v1.9.1

What's Changed

• Add 05 skills/plugins examples to test-examples by @enyst in #1773
• Release v1.9.0 by @all-hands-bot in #1775
• Added more fixes for parsing discriminated unions by @tofarr in #1779

Full Changelog: v1.9.0...v1.9.1

v1.9.0

Added support for experimental critic model

critic-demo.mp4

What's Changed

• Propagate Datadog conversation logging flag to evaluation dispatch by @simonrosenberg in #1703
• Release v1.8.2 by @all-hands-bot in #1722
• Add swebenchmultimodal support to run-eval workflow by @juanmichelini in #1659
• feat: add multiswebench support to run-eval workflow by @juanmichelini in #1693
• feat: load skills from agent server by @hieptl in #1729
• Add GPT-5.2 high reasoning to model configurations by @neubig in #1735
• Replace mocked MCP tests with real integration tests by @neubig in #1678
• feat(plugin): Add Plugin.fetch() for remote plugin fetching and caching by @jpshackelford in #1647
• Fix event ordering in RemoteEventsList by inserting events sorted by timestamp by @neubig in #1737
• fix: move async operations outside sync lock in event subscription by @neubig in #1732
• Remove unsupported models by @simonrosenberg in #1734
• refactor(hooks): add typed fields to HookConfig for better type safety by @xingyaoww in #1726
• Add Marketplace datamodel for marketplace.json support by @neubig in #1744
• Remove push_to_index option from run-eval.yml workflow by @juanmichelini in #1750
• feat: add Gemini 3 models to reasoning effort supported list by @Chesars in #1752
• Fix DiscriminatedUnionEnvParser to use single parser when only one kind exists by @tofarr in #1741
• Add GitHub repository URLs to PyPI package metadata by @xingyaoww in #1753
• Fix cache tag truncation with ports and latest suffix by @simonrosenberg in #1626
• docs: rewrite CONTRIBUTING around SDK architecture by @enyst in #1755
• Fix CORS to allow DOCKER_HOST_ADDR for remote browser access by @shanemort1982 in #1466
• refactor(examples): use .from_dict() for hooks example by @xingyaoww in #1746
• fix: Add JSON formatter for uvicorn access logs by @neubig in #1733
• Fix unhandled ConversationRunError causing non-JSON logs by @neubig in #1680
• remove gpt-5-mini from list of models to evaluate for openhands index by @simonrosenberg in #1766
• Add API-Based Critic for Real-Time Agent Action Evaluation (Experimental) by @xingyaoww in #1269
• feat: Add .pr/ directory convention and auto-cleanup workflow by @jpshackelford in #1764
• feat(delegate): Add create_sub_visualizer method for custom sub-agent visualization by @malhotra5 in #1767
• Add minimax/MiniMax-M2.1 to resolve_model_config.py by @juanmichelini in #1757
• feat: Support full class names in DiscriminatedUnionEnvParser by @tofarr in #1768
• Update idle time on bash, git, and file operations by @neubig in #1770
• Add set_security_analyzer abstract method to BaseConversation by @malhotra5 in #1772

New Contributors

• @shanemort1982 made their first contribution in #1466

Full Changelog: v1.8.2...v1.9.0

v1.8.2

What's Changed

• Release v1.8.1 by @all-hands-bot in #1670
• feat: auto-trigger PR review on new ready-for-review PRs + fix duplicate comments by @xingyaoww in #1669
• Add --no-pager instruction for git commands in system prompt by @xingyaoww in #1673
• fix: mask LLM_API_KEY and GITHUB_TOKEN secrets in PR review agent by @xingyaoww in #1675
• Add per-test timeout to prevent hanging example scripts by @xingyaoww in #1666
• Fix PR review agent multi-line suggestion alignment by @xingyaoww in #1677
• Add visualize() for ConversationErrorEvent by @enyst in #1686
• Fix useless logging in sockets.py by @neubig in #1679
• Fix conversation restore failing due to secret serialization/deserialization mismatch by @tofarr in #1672
• fix(critic): Accept MessageEvent as valid finish signal in AgentFinishedCritic by @neubig in #1695
• fix: add file-based locking to EventLog.append() by @ixchio in #1614
• feat(pr-review): Add priority labels to review comments by @xingyaoww in #1696
• feat: support custom volumes mounting for DockerWorkspace by @Co1lin in #1618
• Add AGENTS.md and link repo.md by @enyst in #1690
• Condenser integration tests by @csmith49 in #1652
• Fix: Disable streaming for sub-agents in delegate tool by @malhotra5 in #1705
• Fix deepseek-v3.2-reasoner model name to use valid API identifier by @neubig in #1707
• fix(condenser): Forgetting range calculation interprets keep_first correctly by @csmith49 in #1708
• llm: log failed Responses calls with request context by @enyst in #1684
• Fix verify method to include builtin tools in event check by @malhotra5 in #1710
• Optimize docker-like behaviour in apptainer with better code quality by @adityasoni9998 in #1711
• Remove .openhands/skills/repo.md symlink by @enyst in #1709
• Add crash diagnostics logging to agent server by @neubig in #1689
• Fix PR review bot to diff against current base commit by @enyst in #1685
• Add ARG USERNAME to base-image stage for podman by @cbagwell in #1691
• Revert "fix(critic): Accept MessageEvent as valid finish signal in AgentFinishedCritic" by @xingyaoww in #1715
• Add configurable startup wait for APIRemoteWorkspace by @simonrosenberg in #1713
• Add user tracking to evaluation workflow by @juanmichelini in #1716
• fix: route qwen-3-coder through Fireworks provider by @neubig in #1720

New Contributors

• @Co1lin made their first contribution in #1618

Full Changelog: v1.8.1...v1.8.2

v1.8.1

Fix the failed PyPI release on v1.8.0

What's Changed

• Release v1.8.0 by @all-hands-bot in #1663
• fix: make publish workflow idempotent for re-runs by @xingyaoww in #1667
• feat: Add inline review comments for PR review workflow by @xingyaoww in #1654
• fix(tests): Enable accurate coverage reporting with pytest-xdist by @jpshackelford in #1658

Full Changelog: v1.8.0...v1.8.1

v1.8.0

Major Feature

• Support for AgentSkills standard

What's Changed

• Add push_to_index dispatch flag by @simonrosenberg in #1588
• feat: support .mcp.json for MCP server configuration by @neubig in #1481
• Release v1.7.4 by @all-hands-bot in #1595
• feat: support resource directories (scripts/, references/, assets/) by @neubig in #1482
• Fix stats being reset when resuming a conversation by @xingyaoww in #1591
• Add dynamic run name title to Run Eval workflow by @juanmichelini in #1612
• refactor: remove reconciliation methods, use runtime agent directly by @enyst in #1542
• fix(git): skip remote check for repos with no commits by @neubig in #1606
• Add code review microagent based on xingyaoww's review history by @xingyaoww in #1150
• feat(sdk): complete hooks implementation with additional context and stop hook by @ak684 in #1547
• feat: Add Plugin data model and basic loading from directories by @neubig in #1611
• feat: Add plugin loading example demonstrating skills, hooks, and MCP by @neubig in #1616
• Expose runtime_resource_factor in run-eval by @simonrosenberg in #1605
• feat: add skill validation improvements by @neubig in #1597
• Add skills support to PR review example with /codereview and /codereview-roasted by @xingyaoww in #1617
• feat: add to_prompt() for XML skill prompt generation by @neubig in #1483
• Fix poetry lock command for Poetry 2.x compatibility by @xingyaoww in #1622
• feat: Add skills example and include skill location in agent prompt by @neubig in #1599
• Multi-step integration tests by @csmith49 in #1613
• Add action summary feature for tool calls by @xingyaoww in #1339
• Add expected index models to resolve_model_config.py by @juanmichelini in #1497
• Revert "Expose runtime_resource_factor in run-eval" by @simonrosenberg in #1627
• fix: resolve UnsupportedFieldAttributeWarning for secret_registry field by @neubig in #1625
• Fix context window displaying as 0 when resuming conversation by @xingyaoww in #1590
• fix: Update PR review workflow to use correct repository name by @xingyaoww in #1623
• Fix AGENTS.md not being loaded when skills directories don't exist by @xingyaoww in #1624
• fix: Pass agent_context at Agent initialization in PR review script by @xingyaoww in #1632
• Allow over-riding system-level configurations from apptainer.conf in apptainer workspace by @adityasoni9998 in #1628
• feat(llm): Set default LLM timeout to 5 minutes (300s) by @neubig in #1638
• feat(pr_review): send complete git diff to agent in first message by @xingyaoww in #1639
• fix: Auto-add skills to_prompt() output to system prompt by @neubig in #1642
• feat: Add include_default_tools option to control built-in tools by @xingyaoww in #1594
• fix: remove duplicate example numbers in 02_remote_agent_server by @xingyaoww in #1637
• Use AGENTS.md instead of repo.md in SDK examples/system prompt by @enyst in #1648
• feat: Add public sandbox_id field to OpenHandsCloudWorkspace for resuming existing sandboxes by @malhotra5 in #1603
• docs: Update environment configuration for secret encryption in Agent Server by @jagzmz in #1465
• feat: add utility to convert legacy OpenHands skills to AgentSkills format by @xingyaoww in #1643
• feat: Implement AgentSkills progressive disclosure for SKILL.md files by @xingyaoww in #1644
• Add alive property to RemoteWorkspace by @malhotra5 in #1655
• Expose max_retries input and dispatch to eval by @simonrosenberg in #1641

New Contributors

• @jagzmz made their first contribution in #1465

Full Changelog: v1.7.4...v1.8.0

v1.7.4

What's Changed

• Release v1.7.3 by @all-hands-bot in #1551
• Update OpenHands version bump workflow to include all required steps by @malhotra5 in #1553
• fix: Wait for PyPI propagation and add Slack notification to release workflow by @xingyaoww in #1552
• feat: support SKILL.md file convention and name validation by @neubig in #1480
• feat: add pause() and resume() methods to workspace classes by @ixchio in #1539
• build(deps): Bump libtmux to >=0.53.0 for v0.51.0 hard deprecations by @tony in #1561
• feat: add DeepSeek v3.2 model support by @ixchio in #1564
• refactor: consolidate duplicated blocking wait logic in RemoteConversation.run() + remove defensive if else by @simonrosenberg in #1569
• Fix DiscriminatedUnionMixin issues with Pydantic C bindings by @tofarr in #1555
• Add test to verify working_dir standardization (closes #211) by @xingyaoww in #1576
• Re-add discriminator field to discriminated union JSON schema by @tofarr in #1578
• fix: Cap thinking budget below max_tokens by @csmith49 in #1580
• Relax needs-triage duration by @enyst in #1567
• Fix RemoteConversation polling on terminal errors by @simonrosenberg in #1572
• More test fixes by @tofarr in #1581
• remove custom_llm_provider from LLM config by @enyst in #1583
• Update cryptography to 46.0.3 by @enyst in #1585
• feat: relax tool matching on resume by @ixchio in #1538
• Set initial execution status to error if it was running by @tofarr in #1554
• More efficient websocket based implementation for async bash execution by @tofarr in #1587
• feat: Create conversation when provided ID doesn't exist by @malhotra5 in #1579
• fix(condenser): Retry on empty condensation by @csmith49 in #1577
• Bump pydantic by @tofarr in #1593

New Contributors

• @tony made their first contribution in #1561

Full Changelog: v1.7.3...v1.7.4

v1.7.3

What's Changed

• feat: Add support for custom tools with remote agent server by @xingyaoww in #1383
• Release v1.7.2 by @all-hands-bot in #1530
• Add ApptainerWorkspace implementation for rootless container support by @neubig in #892
• Add GPT-5 preset using ApplyPatchTool (opt-in) by @enyst in #1462
• feat: add openrouter/minimax-m2 to SEND_REASONING_CONTENT_PATTERNS by @neubig in #1152
• Add forward_env support to APIRemoteWorkspace by @simonrosenberg in #1540
• Raise ValueError when condensing 0 events by @xingyaoww in #1541
• Propagate remote conversation errors by @simonrosenberg in #1546
• Fix save_meta() not being called on conversation creation by @tofarr in #1548
• Fix ValueError from tool execution causing conversation to fail by @xingyaoww in #1550

Full Changelog: v1.7.2...v1.7.3

v1.7.2

What's Changed

• Release v1.7.1 by @all-hands-bot in #1513
• Load model-specific repo instructions (CLAUDE.md for Claude only, GEMINI.md for Gemini only) by @enyst in #1328
• Fix max iterations reached: Set ERROR status with clear message by @simonrosenberg in #1511
• Save screenshots from browser tool by @hamiltop in #1443
• Use ALLHANDS_BOT_GITHUB_PAT for bot-like workflows by @xingyaoww in #1519
• Fix TmuxTerminal.close() to handle dead session exceptions by @xingyaoww in #1517
• ci(integration): update integration LLM matrix to gpt-5.2-codex by @enyst in #1503
• Add automated version bump PRs for downstream repos by @xingyaoww in #1520
• feat: add early stopping/pruner for behavior tests cost optimization by @ixchio in #1433
• Fix condensation for 0 events by adjusting default parameters by @xingyaoww in #1523
• Add pull request template by @neubig in #1528
• Revert "ci(integration): update integration LLM matrix to gpt-5.2-codex" by @xingyaoww in #1524

New Contributors

• @all-hands-bot made their first contribution in #1513
• @hamiltop made their first contribution in #1443
• @ixchio made their first contribution in #1433

Full Changelog: v1.7.1...v1.7.2

v1.7.1

What's Changed

• feat: add AgentSkills standard fields to Skill model by @neubig in #1479
• Release v1.7.0 by @xingyaoww in #1486
• Fix SDK_SHA reference in run-eval.yml Comment on PR step by @simonrosenberg in #1498
• Add optional parallelization parameters to run-eval workflow by @simonrosenberg in #1499
• Add compatibility fixes for Nemotron and similar models by @pankajarm in #1470
• feat(condenser): Token-aware condensation in LLMSummarizingCondenser by @csmith49 in #1380
• fix(condenser): Tool-call aware condensation by @csmith49 in #1412
• feat(localFileStore): Add Cache Layer to LocalFileStore to Avoid I/O Overhead from Read Operations by @CLFutureX in #1274
• feat: Add hooks system for event interception by @ak684 in #1467
• fix: handle union types in py_type for mcp schema conversion by @hieptl in #1509
• Remove tool preambles from gpt-5-codex by @enyst in #1512
• fix(condenser): Tool-loop aware condensation by @csmith49 in #1508

New Contributors

• @pankajarm made their first contribution in #1470

Full Changelog: v1.7.0...v1.7.1

v1.7.0

What's Changed

• Release 1.6.0 by @xingyaoww in #1404
• Make thinking signatures optional by @cbagwell in #1394
• Make integration-test label run integration tests only by @ryanhoangt in #1416
• make eval_limit param optional by @simonrosenberg in #1419
• Exclude '*mini' models from prompt_cache_retention by @enyst in #1345
• Add load_project_skills helper function by @xingyaoww in #1420
• feat(AsyncExecutor): AsyncExecutor allows adding tasks after shutdown, which may cause unpredictable exceptions. by @CLFutureX in #1092
• fix end span in local conversation run by @Rainhunter13 in #1432
• Add iterative refinement example for COBOL to Java refactoring by @neubig in #1415
• Make cronjob run integration test only by @ryanhoangt in #1430
• Add commit0 to CI evaluation workflow by @simonrosenberg in #1429
• delete image when docker workspace closes by @madhur-tandon in #1424
• Include persistence_dir in ConversationRunError for better debugging by @xingyaoww in #1437
• Fix JSON encoding to preserve non-ASCII characters in model_dump_json by @LuneZ99 in #1431
• refactor: add support for description field in sdk secrets by @hieptl in #1428
• include j2 templates in tools package by @madhur-tandon in #1426
• Fix: throw error when viewing PDF files instead of crashing by @xingyaoww in #1423
• Update litellm to 1.80.10 and add tests for Gemini 2.5 thinking blocks by @xingyaoww in #1413
• Suppress LiteLLM INFO logs to clean up output by @xingyaoww in #1369
• Make stuck detection thresholds configurable by @neubig in #763
• feat: Add OpenHandsCloudWorkspace for OpenHands Cloud API by @xingyaoww in #1442
• Add PostHog Debugging Example by @neubig in #1202
• Public APIs return OpenHands types not LiteLLM types - Part 2: SystemPromptEvent by @jpshackelford in #495
• refactor: Add LookupSecret tests for AgentContext.get_secret_infos() by @hieptl in #1456
• Add nova-2-lite to model_features by @BalasubramanyamEvani in #1455
• fix: disable scheduled workflows from running in forks by @Chesars in #1447
• Add skill for writing behavior tests by @ryanhoangt in #1459
• feat: Add configurable security policy support to Agent by @neubig in #427
• Fix batch atomicity when condensation forgets ObservationEvents by @xingyaoww in #1450
• fix: add openrouter/minimax to FORCE_STRING_SERIALIZER_MODELS by @neubig in #1448
• Fix verified_models by @enyst in #1458
• fix: deepseek reasoner model not part of send_reasoning_content_support models (Closes #1343) by @BalasubramanyamEvani in #1464
• Implement gemini-style file editing tools by @neubig in #1199
• terminal: populate TerminalObservation.exit_code (fixes #1468) by @enyst in #1469
• Fix RemoteConversation.run() timeout issues with async polling by @xingyaoww in #1444
• Add accumulated token usage to integration tests output and GitHub CI comments by @xingyaoww in #1471
• Fix prepare-release workflow to use ALLHANDS_BOT_GITHUB_PAT by @xingyaoww in #1485
• Fix prepare-release workflow to use correct label names by @xingyaoww in #1487

New Contributors

• @cbagwell made their first contribution in #1394
• @Rainhunter13 made their first contribution in #1432
• @madhur-tandon made their first contribution in #1424
• @BalasubramanyamEvani made their first contribution in #1455

Full Changelog: v1.6.0...v1.7.0