi ngờ OpenClaw liên quan https://github.com/netease-youdao/LobsterAI/issues/594
- Nhu cầu: xử lý event dedup, ổn định luồng IM → gateway → model.
- Bảo mật/đường dẫn workspace và secrets:
- Issue path traversal trong memory file đã được tác động bởi PR #596 (open) https://github.com/netease-youdao/LobsterAI/pull/596
- PR #591 (CLOSED) đã loại bỏ plaintext secrets khỏi openclaw.json https://github.com/netease-youdao/LobsterAI/pull/591
- Nhu cầu: tiếp tục harden lưu trữ credentials và validate path input.
- Model hot-reload và routing IM:
- #585 报告切换模型后 IM 仍调用旧模型 → đã có PR fix hot-reload #597 https://github.com/netease-youdao/LobsterAI/issues/585 https://github.com/netease-youdao/LobsterAI/pull/597
Lỗi & Ổn định (xếp theo độ nghiêm trọng)
- Cao (blocker / ảnh hưởng nhiều người)
- OpenClaw gateway khởi động/restart liên tục (#540, #572) — ảnh hưởng trực tiếp đến IM integrations và tự động hóa; PR #576 đang hướng tới cải thiện khởi động Windows https://github.com/netease-youdao/LobsterAI/issues/540 https://github.com/netease-youdao/LobsterAI/issues/572 https://github.com/netease-youdao/LobsterAI/pull/576
- IM bot gửi/nhận phát sinh tin nhắn lạ / tự động gửi lịch sử (#594, #561) — gây mất tin cậy, nghi vấn event replay/duplication https://github.com/netease-youdao/LobsterAI/issues/594 https://github.com/netease-youdao/LobsterAI/issues/561
- Trung bình (kinh nghiệm người dùng/gây lo ngại)
- Windows auto-start thất bại, process exit không có logs (#595) — ảnh hưởng UX trên Windows; PR #576 chứa cải thiện diagnostic https://github.com/netease-youdao/LobsterAI/issues/595
- Model switch/channels chưa hot-reload (#585) — đã có PR #597 đang chờ review https://github.com/netease-youdao/LobsterAI/issues/585 https://github.com/netease-youdao/LobsterAI/pull/597
- SQLite blocking main thread khi ghi đồng bộ (#562) — PR #573 chuyển sang async IO đang chờ review https://github.com/netease-youdao/LobsterAI/issues/562 https://github.com/netease-youdao/LobsterAI/pull/573
- Thấp (UI/UX, local issues)
- Ảnh input config khó tìm (#588) https://github.com/netease-youdao/LobsterAI/issues/588
- Workspace bị hiểu nhầm và tạo file .openclaw/SOUL.md ở nhầm chỗ (#579) https://github.com/netease-youdao/LobsterAI/issues/579
Yêu cầu tính năng & tín hiệu lộ trình
- Thống kê token per-model (giám sát chi phí) — issue #582 đang yêu cầu tính năng này; nhiều hệ thống LLM management sẽ coi đây là tính năng ưu tiên để quản lý chi phí https://github.com/netease-youdao/LobsterAI/issues/582
- Hỗ trợ triển khai offline / intranet (enterprise) — issue #587 đề nghị thích ứng môi trường nội bộ không có Internet https://github.com/netease-youdao/LobsterAI/issues/587
- UX: one-click IM bot setup (Feishu) và POPO websocket đã được đưa vào/merge (#558, #556) → cho thấy hướng tập trung vào onboarding IM channels nhanh hơn https://github.com/netease-youdao/LobsterAI/pull/558 https://github.com/netease-youdao/LobsterAI/pull/556
- Dự đoán: các phiên bản tiếp theo có thể ưu tiên (1) độ ổn định OpenClaw, (2) token usage UI, (3) offline deployment và (4) quản lý secrets/secure config.
Tóm tắt phản hồi người dùng
- Điểm đau chính: độ ổn định runtime (gateway restart, khởi động thất bại), routing model không đúng sau đổi cấu hình, và hành vi IM bất thường (tin nhắn tự gửi hoặc mất/treo). Nhiều issue báo cáo ảnh hưởng trực tiếp đến luồng công việc (chat bot trả lời không liên tục hoặc dùng sai model).
- UX: một số cấu hình bị ẩn (ví dụ cấu hình input ảnh), thiếu auto-update/upgrade rõ ràng trên Windows, và hiện tượng icon ứng dụng bị lỗi trên mac intel (#551). Người dùng tỏ ra không hài lòng chủ yếu vì mất tin cậy và gián đoạn công việc.
- Ví dụ: #511 (Feishu bot không trả lời), #560 (app crash khi tương tác), #551 (Mac intel icon lỗi) https://github.com/netease-youdao/LobsterAI/issues/511 https://github.com/netease-youdao/LobsterAI/issues/560 https://github.com/netease-youdao/LobsterAI/issues/551
Theo dõi tồn đọng (cần ưu tiên / phản hồi maintainer)
- Issues có tương tác cao / cần phản hồi:
- #563 对话标题显示乱码 (7 bình luận) — https://github.com/netease-youdao/LobsterAI/issues/563
- #540 open claw 频繁重启 — https://github.com/netease-youdao/LobsterAI/issues/540
- #595 Windows 开机自动启动失败 — https://github.com/netease-youdao/LobsterAI/issues/595
- #594 飞书机器人导致客户端频繁自动发送 — https://github.com/netease-youdao/LobsterAI/issues/594
- PRs lớn / infra cần review sớm:
- #599 fix(settings) — sửa test connection false negative (ảnh hưởng trải nghiệm model settings) https://github.com/netease-youdao/LobsterAI/pull/599
- #597 fix(im) — hot-reload model cho IM channels (giải quyết #585) https://github.com/netease-youdao/LobsterAI/pull/597
- #596 security path sanitize — ưu tiên code-review và audit vì liên quan path traversal https://github.com/netease-youdao/LobsterAI/pull/596
- #573 sqlite async IO — nên review sớm để giải quyết blocking main thread issue (#562) https://github.com/netease-youdao/LobsterAI/pull/573
- #589 infra: full CI/CD & security automation PR (open) — lớn, cần maintainer đánh giá tác động https://github.com/netease-youdao/LobsterAI/pull/589
Kết luận ngắn
- Dự án có cộng đồng tích cực và nhiều PR/issue xử lý nhanh, nhưng hiện đang đối mặt với các vấn đề ổn định runtime (OpenClaw/gateway) và tích hợp IM—những vấn đề này ảnh hưởng trực tiếp tới tin cậy sản phẩm. Ưu tiên cần là: (1) chẩn đoán/ổn định gateway, (2) hot-reload model cho IM, (3) chuyển I/O blocking sang async và (4) củng cố bảo mật secrets/path. Các PR liên quan đã được mở và một số đã được merge; maintainer nên ưu tiên review các PR liên quan đến ổn định và bảo mật.
Tài nguyên tham khảo (trích một số link)
- Release 2026.3.20 changelog: netease-youdao/LobsterAI@2026.3.19...2026.3.20
- Issue OpenClaw restart: https://github.com/netease-youdao/LobsterAI/issues/540
- Issue Feishu/IM 自动发送: https://github.com/netease-youdao/LobsterAI/issues/594
- Issue Windows auto-start: https://github.com/netease-youdao/LobsterAI/issues/595
- PR hot-reload IM model: https://github.com/netease-youdao/LobsterAI/pull/597
- PR sqlite async IO: https://github.com/netease-youdao/LobsterAI/pull/573
- PR sanitize workspace path (security): https://github.com/netease-youdao/LobsterAI/pull/596
Nếu cần, tôi có thể tóm tắt những thay đổi code chi tiết cho từng PR ưu tiên (ví dụ #597, #573, #596) để giúp reviewer nhanh nắm nội dung.
TinyClaw — TinyAGI/tinyclaw
TinyClaw — Bản tin dự án (2026-03-21)
- Tổng quan hôm nay
- Dự án có hoạt động phát triển rõ rệt trong 24 giờ qua mặc dù không có issue mới: 5 PR được cập nhật (4 đã đóng/merged, 1 còn mở) và một phiên bản mới v0.0.16 được phát hành.
- Hành động chính tập trung
⚠️ 内容超过 GitHub Issue 上限,完整报告见提交的 Markdown 文件。