inclusionAI/Avernet v2026.07.30

Avernet v2026.07.30

Repository: inclusionAI/Avernet

Tag: v2026.07.30

Published: 2026-08-01T07:39:24Z

Prerelease: no

Release notes:

Avernet v2026.07.30

> Release branch: REL20260730 · previous release: v2026.07.28

Overview

Avernet v2026.07.30 focuses on runtime reliability and production hardening across Skills Pool delivery, shared session files, bot lifecycle management, and multi-engine integrations.

This release makes Skills Pool promotion and recovery safer across OpenClaw, AICoding, and desktop runtimes; adds a direct engine data path for session files; improves collaborator and bot-chat ownership semantics; and prevents premature chat stream timeouts during long-running tasks.

Releases follow the vYYYY.MM.DD CalVer scheme.

Highlights

• Skills Pool reliability — resumable rollback, stronger cutover evidence,

fail-closed layout probes, safe AICoding repository retirement, parallel engine promotion, and improved handling of obsolete migration wakeups.

• Cross-engine skill delivery — centralized layout planning, logical Pool

mappings, hardened desktop runtime contracts, OpenClaw probe identity pinning, and owner-aware README resolution.

• Session files and workspaces — an engine-direct session-file data plane,

safer resource materialization, and explicit exclusion of generated ClawBench artifacts from OpenClaw workspace synchronization.

• Long-running task stability — chat streams no longer terminate prematurely

while OpenClaw is waiting for subagents or other long-running operations.

• Bot lifecycle and collaboration — template-based collaborator management,

restart-state cleanup, preserved template configuration and file ownership, and more reliable device connection routing.

• Isolation and API correctness — bot-chat metadata is scoped by trace owner,

notify polling is filtered by engine, Teclaw Cloud bots cannot become default bots, and AICoding file trees no longer expose internal Skills Pool paths.

Getting Started

See the documentation for installation and local setup:

• README
• Quick Start
• Docker Guide

Known Limitations

Skills Pool migration and session-file capabilities continue to roll out across heterogeneous engines. Runtime-specific configuration and migration contracts may still evolve in following releases.

Please report problems through GitHub Issues.

Feedback and Contributions

We welcome bug reports, feature requests, documentation improvements, and code contributions from the community.

What's Changed

View the complete list of pull requests merged into REL20260730

• Fs improve by @pfmiles in https://github.com/inclusionAI/Avernet/pull/554
• fix: prevent Teclaw Cloud from being default bot by @liyaqing-l in https://github.com/inclusionAI/Avernet/pull/555
• default use queue for bcn downlink by @xxxxpenny in https://github.com/inclusionAI/Avernet/pull/557
• fix(publish-flow): backport restart flag cleanup to REL20260730 by @evilcoda in https://github.com/inclusionAI/Avernet/pull/575
• feat(backend): cherry-pick draft restoration to REL20260730 by @ws121200 in https://github.com/inclusionAI/Avernet/pull/576
• fix: harden committed skills pool reconciliation by @FreddieSun in https://github.com/inclusionAI/Avernet/pull/570
• refactor(engine): centralize Skills layout planning by @XIEAN99 in https://github.com/inclusionAI/Avernet/pull/586
• feat(workspace): exclude clawbench_* files in openclaw rsync by @quhongwei in https://github.com/inclusionAI/Avernet/pull/590
• feat(skills): resolve logical pool mappings by @XIEAN99 in https://github.com/inclusionAI/Avernet/pull/587
• feat(skills): align Desktop Skills Pool delivery by @XIEAN99 in https://github.com/inclusionAI/Avernet/pull/588
• fix: negotiate pool cutover evidence contract by @FreddieSun in https://github.com/inclusionAI/Avernet/pull/584
• fix(session-resources): proxy materialization safely by @yhzhang35 in https://github.com/inclusionAI/Avernet/pull/593
• fix(workspace): explicitly exclude clawbench_results and clawbench_te... by @quhongwei in https://github.com/inclusionAI/Avernet/pull/598
• fix(backend): preserve template config on bot restart by @duxingxingstar in https://github.com/inclusionAI/Avernet/pull/589
• fix: fence committed pool probe failures by @FreddieSun in https://github.com/inclusionAI/Avernet/pull/595
• fix(backend): keep bot chat relation labels consistent by @3h-william in https://github.com/inclusionAI/Avernet/pull/601
• feat: session file engine direct data plane by @yhzhang35 in https://github.com/inclusionAI/Avernet/pull/597
• fix(engine): harden fresh Desktop Skills Pool rollout by @XIEAN99 in https://github.com/inclusionAI/Avernet/pull/604
• fix(backend): scope bot chat names by trace owner by @3h-william in https://github.com/inclusionAI/Avernet/pull/611
• Fs ctt type fix by @pfmiles in https://github.com/inclusionAI/Avernet/pull/613
• fix(backend): preserve draft restore file ownership by @ws121200 in https://github.com/inclusionAI/Avernet/pull/618
• Enable template-based member management for bot collaborators by @panqiwp in https://github.com/inclusionAI/Avernet/pull/600
• fix(backend): prefer the http_info target in _compose_device_conn_info by @pfmiles in https://github.com/inclusionAI/Avernet/pull/626
• fix(engine): prevent premature chat stream timeout for long-running tasks by @evilcoda in https://github.com/inclusionAI/Avernet/pull/623
• fix(backend): make Skills Pool rollback resumable by @XIEAN99 in https://github.com/inclusionAI/Avernet/pull/628
• fix: retire AICoding active repo corpus after pool cutover by @FreddieSun in https://github.com/inclusionAI/Avernet/pull/624
• Fs double env fix by @pfmiles in https://github.com/inclusionAI/Avernet/pull/633
• fix(backend): normalize Skills Pool runtime timestamps by @XIEAN99 in https://github.com/inclusionAI/Avernet/pull/635
• fix(community): prune skills-pool in aicoding file-tree by @maicaisongxiaocong in https://github.com/inclusionAI/Avernet/pull/639
• fix(engine): fail closed on invalid layout probes by @XIEAN99 in https://github.com/inclusionAI/Avernet/pull/641
• fix(community): filter notify polling by...