Skip to content

🦞 OpenClaw Ecosystem Digest 2026-08-09 #131

Closed
Closed
🦞 OpenClaw Ecosystem Digest 2026-08-09#131
Labels
openclaw-en

Description

@datnguyenquy94
datnguyenquy94
opened on Aug 9, 2026

OpenClaw Ecosystem Digest 2026-08-09

Issues: 166 | PRs: 500 | Projects covered: 12 | Generated: 2026-08-09 02:14 UTC

OpenClaw Deep Dive

OpenClaw Project Digest — 2026-08-09

1. Today's Overview

OpenClaw shows very high velocity with 500 PRs and 166 issues updated in 24 hours. Two patch releases (v2026.6.33, v2026.6.34) shipped recently, both focused on security hardening — sandboxing browser routes, capping hostile response sizes, and preventing credential leakage in diagnostics. The open issue backlog is dominated by P0/P1 reliability bugs: a critical gateway memory leak (350 MB → 15.5 GB), silent model reply failures, OAuth wedging, and message loss across Telegram/Feishu/Codex. Meanwhile, a large batch of PRs from maintainers (notably vincentkoc, steipete) advances Code Mode provider accounting, session trace integrity, and device pairing UX — signaling a push toward production-grade agent observability and multi-device workflows.

2. Releases

v2026.6.34 — Safer Browser & Network Boundaries

  • Sandboxed browser routes — reject unsafe access paths (#97958, #38290, #103075, #110693)
  • Trusted DNS targets — restrict outbound resolution
  • Custom browser origins — tightened origin validation
  • Loopback provider endpoints — block unsafe local access
  • Contributors: @​eleqtrizit, @​brunowowk, @​mosidevv, @​pgondhi987

v2026.6.33 — Safer Network & Secret Boundaries

  • Provider streams / Discord REST / browser fetches / OAuth paths / logs — cap hostile response sizes
  • Telegram credentials — kept out of diagnostics (#96989, #95412, #99428)
  • Contributors: @​wangmiao0668000666, @​Alix-007

No breaking changes noted in either release; both are defensive hardening patches.

3. Project Progress (Merged/Closed Today)

PR Area Summary
#119511 sessions / tasks Fix: tasks maintenance --apply now archives cron-run transcripts as .deleted instead of hard-deleting (closes #119269)
#120803 worker / gateway Fix: Preserve long OpenAI Responses sessions across cloud handoff — prevents silent loss of server-compaction replay
#120802 windows / gateway Fix: Preserve configured child env overrides across key casing differences
#120817 telegram Fix: Restore account-specific replyToMode on beta.1 (release-blocking regression)
#120820 web-ui Fix: Dashboard widgets no longer flash white in dark mode
#120822 google-meet Fix: Stop failed voice gateway connections cleanly (no leaked retry timers)
#120824 msteams Fix: Paginate channel thread replies > 50 (closes #98870)
#120827 cli Fix: Keep QR codes within standard 80-column terminal widths (closes #120758)

Pattern: Rapid closure of recent regressions (beta.1 Telegram, QR CLI, cron transcript loss) + steady progress on Code Mode traceability (see §6).

4. Community Hot Topics

Issue Status Comments Core Need
#116277 CLOSED 179 DeepSeek v4 Flash silent failure → generic fallback; need provider-level error visibility
#91588 OPEN 22 P0: Gateway RSS leaks 350 MB → 15.5 GB over days → OOM kills; blocks long-running deployments
#86215 OPEN 11 Codex OAuth refresh wedges agent for hours; no alerting, no aggressive profile rotation
#84583 OPEN 11 Cron announce delivery collides with active user chat → EmbeddedAttemptSessionTakeoverError
#10687 OPEN 10 Dynamic model discovery for OpenRouter (static catalog is stale)
#114154 OPEN 7 MCP tool passes policy/probe but never bundled — ToolSearch finds nothing
#79293 OPEN 6 openclaw-weixin proactive sends report success but user sees "请稍后再试" / missing chunks
#114020 OPEN 6 Feishu/Telegram dispatch fails: runChannelInboundEvent requires runDispatchLifecycle (post-2026.7.2-beta.4)

Underlying themes:

  • Reliability > features — memory leak, silent failures, and message loss dominate discussion
  • Multi-channel parity — Telegram, Feishu, Weixin, Teams, Discord all show delivery gaps
  • Provider observability — users cannot diagnose why a model/tool/channel failed

5. Bugs & Stability (Ranked by Severity)

Severity Issue Description Fix PR?
P0 / Critical #91588 Gateway memory leak → OOM crash every 2–3 days; launchd-handoff restart loops ❌ No PR yet
P1 / Message Loss #116277 DeepSeek v4 Flash silent reply failure → generic "No reply generated" fallback ✅ Closed (fix implied)
P1 / Message Loss #114020 Feishu/Telegram dispatch broken since 2026.7.2-beta.4 — runDispatchLifecycle missing ❌
P1 / Session State #87327 Isolated agent runs stall in runtime-plugins phase before execution (hourly crons) ❌
P1 / Auth #86215 Codex OAuth refresh wedges agent for hours; no alerting/profile rotation ❌
P1 / Message Loss #120425 Telegram DM reply-to-photo re-fetches media every turn → multi-min latency, races timeout ❌
P1 / Crash Loop #89791 Supervisor mode + dual plist → 30s EADDRINUSE loop on port 18789 ❌
P2 / Data Loss #119269 tasks maintenance --apply hard-deletes cron transcripts without .deleted archive ✅ #119511 merged
P2 / Session State #94536 Commitment marked sent but never delivered (PR #92231 fix incomplete) ❌
P2 / Compatibility #98870 Teams thread context omits replies > 50 (Graph pagination) ✅ #120824 open

Note: Several P1s have no fix PR after weeks/months — maintainer bandwidth appears constrained on deep runtime bugs.

6. Feature Requests & Roadmap Signals

Issue Votes Signal Likely Next Version?
#10687 3 Dynamic model discovery (OpenRouter + beyond) — static catalog is a recurring pain point 🟡 High — multiple providers affected
#13219 1 Per-model usage logging for cost tracking — session JSONL has data but no aggregation 🟡 High — aligns with Code Mode accounting PRs
#49740 0 Cron auto-retry (--retry-count, --retry-delay) — daily crons fail silently until next day 🟢 Medium — clear operator need
#71330 0 Configurable memory promotion target (not hardcoded MEMORY.md) 🟢 Medium — architectural cleanup
#10944 0 Telegram parseMode config (Markdown breaks emojis/formatting) 🟢 Medium — simple, high impact
#71195 1 OpenAI Realtime (speech-to-speech) for macOS Talk Mode — parity with voice-call plugin 🔴 Low — niche, complex
#8299 1 Suppress sub-agent announce (currently requires model to reply ANNOUNCE_SKIP) 🟢 Medium — UX friction

Strongest signals: Dynamic model catalog, per-model cost observability, cron retry, Telegram parse mode — all address daily operator friction.

7. User Feedback Summary

Pain Point Evidence User Impact
Silent failures #116277 (179 comments), #79293, #120425, #111944 Users see "bot went silent" or generic fallbacks; no debug visibility
Memory instability #91588 (22 comments, P0) Requires manual restart / monitoring; breaks unattended automation
OAuth / auth wedging #86215, #107839, #86214 Agents stuck for hours; no alerting; manual intervention needed
Channel delivery gaps #114020 (Feishu/Telegram), #98870 (Teams), #79293 (Weixin) Multi-channel deployments unreliable; trust erodes
Session/transcript loss #119269, #73471, #120567 doctor --fix and tasks maintenance delete history without archive
UI/UX friction #75947 (8 comments), #9637 (accessibility), #106475 (QR in webchat) Config pages dense; TUI not screen-reader friendly; QR codes break in 80-col terminals
Provider observability #13219, #10687, #118673 Cannot track costs, model versions, or stop reasons; "flying blind"

Positive notes: Users cite OpenClaw as "part of daily workflow" (#73537), appreciate family/business automation, and engage deeply with detailed repros.

8. Backlog Watch — Stalled High-Value Items

Item Age Why It Matters Blocked By
#91588 ~2 months P0 memory leak — blocks production stability; no PR, no maintainer assignment Deep runtime profiling needed
#86215 ~2.5 months Codex OAuth wedging — affects all Codex users; needs alerting + rotation logic Product decision on rotation policy
#10687 ~6 months Dynamic model discovery — requested by many; static catalog is stale Architecture decision on provider abstraction
#73537 ~3.5 months Production-readiness label for releases — users cannot distinguish stable vs beta Release process definition
[#87327](https://github​.com/openclaw/openclaw

Cross-Ecosystem Comparison

Cross-Project Comparison Report: Personal AI Assistant / Agent Open-Source Ecosystem

Generated from 2026-08-09 community digests

1. Ecosystem Overview

The personal AI agent ecosystem shows bifurcated maturity: a tier of high-velocity, production-hardening projects (OpenClaw, IronClaw, ZeroClaw, Hermes) operating at 50+ PRs/day with dedicated maintainer teams, and a tier of specialized or earlier-stage projects (NanoBot, PicoClaw, NanoClaw, Moltis, CoPaw, LobsterAI) focused on protocol integration, sandbox stability, and UX polish. No project released a minor/major version today — all are in patch, stabilization, or architectural migration phases. Security hardening (sandboxing, response caps, credential hygiene) and multi-channel reliability (Telegram, Slack, Discord, Matrix, Feishu, Weixin) are universal concerns. The ecosystem is converging on MCP (Model Context Protocol) as the standard tool integration layer, with remote HTTP/SSE MCP support landing in multiple codebases simultaneously.

2. Activity Comparison

Project Issues (24h) PRs (24h) Release Status Health Score*
OpenClaw 166 500 2 patch releases (security) 🟢 High — velocity + releases + maintainer depth
IronClaw 25 50 None (Reborn migration) 🟡 High — velocity + architectural progress, but XL PR backlog
ZeroClaw 11 50 None (consolidation phase) 🟡 High — velocity + security focus, but review bottleneck + S0 bug
Hermes Agent 12 50 None (v0.20.0 2026-08-07) 🟡 Medium-High — velocity but 43 open PRs, P1 bugs unfixed
CoPaw/QwenPaw 1 50 None (pre-release stabilization) 🟡 Medium — high PR throughput but low issue closure, architectural gaps
NanoBot 5 9 None 🟢 Medium — focused delivery (token observability + ephemeral chat shipped)
NanoClaw 8 6 None 🟢 Medium — steady merging, MCP/channel expansion
PicoClaw 3 4 (open) None 🟡 Medium-Low — stale PRs (>30d), review bandwidth constrained
Moltis 2 1 (merged) None 🟢 Medium — sandbox fix delivered, new Apple Container gap
LobsterAI 1 (stale) 3 (stale) None 🔴 Low — high-value work stalled 4+ months, zero community engagement
NullClaw 0 0 None ⚪ Unknown — no activity
ZeptoClaw 0 0 None ⚪ Unknown — no activity

*Health Score: 🟢 Healthy velocity & delivery | 🟡 High velocity but risks (review bottleneck, unfixed P1s, migration) | 🔴 Stalled critical work

3. OpenClaw's Position

Advantages vs Peers

  • Scale of operation: 10x PR volume of nearest peers (500 vs 50/day); 166 issue updates indicates massive community engagement
  • Release discipline: Only project shipping security patches today (v2026.6.33/34) — demonstrates production readiness
  • Maintainer depth: Multiple named maintainers (vincentkoc, steipete, eleqtrizit, brunowowk, mosidevv, pgondhi987, wangmiao0668000666, Alix-007) driving distinct workstreams
  • Observability lead: Code Mode provider accounting, session trace integrity, device pairing UX — features absent in most peers
  • Multi-channel breadth: Active fixes across Telegram, Feishu, Weixin, Teams, Discord, Codex, Google Meet simultaneously

Technical Approach Differences

  • Gateway-centric architecture with explicit sandboxing, response-size caps, and DNS restrictions — peers handle security at channel or tool level
  • Cron/task maintenance as first-class operator concern (transcript archival, retry policies) — others treat scheduling as afterthought
  • Provider abstraction with dynamic discovery (OpenRouter issue #10687) vs static catalogs in NanoBot, Hermes, NanoClaw
  • Session handoff preservation across cloud/edge boundaries (PR #120803) — unique focus on long-running session continuity

Community Size Comparison

  • OpenClaw: 179 comments on single issue (#116277), 22 on P0 memory leak — orders of magnitude higher engagement than any peer
  • IronClaw/ZeroClaw/Hermes: 5–11 comments on hottest items (mostly maintainer-driven RFCs)
  • NanoBot/NanoClaw/PicoClaw/Moltis/CoPaw: 0–4 comments typical — niche or early communities
  • LobsterAI: Zero reactions/comments — dormant community

4. Shared Technical Focus Areas

Requirement Projects Affected Specific Need
MCP remote server support (HTTP/SSE) NanoClaw (merged #2776), IronClaw (Reborn), ZeroClaw (catalog unification), NanoBot (schema budgeting #5298) Standardized remote MCP endpoint discovery, auth, and capability negotiation
Token/cost observability NanoBot (merged #5293, #5299), OpenClaw (#13219), Hermes (session write policy), ZeroClaw (gateway audit logging #9410) Per-iteration, per-model, per-tool token accounting with UI visibility
Multi-channel message delivery reliability OpenClaw (Telegram/Feishu/Teams/Weixin), Hermes (Slack/Telegram), IronClaw (Slack/Telegram/Web Push), NanoClaw (Discord/Mattermost/Google Chat), PicoClaw (WhatsApp/IRC/DeltaChat/SimpleX) Unified delivery acknowledgment, retry, deduplication, and cross-channel threading
Sandbox/container execution hardening OpenClaw (browser routes, loopback), Moltis (Docker fallback merged, Apple Container gap), IronClaw (WASM guest sanitization #7048), ZeroClaw (network guards #9580), CoPaw (shared filesystem persistence #6767) Filesystem access control, host-path translation, runtime detection across Docker, Podman, Apple Container, WASM
Authentication resilience (OAuth refresh, rotation) OpenClaw (Codex OAuth wedging #86215), Hermes (Anthropic content filter #82154, xAI 403 #82052), NanoBot (MCP OAuth #5297), PicoClaw (OAuth 2.1 MCP #3302), ZeroClaw (gateway auth #9744) Automatic token refresh, profile rotation, alerting on auth failure, standards compliance (OAuth 2.1)
Session state integrity across restarts OpenClaw (gateway handoff #120803), Hermes (Codex thread resume #82160, SIGTERM on release #41225), CoPaw (persistence on shared FS #6767), ZeroClaw (SOP cron attachment #9494) Durable session identity, background process survival, atomic state writes

5. Differentiation Analysis

Dimension OpenClaw IronClaw ZeroClaw Hermes NanoBot NanoClaw PicoClaw CoPaw Moltis LobsterAI
Primary Focus Production-grade multi-device agent observability Reborn architectural migration (identity, skills, channels) Crate consolidation, security hardening, catalog unification Desktop/cloud agent with Modal backend, session lifecycle Token observability, ephemeral chat, MCP resilience Multi-channel + MCP ecosystem expansion Protocol bridging (WhatsApp, IRC, SimpleX, DeltaChat) Multi-agent (sub-agent) workspace isolation Sandbox execution (Docker, Apple Container) SQLite perf, LiteLLM gateway, tool config declarativeness
Target User Power users, automation engineers, multi-device deployments Platform builders, channel/tool authors Framework integrators, security-conscious deployers Desktop-first users, cloud (Modal) developers Cost-sensitive developers, privacy-focused users Multi-channel bot operators, MCP integrators Bridge operators (IRC↔AI, WhatsApp↔AI) Multi-agent workflow designers Containerized agent runners Chinese-market developers, embedded/headless automation
Architecture Gateway-centric, session handoff, cron/task first-class Reborn: capability lifecycle, product workflows, channel adapters Workspace: plugins, channels, providers, capabilities unified Agent + gateway + Modal backend + desktop (Tauri) WebUI + gateway, token diagnostics baked in Channel adapters + MCP servers + skills Protocol adapters + MCP client AgentScope 2.0 alignment, Scroll memory Sandbox runtime abstraction + gateway Electron + sql.js (in-browser SQLite)
Unique Strength Scale, release discipline, observability depth Identity model (owner≠actor), model-driven skills, stress infra Unified catalog contract, security-first defaults Modal sandbox integration, session write policy Sub-iteration token UI, ephemeral chat Remote MCP HTTP/SSE, Strava skill demo WhatsApp/IRC/SimpleX/DeltaChat breadth Sub-agent model isolation + shared workspace Docker fallback + Apple Container focus Headless browser config declarativeness

6. Community Momentum & Maturity

Tier 1: Rapid Iteration at Scale (50+ PRs/day, multi-maintainer)

  • OpenClaw — Maturing: Shipping patches, tackling P0 reliability, building observability. Highest community engagement.
  • IronClaw — Migrating: Reborn architectural overhaul (23 issues closed today). High velocity but XL PR backlog indicates review saturation.
  • ZeroClaw — Consolidating: Crate reduction, security hardening, catalog unification. Governance bottleneck (#8692: 11 RFCs queued).
  • Hermes Agent — Stabilizing: v0.20.0 recent, but 43 open PRs and unfixed P1s (gateway drain, Anthropic auth) show release debt.
  • CoPaw/QwenPaw — Pre-release: 50 PRs but only 1 issue; heavy refactoring (AgentScope 2.0 alignment, persistence hardening). Sub-agent UX gap emerging.

Tier 2: Focused Delivery (5–10 PRs/day, clear feature shipping)

  • NanoBot — Shipping: Token observability + ephemeral chat delivered end-to-end in 24h. MCP resilience next.
  • NanoClaw — Expanding: Remote MCP, Strava skill, Mattermost channel merged. Platform integration focus.
  • PicoClaw — Constrained: Protocol work ready (WhatsApp fix, SimpleX, DeltaChat) but review bandwidth stalled (>30d PRs).
  • Moltis — Stabilizing: Docker sandbox fixed, Apple Container detection gap new. Niche but responsive.

Tier 3: Low Velocity / Stalled

  • LobsterAI — Dormant: High-value PRs stale 4+ months (#1193 SQLite perf, #1192 tool config). Zero community signals.
  • NullClaw / ZeptoClaw — Inactive: No 24h activity.

7. Trend Signals for AI Agent Developers

Trend Evidence Strategic Value
MCP is the universal tool integration layer Remote HTTP/SSE MCP merged in NanoClaw

Peer Project Reports

NanoBot — HKUDS/nanobot

NanoBot Project Digest — 2026-08-09

1. Today's Overview

NanoBot shows high maintenance velocity with 14 total updates (5 issues, 9 PRs) in the last 24 hours and zero new releases, indicating active development sprint rather than stabilization. Four PRs were merged/closed today, focusing on code hygiene (dead code removal, UI polish) and a user-facing feature (temporary chat mode). The issue queue surfaces three critical pain points: runaway token consumption, MCP integration fragility (OAuth, crash loops), and Docker deployment friction. No releases since the data cutoff suggests the team is batching fixes for a near-term cut.

2. Releases

None in the last 24 hours.

3. Project Progress — Merged / Closed PRs Today

PR Type Summary Impact
#5252 feat(webui) Adds Temporary Chat mode — ephemeral, multi-turn conversations that never persist to disk or session history. New UX pattern for privacy-sensitive / throw-away interactions.
#5293 feat(usage) Logs per-iteration token diagnostics (input/output/cached) alongside existing daily aggregates. Directly addresses #5266; enables pinpointing token-hungry agent runs.
#5299 feat(webui) Surfaces recent token-usage records (with iteration & tool context) in the WebUI. Makes #5293 data visible to end-users without CLI access.
#5296 refactor/chore Removes 19 dead internal code units, 11 test-only seams, and orphaned frontend assets. Reduces maintenance surface; no user-facing change.
#5294 fix(webui) Eliminates image-hover clipping by dropping hover scaling/ring; keeps zoom cursor & focus ring. Polish fix for assistant image previews.

Net effect: Token observability stack (backend + UI) shipped end-to-end in one day; temporary chat mode lands; codebase trimmed.

4. Community Hot Topics

Item Activity Core Need
#5266 Token consumption logging 13 comments, 2 days old Users see millions of tokens burned in hours with no visibility into which call/iteration caused it. Urgent demand for per-call, per-iteration breakdown.
#5297 MCP OAuth / web auth 2 comments, same day Need to connect to MCP servers that require browser-based OAuth flows (e.g., Xmind). Current gateway cannot complete external auth handshakes.
#5300 MCP crash loop & CPU spike 0 comments, same day Remote MCP HTTP 530 → anyio cancel-scope cross-task RuntimeError → gateway crash + leaked tasks → CPU runaway. High-severity stability hole.
#5271 Stale background task overwrite Open, P0, conflict Background title-generation task can overwrite session after /new clears it. Data-loss risk; marked priority P0.

Signal: Token observability and MCP resilience are the two dominant operational concerns.

5. Bugs & Stability — Ranked by Severity

Severity Issue / PR Status Fix PR?
Critical #5300 MCP cancel-scope crash → gateway death + CPU leak Open No
High #5271 Stale background task overwrites session after /new Open (P0) Yes (#5271)
High #5295 Docker compose: entrypoint.sh permission denied Open No
Medium #5206 Duplicate streamed-response logging Open (P2) Yes (#5206)
Low #5294 Image hover clipping (UI only) Closed Yes (#5294)

Note: #5300 has no fix PR yet and causes full gateway collapse — should be triaged immediately.

6. Feature Requests & Roadmap Signals

Request Source Likelihood Next Release
Per-call / per-iteration token logging + UI #5266, #5293, #5299 Very High — backend + UI already merged.
Temporary / ephemeral chat mode #5252 Done — merged today.
MCP OAuth / gateway-assisted web auth #5297 Medium — requires gateway redesign; no PR yet.
Budgeted MCP schema exposure (large tool sets) #5298 Medium — architectural; correlates with token budget work.
Model-agnostic computer-use / browser tools #4276 Low-Medium — long-open (Jun), large scope, still open.
Matrix room-level reply threading #5292 Medium — niche but clean PR; may land soon.

Prediction: Next cut will ship token observability + temporary chat; MCP OAuth & schema budgeting are next-quarter candidates.

7. User Feedback Summary

Pain Point Evidence Sentiment
Invisible token burn “millions of tokens in 2 hours, no activity” (#5266) 😡 Frustrated — blocks cost control.
MCP unreliability Crash on HTTP 530, OAuth unsupported (#5300, #5297) 😰 Anxious — core integration broken.
Docker deploy friction entrypoint.sh permission denied on stock compose (#5295) 😕 Annoyed — blocks onboarding.
Session data loss risk Background task overwrites after /new (#5271) 😟 Concerned — marked P0 by maintainers.
Positive Temporary chat mode welcomed; token UI praised in PR reviews 🙂 Satisfied — visible progress on observability.

8. Backlog Watch — Stale / High-Leverage Items Needing Attention

Item Age Why It Matters
#4276 Computer-use / browser tools 60 days Flagship agent capability; large diff, needs review bandwidth.
#5206 Duplicate stream logging fix 8 days P2, clean fix, stuck in review — logging noise affects debugging.
#5271 Stale session task (P0) 3 days Critical data-integrity fix; marked conflict — needs rebase/merge.
#5300 MCP cancel-scope crash 0 days New but critical; no PR, no workaround — gateway instability.
#5297 MCP OAuth 0 days Unlocks whole class of MCP servers; requires gateway auth proxy design.

Bottom line: NanoBot is in a high-throughput fix/feature sprint — token observability and ephemeral chat delivered today, but MCP stability (crash + OAuth) and Docker onboarding are the next burning fires. Maintainer bandwidth on #4276 and #5271 will signal release cadence.

Hermes Agent — nousresearch/hermes-agent

Hermes Agent Project Digest — 2026-08-09

1. Today's Overview

Hermes Agent shows high velocity with significant stability concerns on 2026-08-09. The project processed 62 total updates (12 issues, 50 PRs) in 24 hours with zero releases, indicating active development but a backlog of unreleased fixes. Critical P1 bugs affect gateway shutdown behavior and Anthropic authentication, while multiple P2 issues span desktop updates, Modal backend reliability, and session state management. The PR pipeline is heavily loaded (43 open PRs) with several salvaged fixes from prior attempts, suggesting technical debt in review processes. Community engagement is moderate—most new issues have 0–1 comments—indicating either rapid triage or limited external contributor bandwidth.

2. Releases

No new releases today. The last release (v0.20.0) was referenced in issue #82166 as having been available since 2026-08-07, but the update-check endpoint incorrectly reported "up to date" for consecutive days.

3. Project Progress — Merged/Closed PRs Today

PR Type Component Summary
#82160 Bug fix codex_app_server Resume persisted Codex thread on agent re-instantiation instead of creating fresh thread
#79343 Bug fix Memory provider Fix trivial-prompt classifier incorrectly skipping memory recall for workflow commands (continue, proceed, next, etc.)
#79325 Bug fix Models.dev catalog Add SiliconFlow provider to PROVIDER_TO_MODELS_DEV mapping (49/47 models now visible in picker)
#82158 Bug fix Desktop update Fix venv-blocker scan truncating cmdlines, breaking gateway exemption and dead-ending Desktop updates
#79723 Feature Session write policy Integrate v0.20 session write policy migration (29 paths, 27 with effective diffs)
#80943 Feature Runtime/ACP Propagate and enforce session write policy into delegated child agents and Copilot ACP subprocess
#82158 Bug fix (duplicate) CLI/Windows Same as above — closed as duplicate

Net progress: 7 PRs merged/closed, addressing session persistence, memory classification, provider catalog completeness, Windows desktop update reliability, and session write policy enforcement.

4. Community Hot Topics

Item Activity Underlying Need
#82167 — Desktop ignores config.yaml provider 1 comment, created today Configuration authority: Users expect config file to be source of truth; desktop app persists stale provider selection causing billing to wrong endpoint
#82165 — Add Spanish locale 1 comment, created today Global accessibility: Spanish (600M speakers) missing from desktop language picker while Chinese, Japanese, Arabic present
#41225 — Background processes killed by SIGTERM on release() 4 comments, updated today (open since Jun) Session lifecycle integrity: Long-running background terminals die during session end, compression, error recovery — breaks workflows relying on persistent processes
#82166 — Update-check endpoint false "up to date" 0 comments, created today Release reliability: Inverse of known stale-cache bug; automated update checks silently fail for days

Pattern: Configuration drift, i18n gaps, and session-state fragility dominate user pain.

5. Bugs & Stability — Ranked by Severity

P1 (Critical)

Issue Component Status Fix PR
#82161 Gateway drain exits 0.00s with in-flight cron job, killing it mid-run Gateway/Cron Open None
#82154 Anthropic content filter rejects built-in SKILLS_GUIDANCE prompt — subscription OAuth fails with misleading "out of extra usage" 400 Agent/Anthropic/Auth Open None

P2 (High)

Issue Component Status Fix PR
#41225 Background processes killed by SIGTERM during release() Agent/Tools/Terminal Open (needs repro) None
#82167 Desktop sessions ignore config.yaml model.provider — bills old endpoint Desktop/Config/Billing Open None
#82164 Modal backend: sandbox silently dies mid-task — generic errors, no reconnect, file bridge dies Backend/Modal/Terminal/File Open (needs repro) None
#82168 Windows: Both updating and reinstalling fail Desktop/Windows/Install Open (needs repro) #82143 (self-heal get-windows binding)
#82152 FTS5 sanitizer missing special chars — search returns zero for it's, user@​host, 50% Agent/Search Open #82152 (salvaged from #79285)
#82151 Model-switch reads keys outside per-profile secret scope — cross-profile key leakage CLI/Auth/Profiles Open #82151 (salvaged from #79222)

P3 (Medium)

Issue Component Status Fix PR
#82052 xAI 403 classified non-retryable — long-lived workers never refresh expired OAuth token Agent/xAI/Auth Open None
#79343 Memory gate treats workflow commands as trivial — skips provider recall Agent/Memory Closed #79343
#79325 SiliconFlow missing from PROVIDER_TO_MODELS_DEV — picker empty Agent/Models.dev Closed #79325
#82166 Update-check endpoint false "up to date" for days while release exists CLI/Install Open None
#82165 Spanish locale missing from desktop Desktop/i18n Open None

Stability signal: 3 P1/P2 bugs with no fix PRs yet; Modal backend and gateway drain are single points of failure for cloud users.

6. Feature Requests & Roadmap Signals

Request Source Likelihood for Next Version
Spanish (es) locale for desktop #82165 High — low complexity, high impact, explicit demand
Per-child memory & toolset permission boundary in delegation #82157 High — PR open, addresses multi-agent security/isolation
Configurable background review tools (auxiliary.background_review.extra_tools) #82146 Medium — PR open, needs design decision
Search-only toolset exposed in configurator (web_search without web_extract) #82155 High — PR open with regression test
Bidirectional Telegram reactions #81709 Medium — PR open, platform-specific
Configurable human-facing timestamps (CLI/TUI/Desktop) #81439 Medium — PR open, UX polish
Deterministic MCP record/replay fixtures #80475 Medium — PR open, testing infrastructure

Roadmap inference: Next version (v0.21+) likely to ship i18n, delegation security boundaries, search tooling granularity, and MCP test fixtures. Gateway/Modal stability fixes are prerequisites.

7. User Feedback Summary

Pain Point Evidence Impact
Desktop config drift #82167: "new sessions ignore config.yaml model.provider — bills old endpoint" Financial (wrong billing), trust erosion
Session state loss #41225: background processes SIGTERM'd on release(); #82160: Codex thread identity lost on restart Workflow interruption, data loss
Update unreliability #82166: endpoint lies "up to date" for days; #82168: Windows update/reinstall both fail Stale deployments, manual intervention required
Cloud backend fragility #82164: Modal sandbox dies silently, no reconnect, file bridge dies Work loss, no recovery path
Auth token staleness #82052: xAI 403 non-retryable; #82154: Anthropic misleading 400 Long-lived sessions break silently
Search broken for common queries #82152: FTS5 rejects it's, user@​host, 50% Core feature unusable for technical queries

Positive signals: Active PR pipeline shows maintainers addressing memory classification, provider catalog, Windows native deps, and session write policy. Salvaged PRs (#82151, #82152, #82162, #82163) indicate commitment to closing old loops.

8. Backlog Watch — Needing Maintainer Attention

Item Age Why It Matters
#41225 Background processes killed by SIGTERM on release() 63 days (since Jun 7) Core session lifecycle bug; blocks persistent terminal workflows; labeled sweeper:risk-session-state, needs-repro
#82161 Gateway drain kills in-flight cron jobs 0 days (P1, new) Data loss risk for scheduled tasks; no fix PR; sweeper:risk-message-delivery
#82154 Anthropic content filter blocks built-in prompt 0 days (P1, new) Blocks all Anthropic subscription OAuth users; misleading error message
#82164 Modal backend silent sandbox death 0 days (P2, new) Cloud execution reliability; no reconnect/recovery; needs-repro
#82166 Update-check endpoint false negative 0 days (P3, new) Inverse of known bug (#11007, #1620); breaks automated update pipelines
#53040 Preserve last-good build during before-pack cleanup 44 days (open) Prevents hermes update destroying binary on failure; AI-authored, needs verification
#80475 MCP deterministic record/replay fixtures 3 days (open) Closes real-protocol testing gap; infrastructure investment

Recommendation: Prioritize P1 gateway/cron and Anthropic auth fixes for hotfix release; assign repro efforts for Modal and SIGTERM issues; schedule Spanish locale and search-only toolset for v0.21.

Digest generated from GitHub API data for NousResearch/hermes-agent as of 2026-08-09. All links point to live GitHub items.

PicoClaw — sipeed/picoclaw

PicoClaw Project Digest — 2026-08-09

1. Today's Overview

PicoClaw shows steady maintenance activity with 3 issue updates and 4 open PRs in the last 24 hours, though no merges or releases occurred. The project is actively addressing protocol compatibility (WhatsApp, IRC, SimpleX, DeltaChat) and performance optimization (prefix caching, CPU usage). Two stale PRs (#3222, #3193) remain open for over a month, suggesting review bandwidth constraints. The closed CPU-usage bug (#3292) indicates responsive triage for user-reported regressions.

2. Releases

No new releases in the last 24 hours. Current latest version remains unlisted in provided data.

3. Project Progress

No PRs merged or closed today. All 4 active PRs are in open/review state:

  • #3320 — Critical dependency bump for WhatsApp connectivity (whatsmeow)
  • #3321 — Agent context reordering to enable prefix caching optimization
  • #3222 — DeltaChat refactor (-200 LOC, cleanup, security hardening)
  • #3193 — New SimpleX channel implementation

4. Community Hot Topics

Item Type Activity Core Need
#3287 Issue 4 comments IRCv3 long-message reassembly — Users need transparent handling of fragmented IRC messages (>512 bytes) as single logical units
#3302 Issue 2 comments OAuth 2.1 for MCP servers — Alignment with evolving auth standards for Model Context Protocol integrations
#3320 PR Recent update WhatsApp client version block — Urgent: WhatsApp rejecting outdated client (405 error), channel non-functional without bump
#3321 PR Recent update Prefix caching efficiency — Reordering dynamic context to preserve cache hits across requests

Analysis: Protocol compatibility (WhatsApp, IRC, MCP) dominates immediate needs. Performance optimization (caching, CPU) shows maturing focus on production efficiency.

5. Bugs & Stability

Severity Issue Status Fix PR
Critical WhatsApp channel dead — client outdated (405) Open #3320 (ready)
High CPU spike on chat input focus (web) Closed (#3292) Fixed in unreleased code
Medium IRC message fragmentation breaks cohesion Open (#3287) None yet

Note: The WhatsApp blocker is production-breaking for affected users; #3320 should be prioritized for merge.

6. Feature Requests & Roadmap Signals

Request Signal Strength Likelihood for Next Version
OAuth 2.1 for MCP (#3302) Explicit roadmap alignment checkbox ticked; references prior issue #2546 High — standards compliance
SimpleX channel (#3193) PR open 43 days; new protocol support Medium — pending review
DeltaChat modernization (#3222) PR open 37 days; security & cleanup focus Medium — technical debt reduction
IRC long-message support (#3287) Stale tag but active discussion Low-Medium — niche but requested

Prediction: OAuth 2.1 MCP support and WhatsApp fix are highest-probability inclusions. SimpleX/DeltaChat depend on maintainer review capacity.

7. User Feedback Summary

  • Pain points: WhatsApp non-functional (blocker), CPU spike on input focus (resolved), IRC message splitting breaks UX
  • Use cases: Multi-protocol bridging (IRC, WhatsApp, DeltaChat, SimpleX, MCP), AI agent context optimization
  • Sentiment: Constructive — users file detailed issues with env specs; maintainers close bugs promptly (#3292 closed in 15 days)
  • Unmet needs: Modern auth (OAuth 2.1), protocol parity for emerging networks (SimpleX)

8. Backlog Watch

Item Age Risk Action Needed
#3222 37 days Stale refactor; security-relevant (drops password config) Review/merge — reduces attack surface, modernizes DeltaChat
#3193 43 days New channel type; expands protocol coverage Review/merge — first SimpleX support
#3287 18 days Stale feature request; IRC power-user need Triage — assign or clarify scope
#3302 10 days OAuth 2.1 spec alignment; referenced prior work Prioritize — standards-driven, likely low-effort if #2546 done

Maintainer attention needed: Two month-old PRs (#3222, #3193) represent completed work blocked only by review. The WhatsApp fix (#3320) is time-sensitive due to upstream API enforcement.

NanoClaw — qwibitai/nanoclaw

NanoClaw Project Digest — 2026-08-09

1. Today's Overview

NanoClaw saw moderate maintenance activity in the last 24 hours with 8 issue updates and 6 PR updates, but no new release. The project is actively resolving platform integration bugs (Discord approvals, Signal attachments, Google Chat message-ID handling) and advancing multi-channel support (Mattermost, Telegram rich messages, Strava MCP). Three PRs were merged — notably remote HTTP/SSE MCP server support and Strava integration — signaling continued expansion of the MCP/skills ecosystem. Several open issues point to architectural friction around secret management, provider event typing, and skill documentation drift.

2. Releases

No new releases published today. The main branch continues to receive fixes and features; next release will likely bundle Discord approval fix, Mattermost channel, Telegram rich rendering, and MCP remote-server support.

3. Project Progress — Merged/Closed PRs Today

PR Type Summary Impact
#2776 Feature Remote HTTP/SSE MCP servers now supported via new McpServerRemoteConfig (type, url, headers, instructions) and CLI flags --type, --url, --header High — enables hosted/managed MCP endpoints; unblocks SaaS MCP integrations
#2777 Feature /add-strava skill: official Strava MCP over HTTP transport + host-side OAuth flow with token refresh Medium — adds fitness-data MCP; demonstrates remote-MCP pattern
#3199 Feature (superseded) Mattermost v2 ChannelAdapter implementation (community chat-adapter-mattermost) Medium — replaced by #3202 which is now the active PR

4. Community Hot Topics

Item Activity Underlying Need
#3201 Discord approval buttons broken 2 comments, closed same day Critical UX blocker: owners cannot approve config updates; root cause = \n in custom_id parsing (fix in #3185)
#3177 SQLite lock contention on Docker mounts 1 comment, closed Stability on macOS/Linux: 29k+ readonly errors; fixed by switching journal mode to WAL for cross-mount filesystems
#3206 Google Chat attachments dropped 0 comments, new Channel compatibility: isSafeAttachmentName rejects / in message IDs; blocks Google Chat file ingestion
#3205 Persistent group-scoped OneCLI secrets 0 comments, new Architectural clarity: two contradictory secret-assignment models exist; needs persistent per-group vault mapping

5. Bugs & Stability — Ranked by Severity

Severity Issue Status Fix PR
Critical #3201 Discord approvals always reject (owner clicks ignored) Closed #3185 (open, strips \n from custom_id)
High #3206 Google Chat inbound attachments silently dropped Open None yet
High #2528 Signal images/PDFs unreachable in agent container Open (since May) None
Medium #3203 Codex provider emits undeclared file event → typecheck fail + dropped images Open None
Medium #3177 SQLite lock contention on VirtioFS (macOS/Linux Docker) Closed Fixed via WAL journal mode

6. Feature Requests & Roadmap Signals

Signal Source Likelihood for Next Release
Mattermost channel #3202 (open PR) High — follows Slack adapter pattern; supersedes stale #546
Telegram rich messages (Bot API 10.1) #2877 (open PR) Medium — uses sendRichMessage; needs review
Remote HTTP/SSE MCP servers #2776 merged Done — will ship in next release
Persistent group-scoped OneCLI secrets #3205 Medium — design fork must be resolved first
Strava MCP integration #2777 merged Done — demonstrates remote-MCP pattern

7. User Feedback Summary

  • Discord admins blocked: approval workflow unusable until #3185 merges; immediate pain for multi-user groups.
  • Google Chat users: file attachments vanish silently — no error, no visibility (#3206).
  • Signal users (long-standing): media files arrive on host but container cannot read them (#2528, 3 months open).
  • Developers/maintainers: skill documentation drift (#3204 — add-opencode still references deleted Dockerfile blocks) and provider typing gaps (#3203) slow onboarding.

8. Backlog Watch — Stale/Important Items Needing Attention

Item Age Why It Matters
#2528 Signal attachments unreachable in container ~3 months Blocks media-rich Signal workflows; no PR, no recent movement
#3205 Persistent group-scoped OneCLI secrets New but architectural Two contradictory secret models exist; resolution required before multi-tenant hardening
#3203 Codex provider file event undeclared New Breaks typecheck on main; drops generated images silently
#3204 add-opencode skill doc drift New Skill guard test asserts old Dockerfile shape; misleads contributors
#2877 Telegram rich rendering PR ~6 weeks Large PR, awaiting review; enables modern Telegram UX

Health Indicators: ✅ Active merging of MCP/channel features; ✅ Critical Discord bug has fix PR; ⚠️ Several high-severity bugs lack fixes; ⚠️ Architectural secret-management fork unresolved; ⚠️ Long-standing Signal attachment issue untouched.
Next-release candidates: Discord approval fix, Mattermost channel, remote MCP support, Strava skill, Telegram rich messages (if reviewed).

NullClaw — nullclaw/nullclaw

No activity in the last 24 hours.

IronClaw — nearai/ironclaw

IronClaw Project Digest — 2026-08-09

1. Today's Overview

IronClaw shows high velocity with 50 PRs and 25 issues updated in the last 24 hours. The project is in the midst of a major "Reborn" architectural migration — 23 issues were closed today, nearly all tied to Reborn workstreams (product workflows, channel porting, CLI/TUI migration, capability lifecycle, approval parity). Meanwhile, 18 PRs remain open, several of them XL-sized (web push notifications, progressive Slack previews, presence-based shared conversations, inspector completion, stress testing). No new releases were cut. The merge rate (32 PRs merged/closed) indicates healthy throughput, but the concentration of XL PRs suggests reviewers are under pressure.

2. Releases

No new releases published today.

3. Project Progress — Merged/Closed PRs (Selected)

PR Title Area Status
#7377 feat!: a run acts as its invoker — remove shared-route subject binding Auth/Identity, Slack & Telegram Merged — Core identity model change; enables owner ≠ actor runs
#7382 feat(stress): scripted tool-call workload with durable write read-back Stress testing, API capacity Merged — Adds --api-scripted-tool mode for deterministic tool sequences
#7280 test(inspector): add browser, security, and operator coverage Web Debug Inspector Merged — Security & browser test coverage for ?debug=true inspector
#7389 fix(live-qa): verify triggered Slack delivery through the two-lane contract Slack delivery, Live QA Merged — Fixes post-#7157 regression in delivery verification
#7393 test(disclosure): measure the Core delivery pair in the wide-catalog benchmark Benchmarks, Outbound Merged — Updates disclosure benchmark for Core-tier tools
#6938 fix(skills): the model chooses the skill, not a keyword scorer Skills, Agent Loop Merged — Removes host-side keyword scoring; model-driven activation via builtin.skill_activate
#7029 fix(product): restore durable delivery claim Outbound delivery Merged — Restores CAS-based Prepared → Sending claim as sole authority
#4118 Reborn CLI provider add/login parity Reborn CLI Closed — Tracks provider registry/auth parity using ironclaw_llm crate

Key advances: Identity model unification (owner ≠ actor), model-driven skill selection, stress-test infrastructure, inspector hardening, outbound delivery durability, and Reborn CLI parity.

4. Community Hot Topics — Most Active Items

Item Type Comments Signals
#3280 Issue 7 Reborn ProductWorkflow facade — Central piece connecting ProductAdapters to host-layer services; 7 related issues show high coordination need
#6989 Issue 5 Token accounting bug — ModelWorkRequest estimates from reference string length, not actual content; part of P1 pi-harness adoption
#7398 PR — Web Push + PWA — XL PR making web app a first-party notification channel (VAPID, RFC 8030/8291/8292); parity with Slack/Telegram
#7397 PR — Presence-based shared conversations — Builds on #7377's acting-identity ladder for Slack & Telegram
#7396 PR — Generic progressive previews for Slack — Channel-neutral contract mapped to chat.startStream/appendStream/stopStream
#7291 PR — Inspector completion — Stats, navigation, localization, stream health metrics

Underlying needs:

  • Reborn completion — The 23 closed Reborn issues signal a push to finish the migration.
  • Multi-channel parity — Web push, Slack progressive previews, shared conversations all aim for feature parity across channels.
  • Observability — Inspector and stress-test work address operator/debugging gaps.
  • Token accounting correctness — Critical for cost control and model routing.

5. Bugs & Stability — Today's Reports

Severity Item Description Fix PR
High #6989 Token estimation uses content_ref.as_str().len() (reference string) instead of referenced content None yet
High #7395 TOCTOU race in claim_delivery_attempt_for_send + failed-row misclassification Open PR #7395
Medium #7352 Gate projection identity collision (run-notification:approval:<run_id> duplicates across gates) Open PR #7352
Medium #7389 Live QA Slack delivery verification broken post-#7157 (retired triggered-run-delivery outcome) Merged #7389
Low #7341 WebUI scoped attachment reads & SSE tests broken after fetch/ReadableStream migration Open PR #7341
Low #7048 WASM guest diagnostics not sanitized before tracing (log injection risk) Open PR #7048 (depends on #7063)

Note: Several bug-fix PRs are open and XL-sized (#7395, #7048), indicating non-trivial fixes.

6. Feature Requests & Roadmap Signals

Feature Evidence Likelihood for Next Version
Web Push / PWA notifications PR #7398 (XL, core contributor) High — Near completion, first-party channel parity goal
Progressive Slack previews PR #7396 (XL, core contributor) High — Channel-neutral contract designed
Presence-based shared conversations (Slack/Telegram) PR #7397 (XL, core contributor) High — Builds on merged #7377 identity model
Web Debug Inspector (operator-only) Issue #7218 (epic), PR #7291, #7280 High — Tests merging, feature behind ?debug=true
Model-driven skill selection Merged #6938 Done — Shipped
Reborn approvals parity (approve once / deny / always allow) Issue #4539 (epic) Medium — Epic tracked, core substrate exists
Declarative capability policy (local-dev grants) Issue #4120 Medium — Config-as-code dependency
Generic channel progressive preview contract PR #7396 Medium — Slack first, then extensible
Stress-test scripted tool workloads Merged #7382 Done — Infrastructure landed

7. User Feedback Summary

No direct user comments in the provided data (all 👍: 0). Inferred pain points from issue/PR content:

Pain Point Source
Skill installation appears to succeed but skill disappears PR #7171: "installed: true but gone from Settings, unactivatable"
Token accounting wildly inaccurate Issue #6989: estimating from reference string length
Live QA delivery verification broken after identity change PR #7389: scheduled lane failing every run since #7157
Gate notification collisions block multi-gate runs PR #7352: identical projection IDs for approval/auth gates
WebUI attachment reads & SSE tests regressed PR #7341: legacy Playwright suites incompatible with new transport
WASM guest logs can inject into tracing PR #7048: unsanitized diagnostics
Reborn CLI provider add/login missing parity Issue #4118: can't manage providers via CLI post-LLM config wiring

Satisfaction signals: High contributor velocity, systematic Reborn closure, operator tooling (Inspector) investment.
Dissatisfaction signals: Regression rate (multiple "fix regression" PRs), XL PR backlog, token accounting bug in P1 path.

8. Backlog Watch — Stale / Needing Attention

Item Age Why It Matters
#3484 ~90 days EPIC: Reborn Contributor Runway — Enables parallel skill/tool/channel porting; blocked on stable runway
#4470 ~65 days Refactor ironclaw_reborn_composition — Oversized crate owning auth, WebUI, Slack, OAuth, extensions; CI-enforced boundaries needed
#4088 ~75 days Decompose oversized Reborn integration files — 4 files called out in PR #4066 review; technical debt
#4091 ~75 days Production/multi-tenant extension lifecycle wiring — Currently only local single-user; blocks production reuse
#3905 ~80 days Safe user-scoped tool installs — Scoped model exists for skills but not tools; security boundary
#3571 ~85 days Refactor HostHttpEgressService to accept Arc<dyn SecretStore> — Boilerplate delegation; API ergonomics
#7028 6 days fix(outbound): preserve terminal status during recovery — Small, open, dependency for #7029 (merged)
#7048 6 days fix(wasm): sanitize guest diagnostics — XL, depends on #7063, security-relevant

Maintainer action suggested: Prioritize #4470 (composition crate split) and #4088 (file decomposition) to unblock parallel Reborn work; land #7028/#7048 to clear security/stability debt.

*Data source: GitHub API snapshot for nearai/i

⚠️ Nội dung vượt quá giới hạn của GitHub Issue, xem báo cáo đầy đủ trong file Markdown đã commit.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    openclaw-en

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions