agentscope-ai/AgentTeams v1.2.6

on GitHub

What's Changed

• fix(matrix): send-gate retrigger carries the placeholder root flag by @LUOSENGWA in #1393
• feat(teamharness): map loop_id across project state and task meta by @LUOSENGWA in #1401
• feat(events): signal emission — shared contract, Go five endpoints, Python task/loop emitters, list_events by @LUOSENGWA in #1405
• feat(server): expose projection freshness and strong-consistency reads by @LUOSENGWA in #1407
• fix(matrix): send-gate retrigger only fires on relevant room events by @LUOSENGWA in #1408
• feat(teamharness): structured illegal-transition errors + SIG-MISMATCH emission by @LUOSENGWA in #1404
• feat(storage): external S3-compatible storage backend (provider=s3) by @LUOSENGWA in #1387
• deps(controller): bump x/net and grpc for reachable advisories by @Portll in #1234
• docs(readme): add v1.2.5 news, sync ja-JP news, drop broken zh-CN image by @max-wc in #1399
• fix(manager-bridge): activate runtime ACL flags for the allowlist policy (#1388) by @LUOSENGWA in #1419
• fix(project-management): surface worker auto-join outcome (#1390) by @LUOSENGWA in #1420
• fix(openclaw): restrict elevated.allowFrom to trusted Matrix IDs by @nillikechatchat in #1425
• fix(upgrade): block upgrade when explicit --check-deps dry-run is unrunnable by @LUOSENGWA in #1398
• docs: bump pinned AGENTTEAMS_VERSION in all three READMEs to v1.2.3 by @RerankerGuo in #1370
• fix(matrix-channel): look up retryable errcodes via the live class name by @RerankerGuo in #1368
• fix(teamharness): whitelist markdown-wrapped placeholder templates in the artifact scan by @RerankerGuo in #1372
• chore(migrate): drop dead CMD_SOURCES map from skill analyze.sh by @RerankerGuo in #1374
• test(teamharness): fill the mcp_server fixture stubs needed by the full lifecycle by @RerankerGuo in #1373
• fix(matrix-channel): clamp sync_timeout_ms to a sane minimum by @RerankerGuo in #1369
• refactor(teamharness): split mcp/server.py into modules (no behaviour change) by @LUOSENGWA in #1452
• docs(coding-cli): taskflow visibility by @LUOSENGWA in #1410
• feat(controller): worker-level read-only MCP proxy (GET /api/v1/workers/{name}/mcp) by @LUOSENGWA in #1453
• docs(usage): single- and multi-user deployments share one deployment model by @LUOSENGWA in #1416
• feat(controller): add a worker-level read-only plugin registry proxy by @LUOSENGWA in #1454
• feat(plugins): release any manifest-carrying plugin from one workflow by @LUOSENGWA in #1455
• feat(teamharness): record structured answer payload on request_attention by @LUOSENGWA in #1449
• fix(teamharness): publish artifacts with an @ summary event (m.mentions) by @LUOSENGWA in #1435
• fix(matrix): propagate display flags from CRD to worker config by @LUOSENGWA in #1450
• feat: dual ledger — loop budget (Python) + control-action audit with kind filter by @LUOSENGWA in #1406
• feat(teamharness): attention record fields + notification outcome log (S0+S4a) (#1394) by @LUOSENGWA in #1418
• feat(plugins): add runtime adapter bundle skeletons by @LUOSENGWA in #1427
• fix(coding-cli): persist tab installs in the runtime's persisted home (QwenPaw working dir; OpenClaw/Hermes home) by @LUOSENGWA in #1409
• refactor(install): extract a safe, layered upgrade orchestrator (P1: preflight/backup/controller) by @LUOSENGWA in #1430
• fix(sync): push manifest protects unpushed writes (#1391) by @LUOSENGWA in #1417
• fix(coding-cli): adapt wrapper to qwen-code 0.25.3 --json-file semantics drift by @LUOSENGWA in #1397
• fix(controller): wake a sleeping team leader on mention by @RerankerGuo in #1462
• feat(install): support external Higress adapter and dashboard persistence by @nillikechatchat in #1396
• feat(plugins): add OpenClaw and Hermes adapters to the matrix bundle by @LUOSENGWA in #1458
• build(node): use latest Node 24 LTS (floor 24.22.1) by @LUOSENGWA in #1422
• build(openclaw): pin upstream v2026.10.6 || build(openclaw): pin upstream v2026.10.6 by @LUOSENGWA in #1460
• fix(matrix): finish final-mode cron reply delivery by @shiyiyue1102 in #1466
• test(qwenpaw): fix stale source tests and wire CI by @shiyiyue1102 in #1484
• fix(controller): sync Human grants to Worker allowlists by @benjamin-zhao-075 in #1456
• feat(server): read endpoint for signal events (#1405 follow-up) by @LUOSENGWA in #1433
• feat(plugins): implement OpenClaw and Hermes adapters for the teamharness bundle by @LUOSENGWA in #1459
• feat(controller): read-only project attention endpoint by @LUOSENGWA in #1434
• fix(build): derive the OpenClaw base tag from the pin, and assert the gateway version in the probe and at manager startup by @LUOSENGWA in #1479
• test(legacy-copaw): gate seeding on readiness and keep failure evidence by @LUOSENGWA in #1476
• feat(teamharness): pause and resume task dispatch and loop progression by @LUOSENGWA in #1402
• docs(zh-cn): add Chinese translation of worker-environment-and-gateway-probes by @RerankerGuo in #1378
• fix(controller): persist team human membership updates by @shiyiyue1102 in #1487
• fix(install): separate WASM plugin and runtime image registries by @shiyiyue1102 in #1488
• fix(gateway): repair custom LLM IP endpoints across bootstrap paths by @shiyiyue1102 in #1489
• feat(matrix): runtime-neutral core and Python adapters for the AgentTeams Matrix channel (#1468) by @LUOSENGWA in #1469
• feat(human): add per-human file area, trash, and quota by @LUOSENGWA in #1426
• fix(matrix): prevent access-denied reply loops between agents by @shiyiyue1102 in #1491
• fix(controller): grant admin power in existing team rooms by @shiyiyue1102 in #1490
• feat(teamharness): report per-recipient delivery gaps on attention notifications by @LUOSENGWA in #1495
• fix(ci): make the human-files job wait for RustFS, and stop publishing empty image tars by @LUOSENGWA in #1504
• feat(project-management): create-project.sh --team invites the whole Team by @dchaudhari7177 in #1429
• feat(teamharness): protocol guards for decision classes, attention stalls and review gates (#1451) by @LUOSENGWA in #1467
• fix(controller): atomically persist worker credential files by @shiyiyue1102 in #1517
• feat(runtimeconfig): per-runtime config surface types and registry by @LUOSENGWA in #1492
• feat(matrix): record silent inbound drops in a per-workspace ledger by @LUOSENGWA in #1497
• fix(controller): register the worker plugin registry route by @LUOSENGWA in #1525
• feat(runtimeconfig): openclaw config surface, member config routes, reconcile & rollback (P8a) by @LUOSENGWA in #1493
• feat(hermes): upgrade the Worker runtime to hermes-agent v2026.9.24 by @LUOSENGWA in #1472
• feat(workers): restart endpoints for workers and managers, and hot-apply the primary model by @LUOSENGWA in #1473
• fix(controller): worker list — single indexed team lookup + opt-in fields projection by @LUOSENGWA in #1477
• feat(controller): ETag/304 for stable spec projections (#1475) by @LUOSENGWA in #1480
• test(plugins): bundle L3/L4 lifecycle conformance (P-b / P-c) by @LUOSENGWA in #1530
• feat(controller): team-level attention inbox with lazy overdue events by @LUOSENGWA in #1503
• feat(human): file_sync capability + manifest endpoints (#1414 P3) by @LUOSENGWA in #1499
• chore(teamharness): harden loader smoke, contract guards and agent guardrails by @LUOSENGWA in #1523
• fix(dsh): backfill limited Matrix timelines by @bingdongni in #1528
• feat(controller): per-human model grants — Human.spec.modelAccess + ModelAccessTable by @LUOSENGWA in #1513
• feat(controller): presigned download URLs for human file area (?presign=1) by @LUOSENGWA in #1501
• feat(qwenpaw): upgrade the Worker and QwenPaw Manager runtimes to 2.2.2.post1 by @LUOSENGWA in #1471
• feat(sync): blocking flush barrier with per-file receipts (#1391) by @LUOSENGWA in #1500
• Add OpenClaw channel plugin bridging AgentTeams Matrix via the sidecar by @LUOSENGWA in #1470
• feat(human-files): presign uploads with server-side commit accounting by @LUOSENGWA in #1519
• feat(human-files): list the file area with size/etag and a recursive key cursor by @LUOSENGWA in #1518
• feat(runtimeconfig): hermes config surface, delegation projection, adapter & restart window (P8b) by @LUOSENGWA in #1494
• feat(teamharness): commit task metadata through the controller's conditional endpoint by @LUOSENGWA in #1498
• feat(controller): add module update endpoints for index, teams, workers and managers by @LUOSENGWA in #1526
• feat(matrix): add the bundle manifest and QwenPaw adapter package by @LUOSENGWA in #1524
• feat(human): workspace access grants and restrictions (P2) by @LUOSENGWA in #1502
• feat(plugin): QwenPaw × QwenCode bridge — dispatch, wake, worker-proxy approval mapping by @LUOSENGWA in #1411
• ci: verify downloaded image archives and retry once by @LUOSENGWA in #1534
• build(plugins): publish QwenPaw adapter zips with checksums and manifests by @LUOSENGWA in #1527
• fix(controller): reuse the shared team index in the MCP catalog by @LUOSENGWA in #1511
• feat(teamharness): accountable completion, room invites, report caps and review context by @LUOSENGWA in #1521
• deps(controller): move to Go 1.26 and bump modules for reachable advisories by @Portll in #1529
• feat(controller): team and human list summaries + keep initialPassword out of lists (#1475) by @LUOSENGWA in #1481
• feat(matrix): steer capability, optimistic config writes, isolation bundle and setup playbook by @LUOSENGWA in #1522
• fix(auth): reserve resource control for Manager and Admin by @shiyiyue1102 in #1536
• feat(observability): add optional native Worker session tracing by @shiyiyue1102 in #1532
• fix(oss): list directory entries through the SDK so modification times survive by @LUOSENGWA in #1539
• feat(files): add change-log based ?since= delta endpoint by @LUOSENGWA in #1541
• feat(meta): add GET project/task meta with ETag (read half of the CAS pair) by @LUOSENGWA in #1540
• perf(oss): serve the object hot path from the S3 SDK by @LUOSENGWA in #1537
• fix(oss): map store-availability faults to 503 storage_unavailable by @LUOSENGWA in #1538
• feat(controller): enforce per-human model grants on worker and coding-CLI writes by @LUOSENGWA in #1514
• test: repair 1.2.6 shell regression fixtures by @shiyiyue1102 in #1542
• fix(cli): make default Human creation use a valid permission level by @shiyiyue1102 in #1543
• fix(manager): guide users through Dashboard tracing setup by @shiyiyue1102 in #1546
• fix(tracing): derive team identity from projected runtime by @shiyiyue1102 in #1547
• fix(deepseek-harness): pin compatible Cordis service dependencies by @shiyiyue1102 in #1548
• fix(deepseek-harness): preserve streamed tool IDs for session replay by @shiyiyue1102 in #1549
• fix(deepseek-harness): avoid duplicate outbox file delivery by @shiyiyue1102 in #1550
• fix(deepseek-harness): apply projected model changes per turn by @shiyiyue1102 in #1551
• feat(files): add 304/Range/HEAD/If-Range to human file GET and manifest GET by @LUOSENGWA in #1544
• fix(humans): preserve initial credentials during concurrent updates by @shiyiyue1102 in #1554
• perf(sync): answer sync/status from manifest user metadata by @LUOSENGWA in #1545
• fix(install): harden controller upgrade port and wasm checks by @LUOSENGWA in #1553
• docs(modules): add the module update & distribution guide and refresh plugin docs by @LUOSENGWA in #1555
• fix(build): provide Hermes trace context for image publishing by @shiyiyue1102 in #1556
• fix(build): avoid emulated OpenClaw base compilation hangs by @shiyiyue1102 in #1558
• chore: prepare v1.2.6 component release pins by @shiyiyue1102 in #1552

New Contributors

• @Portll made their first contribution in #1234
• @benjamin-zhao-075 made their first contribution in #1456
• @bingdongni made their first contribution in #1528

Full Changelog: v1.2.5...v1.2.6

Docker Images

Multi-architecture images (amd64 + arm64):

# Embedded all-in-one (infra + controller; pulled by the installer)
docker pull higress-registry.cn-hangzhou.cr.aliyuncs.com/agentteams/agentteams-embedded:v1.2.6

# Manager (lightweight; spawned by the embedded controller)
docker pull higress-registry.cn-hangzhou.cr.aliyuncs.com/agentteams/agentteams-manager:v1.2.6

# Worker
docker pull higress-registry.cn-hangzhou.cr.aliyuncs.com/agentteams/agentteams-worker:v1.2.6

# QwenPaw Worker
docker pull higress-registry.cn-hangzhou.cr.aliyuncs.com/agentteams/agentteams-qwenpaw-worker:v1.2.6

# DeepSeek Harness Worker (experimental)
docker pull higress-registry.cn-hangzhou.cr.aliyuncs.com/agentteams/agentteams-deepseek-harness-worker:v0.1.1

# Controller (used standalone in k8s; bundled inside agentteams-embedded for docker installs)
docker pull higress-registry.cn-hangzhou.cr.aliyuncs.com/agentteams/agentteams-controller:v1.2.6

Quick Start

Latest stable (v1.2.6)

macOS / Linux

bash <(curl -fsSL https://raw.githubusercontent.com/agentscope-ai/AgentTeams/v1.2.6/install/agentteams-install.sh)

Windows (PowerShell 7+)

Set-ExecutionPolicy Bypass -Scope Process -Force; $wc=New-Object Net.WebClient; $wc.Encoding=[Text.Encoding]::UTF8; iex $wc.DownloadString('https://raw.githubusercontent.com/agentscope-ai/AgentTeams/v1.2.6/install/agentteams-install.ps1')

Documentation

For more details, see the installation guide.