Enable ACPAgent on RemoteRuntime API by simonrosenberg · Pull Request #2190 · OpenHands/software-agent-sdk

simonrosenberg · 2026-02-23T19:48:06Z

Summary

Polymorphic agent deserialization in EventService so ACPAgent payloads work
Eager ACPAgent import in api.py to register in discriminated union
Agent lifecycle cleanup (close() on AgentBase, called from LocalConversation)
Pre-install claude-code-acp and codex-acp in Docker image
Add remote runtime example for ACPAgent

Test plan

Unit tests pass (72/72): test_acp_agent.py + test_agent_loading.py
CI image build (triggered by this PR)
SWE-bench validation with ACPAgent on 5 instances

Agent Server images for this PR

• GHCR package: https://github.com/OpenHands/agent-sdk/pkgs/container/agent-server

Variants & Base Images

Variant	Architectures	Base Image	Docs / Tags
java	amd64, arm64	`eclipse-temurin:17-jdk`	Link
python	amd64, arm64	`nikolaik/python-nodejs:python3.13-nodejs22`	Link
golang	amd64, arm64	`golang:1.21-bookworm`	Link

Pull (multi-arch manifest)

# Each variant is a multi-arch manifest supporting both amd64 and arm64
docker pull ghcr.io/openhands/agent-server:3dcb3dd-python

Run

docker run -it --rm \
  -p 8000:8000 \
  --name agent-server-3dcb3dd-python \
  ghcr.io/openhands/agent-server:3dcb3dd-python

All tags pushed for this build

ghcr.io/openhands/agent-server:3dcb3dd-golang-amd64
ghcr.io/openhands/agent-server:3dcb3dd-golang_tag_1.21-bookworm-amd64
ghcr.io/openhands/agent-server:3dcb3dd-golang-arm64
ghcr.io/openhands/agent-server:3dcb3dd-golang_tag_1.21-bookworm-arm64
ghcr.io/openhands/agent-server:3dcb3dd-java-amd64
ghcr.io/openhands/agent-server:3dcb3dd-eclipse-temurin_tag_17-jdk-amd64
ghcr.io/openhands/agent-server:3dcb3dd-java-arm64
ghcr.io/openhands/agent-server:3dcb3dd-eclipse-temurin_tag_17-jdk-arm64
ghcr.io/openhands/agent-server:3dcb3dd-python-amd64
ghcr.io/openhands/agent-server:3dcb3dd-nikolaik_s_python-nodejs_tag_python3.13-nodejs22-amd64
ghcr.io/openhands/agent-server:3dcb3dd-python-arm64
ghcr.io/openhands/agent-server:3dcb3dd-nikolaik_s_python-nodejs_tag_python3.13-nodejs22-arm64
ghcr.io/openhands/agent-server:3dcb3dd-golang
ghcr.io/openhands/agent-server:3dcb3dd-java
ghcr.io/openhands/agent-server:3dcb3dd-python

About Multi-Architecture Support

Each variant tag (e.g., 3dcb3dd-python) is a multi-arch manifest supporting both amd64 and arm64
Docker automatically pulls the correct architecture for your platform
Individual architecture tags (e.g., 3dcb3dd-python-amd64) are also available if needed

- Polymorphic agent deserialization in EventService (type(self.stored.agent) instead of hardcoded Agent) - Eager import of ACPAgent in api.py to register in discriminated union - Add close() lifecycle method to AgentBase, call from LocalConversation.close() - Await conversation close in EventService to ensure subprocess cleanup - Pre-install claude-code-acp in Docker image for ACPAgent support - Add remote runtime example for ACPAgent Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

github-actions · 2026-02-23T19:48:56Z

API breakage checks (Griffe)

Result: Passed

Action log

all-hands-bot

🟢 Good taste - Elegant solution that eliminates special cases

KEY INSIGHT: The polymorphic deserialization (agent_cls = type(self.stored.agent)) is textbook good design - it preserves type information naturally without conditionals, making ACPAgent work seamlessly alongside the base Agent class.

What I like:

Data structure wins: Using Python's type system to handle polymorphism instead of switch statements
Proper lifecycle management: The close() hook is clean extensibility (no-op by default, override where needed)
Correctness fix: The await on run_in_executor (line 609) - you actually need to wait for the cleanup
Solves real problem: ACPAgent subprocess cleanup is necessary, not theoretical

Minor notes:

The eager import in api.py is a necessary hack for Pydantic discriminated unions - acceptable pragmatism
Could add a type guard on agent_cls, but given stored.agent is pre-validated, this is safe

✅ LGTM - Core logic is sound, changes are minimal and well-targeted.

openhands-agent-server/openhands/agent_server/event_service.py

openhands-sdk/openhands/sdk/agent/base.py

openhands-agent-server/openhands/agent_server/api.py

The claude-code-acp npm install fails on golang and java base images that don't have npm. Make it conditional on npm availability. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

…yloads With ACPAgent registered in the discriminated union, agent payloads now require an explicit "kind" field. Updated 6 test payloads and renumbered the ACP example from 07 to 09 to avoid duplicate. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

github-actions · 2026-02-23T20:06:48Z

Coverage Report •

File	Stmts	Miss	Cover	Missing
openhands-agent-server/openhands/agent_server
api.py	160	12	92%	75, 87, 102, 108, 269, 272, 276–278, 280, 286, 327
event_service.py	320	81	74%	55–56, 74–76, 85–89, 92–95, 115, 219, 236, 290–291, 295, 303, 306, 352–353, 369, 371, 375–377, 381, 390–391, 393, 397, 403, 405, 413–418, 555, 557–558, 562, 576–578, 580, 584–587, 591–594, 602–605, 625, 629–634, 646–647, 649–650, 657–658, 660–661, 665, 671, 688–689
openhands-sdk/openhands/sdk/agent
acp_agent.py	389	78	79%	194–196, 250–253, 255–256, 283, 285, 289, 295, 306–307, 312, 379, 481–482, 493, 498, 529, 539, 544, 555–558, 564–566, 569–571, 573, 575–576, 578, 580, 585, 594–595, 599–600, 604, 611–617, 627–632, 634, 643–645, 648–649, 655–659, 661, 663–664, 672, 709, 713–714, 958–959
base.py	193	22	88%	200, 257–259, 289, 293–297, 345–347, 357, 367, 375–376, 486, 523–524, 534–535
openhands-sdk/openhands/sdk/conversation/impl
local_conversation.py	400	26	93%	288, 293, 321, 364, 382, 398, 463, 641–642, 645, 797, 805, 807, 811–812, 823, 825–827, 852, 924, 1050, 1054, 1124, 1131–1132
TOTAL	20509	5226	74%

The CI Agent Server workflow builds binary target images, not source. Update the example to use target_type="binary" accordingly. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Add 'agent_type' input (default/acp) to run-eval.yml and pass it through to the evaluation repo's eval-job workflow dispatch payload. This enables dispatching SWE-bench evaluations using ACPAgent (Claude Code via ACP). Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

SWE-bench prebaked images use Python-only base images without Node.js. The conditional `if command -v npm` silently skipped claude-code-acp installation. Now installs Node.js from nodesource when npm is absent. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

ACPAgent now emits a FinishAction ActionEvent + ObservationEvent after each step() completes. This makes it compatible with evaluation frameworks (SWE-bench, etc.) that detect task completion by scanning for FinishAction in the event history. Without this, the benchmarks fake_user_response loop didn't detect completion and sent up to 10 "please continue" messages, and the AgentFinishedCritic marked results as failed. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Claude Code ACP defaults to "ask" mode for tool permissions when no settings file exists. This causes the agent to hang waiting for interactive permission approval in headless/eval environments. Add /etc/claude-code/managed-settings.json with allow rules for Edit, Read, and Bash tools so Claude Code can operate without human approval. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Claude Code's internal permission system blocks tools like Write when running in default mode, causing the agent to hang waiting for interactive approval. This happens even when the ACP server's settings allow the tools, because Claude Code checks permissions before making MCP tool calls. Call set_session_mode("bypassPermissions") after creating the session so Claude Code skips all permission checks in headless mode. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Add acp_prompt_timeout field (default 1800s) to ACPAgent that wraps the prompt() call with AsyncExecutor's timeout. This prevents the eval from hanging forever when claude-code-acp fails to send the JSON-RPC response after completing its work. Root cause analysis: claude-code-acp's prompt() method awaits sessionUpdate notifications end-to-end through the stdout pipe. If the response is never sent (e.g., due to stdout backpressure or an unhandled result subtype), the Python ACP client blocks indefinitely. Also fixes test assertions for FinishAction emission added in d72fd08. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

The default asyncio.StreamReader limit (64 KiB) is too small for ACP session_update notifications that carry large tool-call outputs. When a JSON-RPC line exceeds the limit, readline() raises LimitOverrunError which silently kills the filter pipeline, leaving the prompt() future unresolved forever and eventually deadlocking the subprocess stdout pipe. Root cause confirmed on v10 eval: 59,496 bytes of unread data stuck in the subprocess pipe, both processes sleeping in ep_poll. Increase both the subprocess pipe and filtered_reader limits to 100 MiB. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

- Auto-detect ACP server type from InitializeResponse.agent_info.name - Map session modes: claude-code → bypassPermissions, codex-acp → full-access - Add acp_session_mode field for manual override - Install codex-acp alongside claude-code-acp in Docker image - Add acp-codex to run-eval.yml agent_type choices Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

codex-acp requires an explicit `conn.authenticate()` call after initialize and before session creation. Auto-detect the auth method from environment variables (CODEX_API_KEY or OPENAI_API_KEY). Without this, codex-acp returns "Authentication required" when trying to create a session in the remote runtime container. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Aligns naming with 'acp-codex': both now follow the acp-{provider} pattern, making the agent type self-documenting. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

- Add elapsed time tracking to _record_usage() via add_response_latency() for both main prompt and fork (ask_agent) paths - Capture agent_version from InitializeResponse.agent_info alongside agent_name - Expose agent_name and agent_version as public properties on ACPAgent so benchmarks can include ACP server identity in eval output Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

The _BYPASS_MODE_MAP only matched "claude-code" but the actual agent name is "@zed-industries/claude-agent-acp", which doesn't contain that substring. Add "claude-agent" as an additional key so the mode correctly resolves to "bypassPermissions" instead of falling through to the default "full-access" (which claude-agent-acp rejects). Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

The npm package was renamed from @zed-industries/claude-code-acp to @zed-industries/claude-agent-acp. Update all references: bypass mode map key, Dockerfile, examples, tests, and docstrings. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

github-actions · 2026-02-27T12:32:19Z

Agent server REST API breakage checks (OpenAPI)

Result: Failed

Log excerpt (first 1000 characters)

{"asctime": "2026-03-11 20:32:47,987", "levelname": "WARNING", "name": "openhands.agent_server.config", "filename": "config.py", "lineno": 173, "message": "\u26a0\ufe0f OH_SECRET_KEY was not defined. Secrets will not be persisted between restarts."}
::error title=openhands-agent-server REST API::Breaking REST API change detected without MINOR version bump (1.13.0 -> 1.13.0).

Breaking REST API changes detected compared to baseline release:
- added '#/components/schemas/ACPAgent-Output, #/components/schemas/Agent-Output' to the '/items/anyOf[subschema #1: ConversationInfo]/agent' response property 'oneOf' list for the response status '200'
- the '/items/anyOf[subschema #1: ConversationInfo]/agent' response's property type/format changed from 'object'/'' to ''/'' for status '200'
- removed the required property '/items/anyOf[subschema #1: ConversationInfo]/agent/kind' from the response with the '200' status
- removed the required property '/items/anyOf[subschema #1: ConversationInfo]/age

Action log

all-hands-bot

🟢 Good taste - Pragmatic engineering that solves real problems.

The complexity growth (auth methods, bypass modes, timeout handling, 100 MiB buffer) is justified - each addresses actual production issues rather than theoretical edge cases. The one-shot ACPAgent design (emitting FinishAction per step) correctly models how ACP servers work internally.

Trade-offs are reasonable: 100 MiB stream buffers solve real LimitOverrunError crashes for large tool outputs; Docker managed-settings.json bypass is documented and appropriate for eval environments. Good engineering.

Add ANTHROPIC_BASE_URL, ANTHROPIC_API_KEY, OPENAI_BASE_URL, and OPENAI_API_KEY environment variables to the run-examples workflow. This allows Claude Code and Codex ACP servers to route their API calls through the LiteLLM proxy using the existing LLM_API_KEY, eliminating the need to provision separate Anthropic/OpenAI API keys. See: OpenHands/evaluation#297 Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

The test framework requires each example to emit an EXAMPLE_COST marker to stdout. Add cost reporting using agent.llm.metrics for the standalone ACP example and conversation_stats for the remote ACP example. Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

Remove backward-compatible ANTHROPIC_API_KEY requirements from ACP examples. Instead, route all ACP agent requests through the LiteLLM proxy by setting ANTHROPIC_BASE_URL and ANTHROPIC_API_KEY from the existing LLM_BASE_URL and LLM_API_KEY environment variables. This approach: - Eliminates the need for separate Anthropic/OpenAI API key secrets - Enables full cost tracking through LiteLLM - Provides model flexibility (can test with different backends) - Centralizes observability and rate limiting Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

When ACP prompt() raises, the agent now: 1. Emits a ConversationErrorEvent (in addition to the existing MessageEvent) so RemoteConversation._get_last_error_detail() can report the actual error instead of the generic fallback message 2. Tags usage/content policy errors with code "UsagePolicyRefusal" 3. Re-raises the exception so LocalConversation.run() breaks the loop immediately instead of spinning until max_iteration_per_run Also fixes pre-existing ruff E501 lint error on line 385. Fixes OpenHands/benchmarks#495 Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Resolve conflict in acp_agent.py: keep agent_name/agent_version properties from feat/acp-remote-runtime and adopt Generator[LLM] return type from main. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Allows callers to specify which model the ACP server should use. The model is passed via session _meta.claudeCode.options.model when creating a new ACP session. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

new_session() assigns **kwargs directly to the _meta field. Passing _meta={"claudeCode": ...} wraps it in an extra _meta key, producing {"_meta": {"_meta": {"claudeCode": ...}}} — the ACP server never sees the model config. Fix: unpack the meta dict as direct kwargs so they become _meta content. Also skip claudeCode meta for non-Claude ACP servers (e.g. codex-acp). Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

The step() error handler now re-raises after emitting events and emits both a MessageEvent and a ConversationErrorEvent. Update the test to expect the re-raise (pytest.raises) and verify both emitted events. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

The ACPAgent uses a sentinel LLM with model="acp-managed" that cannot make direct LLM calls. When generate_title() was called for ACP conversations, it would fall back to this sentinel LLM and fail with: "LiteLLM Provider NOT provided. You passed model=acp-managed" Add a guard to detect the sentinel model and pass None to generate_conversation_title(), which causes it to use simple truncation fallback instead of LLM-based title generation. Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

all-hands-bot · 2026-03-11T12:24:57Z

[Automatic Post]: It has been a while since there was any activity on this PR. @simonrosenberg, are you still working on it? If so, please go ahead, if not then please request review, close it, or request that someone else follow up.

Add automatic retry for ACP prompt failures caused by transient connection errors (OSError, ConnectionError, BrokenPipeError, EOFError). Changes: - Wrap prompt() call in retry loop for connection exception types - Retry up to 3 times with exponential backoff (5s, 15s, 30s) - Configurable via ACP_PROMPT_MAX_RETRIES env var - Reset client accumulators between retries - Timeout errors are NOT retried (handled separately) This preserves session state when connection errors occur, avoiding the need to restart instances from scratch in the evaluation framework. Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

enyst · 2026-03-13T23:18:19Z

This PR actually failed the OpenAPI check:

Fix Python selection in version-bump PR workflow #2430 (comment)
the workflow has a comment here: Enable ACPAgent on RemoteRuntime API #2190 (comment)

I'll fix it to fail more loudly. I'm not sure if it's time to make it mandatory for merging, I think maybe we should, because it's too easy to miss only in a comment... 🤔

@simonrosenberg Please correct me if I see wrong, serialization, and the REST schema, are broken from older clients?

We have a policy to give a deprecation period (~5 releases) before we make breaking changes to public APIs. 🤔

Maybe we could revert this PR, and rebuild it with deprecation markers?

enyst · 2026-03-13T23:30:56Z

Failures log: OpenAPI checks

enyst · 2026-03-13T23:31:59Z

I'll fix it to fail more loudly.

Done

enyst · 2026-03-13T23:43:18Z

Hi — I’m OpenHands-GPT-5.4.

I looked back at this PR in light of the OpenAPI breakage that later surfaced.

What broke

This PR changed the REST contract of an existing agent-server endpoint in a breaking way: the agent request/response shape stopped being a plain object and became a polymorphic oneOf union (Agent-Output / ACPAgent-Output), so fields that older clients could previously rely on (kind, llm, etc.) were no longer guaranteed in the old shape.

So the break here was not primarily “an endpoint was removed without deprecation.” It was a breaking request/response contract change on an existing endpoint without a deprecation runway.

Relevant policy lines

I opened a follow-up policy PR to make this explicit: #2433.

Relevant line references there:

AGENTS.md:114-122 — repo-wide rule that public REST contract breaks need a deprecation notice and a runway of 5 minor releases.
openhands-agent-server/AGENTS.md:17-24 — agent-server REST API is public/backward-compatible, and all REST contract breaks need a deprecation notice plus a 5 minor release runway.
openhands-agent-server/AGENTS.md:50-69 — incompatible request/response schema changes need a migration path, and endpoints / legacy contracts should only be removed after that runway.

Practical options to handle this kind of change

A few reasonable paths would have been:

Keep the old contract working in place
- Make the change additive.
- Keep accepting / returning the old shape while introducing ACP support.
Add a parallel field
- Keep the old agent shape stable.
- Introduce a new field for the new polymorphic contract.
- Deprecate the old field and keep it for the runway.
Add a versioned endpoint or versioned contract
- Keep the old endpoint/contract stable.
- Introduce a new version for the ACP-capable shape.
- Mark the old one deprecated and keep it around for the runway.
If you truly must break immediately, do a MINOR bump
- That would satisfy the existing CI policy better than shipping the break under the same minor version.
- But with the stronger policy in Clarify REST contract deprecation policy #2433, the expectation would still be to provide a deprecation notice + migration runway, not just the version bump.

Bottom line

The core issue here was changing an existing REST contract incompatibly without a migration/deprecation path for clients. The follow-up policy PR (#2433) codifies the intended expectation more clearly: all REST contract breaks need a deprecation notice and a runway of 5 minor releases.

This reverts merge commit ca621a4. Co-authored-by: openhands <openhands@all-hands.dev>

Co-authored-by: openhands <openhands@all-hands.dev>

all-hands-bot approved these changes Feb 23, 2026

View reviewed changes

simonrosenberg and others added 2 commits February 23, 2026 16:57

Fix Dockerfile: conditional npm install for non-Node.js base images

a01d2d8

The claude-code-acp npm install fails on golang and java base images that don't have npm. Make it conditional on npm availability. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

simonrosenberg marked this pull request as draft February 23, 2026 20:04

simonrosenberg and others added 14 commits February 23, 2026 17:37

Fix example: use binary target type to match CI-built images

aa3bad9

The CI Agent Server workflow builds binary target images, not source. Update the example to use target_type="binary" accordingly. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Fix line length in AgentBase.close() docstring

dd7737b

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

rename agent_type 'acp' to 'acp-claude' for consistency

3d9803a

Aligns naming with 'acp-codex': both now follow the acp-{provider} pattern, making the agent type self-documenting. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

simonrosenberg force-pushed the feat/acp-remote-runtime branch from 4176666 to b9c42fd Compare February 24, 2026 13:55

simonrosenberg marked this pull request as ready for review February 27, 2026 12:31

Merge branch 'main' into feat/acp-remote-runtime

33aef1b

all-hands-bot approved these changes Feb 27, 2026

View reviewed changes

simonrosenberg and others added 2 commits February 27, 2026 09:47

keep main branch behavior - dont add await

d23c4dc

Merge branch 'main' into feat/acp-remote-runtime

ae237b1

simonrosenberg self-assigned this Feb 27, 2026

simonrosenberg and others added 9 commits March 5, 2026 10:56

Merge branch 'main' into feat/acp-remote-runtime

e096099

Resolve conflict in acp_agent.py: keep agent_name/agent_version properties from feat/acp-remote-runtime and adopt Generator[LLM] return type from main. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Add acp_model field to ACPAgent for model passthrough via _meta

f131e90

Allows callers to specify which model the ACP server should use. The model is passed via session _meta.claudeCode.options.model when creating a new ACP session. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

simonrosenberg force-pushed the feat/acp-remote-runtime branch from d0127c4 to bda77a6 Compare March 11, 2026 19:13

Fix ruff-format: wrap long assert line in test_acp_agent.py

fade5af

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

simonrosenberg mentioned this pull request Mar 11, 2026

Add ACPAgent remote runtime example docs OpenHands/docs#392

Merged

1 task

Merge branch 'main' into feat/acp-remote-runtime

0f438b5

simonrosenberg merged commit ca621a4 into main Mar 13, 2026
28 checks passed

simonrosenberg deleted the feat/acp-remote-runtime branch March 13, 2026 21:33

VascoSch92 mentioned this pull request Mar 13, 2026

Eval dispatch broken: agent_type input not accepted by evaluation workflow #2431

Closed

enyst mentioned this pull request Mar 13, 2026

Fix Python selection in version-bump PR workflow #2430

Merged

enyst added a commit that referenced this pull request Mar 15, 2026

Revert "Enable ACPAgent on RemoteRuntime API (#2190)"

6a72c27

This reverts merge commit ca621a4. Co-authored-by: openhands <openhands@all-hands.dev>

enyst mentioned this pull request Mar 15, 2026

Revert PR #2190: Enable ACPAgent on RemoteRuntime API #2451

Merged

all-hands-bot mentioned this pull request Mar 15, 2026

chore(agent-server): bump version for REST API breakage checks #2456

Closed

5 tasks

enyst added a commit that referenced this pull request Mar 16, 2026

Revert PR #2190: Enable ACPAgent on RemoteRuntime API (#2451)

ed924a3

Co-authored-by: openhands <openhands@all-hands.dev>

This was referenced Mar 16, 2026

Enable ACPAgent on RemoteRuntime API via v2 conversations contract #2461

Closed

feat(agent-server): reintroduce ACP remote runtime via v2 conversations API #2462

Closed

Enable ACPAgent on RemoteRuntime API via ACP conversation endpoints #2465

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Enable ACPAgent on RemoteRuntime API#2190

Enable ACPAgent on RemoteRuntime API#2190
simonrosenberg merged 40 commits intomainfrom
feat/acp-remote-runtime

simonrosenberg commented Feb 23, 2026 •

edited by github-actions bot

Loading

Uh oh!

github-actions bot commented Feb 23, 2026 •

edited

Loading

Uh oh!

all-hands-bot left a comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

github-actions bot commented Feb 23, 2026 •

edited

Loading

Uh oh!

github-actions bot commented Feb 27, 2026 •

edited

Loading

Uh oh!

all-hands-bot left a comment

Uh oh!

all-hands-bot commented Mar 11, 2026

Uh oh!

Uh oh!

enyst commented Mar 13, 2026 •

edited

Loading

Uh oh!

enyst commented Mar 13, 2026

Uh oh!

enyst commented Mar 13, 2026

Uh oh!

enyst commented Mar 13, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Conversation

simonrosenberg commented Feb 23, 2026 • edited by github-actions bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Test plan

Uh oh!

github-actions bot commented Feb 23, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

API breakage checks (Griffe)

Uh oh!

all-hands-bot left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

github-actions bot commented Feb 23, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

github-actions bot commented Feb 27, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Agent server REST API breakage checks (OpenAPI)

Uh oh!

all-hands-bot left a comment

Choose a reason for hiding this comment

Uh oh!

all-hands-bot commented Mar 11, 2026

Uh oh!

Uh oh!

enyst commented Mar 13, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

enyst commented Mar 13, 2026

Uh oh!

enyst commented Mar 13, 2026

Uh oh!

enyst commented Mar 13, 2026

What broke

Relevant policy lines

Practical options to handle this kind of change

Bottom line

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

simonrosenberg commented Feb 23, 2026 •

edited by github-actions bot

Loading

github-actions bot commented Feb 23, 2026 •

edited

Loading

github-actions bot commented Feb 23, 2026 •

edited

Loading

github-actions bot commented Feb 27, 2026 •

edited

Loading

enyst commented Mar 13, 2026 •

edited

Loading