Repository navigation
docs(verify): map maintenance 2026-11-17 - #6931
Conversation
…TS clients, public-skills switch, message edits and mentions Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013ArwwKRe5R7V5KLRSjxeU5
…6926 and #6927 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013ArwwKRe5R7V5KLRSjxeU5
…and decision cards Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013ArwwKRe5R7V5KLRSjxeU5
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013ArwwKRe5R7V5KLRSjxeU5
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013ArwwKRe5R7V5KLRSjxeU5
…ram output A grep observation repeats the absolute workspace path, so a long SDK_VERIFY_HOME pushed the matched file past the 400-character clip and S14's search bullet failed on the run directory's length. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013ArwwKRe5R7V5KLRSjxeU5
…nothing was added Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013ArwwKRe5R7V5KLRSjxeU5
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013ArwwKRe5R7V5KLRSjxeU5
… by #6866) Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013ArwwKRe5R7V5KLRSjxeU5
REST API breakage checks (OpenAPI) — ✅ PASSEDResult: ✅ PASSED |
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013ArwwKRe5R7V5KLRSjxeU5
all-hands-bot
left a comment
There was a problem hiding this comment.
This review was posted by an AI agent (OpenHands).
Reviewed the current head 5d6c8c9888231b0de2d4df7101eae59dd657ec93 (docs/chore: daily feature-map maintenance for main@0275e714, per #5751). The change belongs in this repository — it updates only the in-repo verify skills (verify-agent-server, verify-sdk), their feature maps/READMEs, and new verify programs; no production code is touched.
Verification performed on this head:
control-agent-server map check→ ok (38 files, 1046 IDs, no problems);map coverage→ ok (219 routes, 209 owned, 209 driven, 0 unowned, 0 owned-but-not-driven, no stale exclusions);sdk-map check→ ok (21 files, 496 IDs, 34 untested) — all match the PR description's claims exactly.- Cross tests re-run locally:
tests/cross/test_verify_sdk_skill.py,test_control_sdk.py,test_verify_agent_server_skill.py→ 98 passed. - Unit tests cited by the new/changed bullets re-run and passing: ask_user/message_edit/conversation_mentions/mention/MCP (86 passed), terminal
test_observation_wait.py(21 passed), goal tests (20 passed). All 432 test-file citations across both maps resolve to existing files; spot-checked cited test functions all exist. - New map content grounded against the code at this head:
POST /api/conversations/{id}/mcp(mcp_router), theask_user/answerroute and its 400/409/404/422/429 behavior,POST .../events/batch(event_read_router),pending_ask400 guards on/goaland/goal/resume(event_service.py), skill-body trimming on socket/webhook frames (event_service.py),OH_PUBLIC_SKILLS_ENABLED/OH_PUBLIC_CATALOG_ENABLEDand the marketplacedisabledflag (skills_router.py,plugins_router.py),GoalAsked/find_pending_ask,AskUserTool/EditMessageTool/mention tools withOH_INTERNAL_SERVER_URL,add_mcp_servers/mcp_startup_budget,max_observation_wait_seconds(incl.isfinitevalidation), and the TypeScript client methods cited by F34 (batchGetEvents,loadPlugin,answerAskUser,getCommits/getCommitChanges,downloadArchive/downloadArchiveBuffer,InitClientwithX-Init-API-Key,ConversationManager.git/AgentServerClient.git) — all exist and behave as the map documents. The false-value list for the env switches matches the pydanticTypeAdapter(bool)parsing (false/off/0/nooff; unrecognized keeps it on). - README family counts match actual bullet counts for every edited family (e.g. F19 34→41 with the 7 new IDs; new F40 with 10 bullets; SDK S04/S07/S09/S14/S16 counts match); both baselines move to main@0275e714 as proposed.
- CI on this exact head is green (29 success, 1 skipped), including the OpenAPI breakage check and cross-tests.
- Issue references check out: #6926 closed (fix #6928), #6927 open (fix #6929), #6907 open; the map marks them as known bugs with xfails/failures accordingly, consistent with the SDK-side linked pairs.
One non-blocking observation, not introduced by this PR: a handful of IDs appear twice within their family files (e.g. F07.reject-non-user, F14.home, F34.version-gate-old) where a Gotcha bullet reuses a sub-feature ID; these duplicates pre-exist at the base and map check passes on them (the README counts unique sub-features), so no action is required in this PR.
No material issues found; the map updates are accurate against the code, the check/coverage gates pass, and the CI is green.
✅ APPROVED
HUMAN:
Engel Nyst (@enyst) set up a daily automation that keeps the verify-agent-server and verify-sdk feature maps current. This PR is its run for main@0275e71, after PR #6923.
AGENT:
I'm an AI agent helping Engel Nyst (@enyst) with project work.
Why
35 commits merged since the maps' baseline (2aae48b). They own paths in 35 of 37 agent-server families and 20 of 21 SDK families, and they add three routes no family owned, so this run is a full pass of both maps rather than a delta.
Summary
Agent-server map: full pass, 2aae48b..0275e71
agent_context.skills: [](perf(agent-server): leave skill bodies out of full-state snapshots #6881). Map drift, fixed.SystemPromptEventcan miss a slow server (perf: start conversations without waiting for slow MCP servers #6877). Map drift, fixed; the second failure was a cascade.Requires:, now fixed.ask_agentstats event lands just after the call returns. The recipe now waits for it.ask_user, feat: ask_user tool for decision cards, behind an experimental feature #6852/fix: preserve pending user questions across completion and goal loops #6902): 10 sub-features, with REST, socket and TypeScript-client entry points.POST /api/conversations/{id}/mcp(Add runtime MCP server updates for modern protocol #6802) and slow or failing MCP startup (perf: start conversations without waiting for slow MCP servers #6877).OH_PUBLIC_SKILLS_ENABLED,OH_PUBLIC_CATALOG_ENABLED,disabledandload_registered_marketplaces(feat(skills): allow disabling default public skills and catalogs #6739).https://github.com/enyst/openhands-evidence/blob/8f33a10f673cae03559f03fd95f40a3bd65cc658/software-agent-sdk/map-maintenance/2026-11-17/agent-server-report-edited.md)).The one fail is F07.error-non-json, the [Bug]: Session socket can send an error frame before sync while a run holds the state lock #6745 race; it passed 3/3 on re-run.map checkok (1046 IDs).map coverageok: 0 unowned, 0 owned but not driven.SDK map: full pass, 2aae48b..0275e71
https://github.com/enyst/openhands-evidence/blob/8f33a10f673cae03559f03fd95f40a3bd65cc658/software-agent-sdk/map-maintenance/2026-11-17/sdk-report-main.md)).Actionmodels droppattern,constandadditionalPropertiesbefore the request #6766, [Bug]: RemoteWorkspace rejects a pathlib.Path working_dir that LocalWorkspace accepts #6654, [Bug]: ACP ask_agent loses the fork's cost and stalls 2 s waiting for a UsageUpdate it dropped #6646, [Bug]: ACP npx cache warm uses --prefer-offline with an outdated cache, hides npm's error, and startup fails with 'Connection closed' #6809 and [Bug]: ACP agent re-sends a timed-out prompt until stuck detection or max iterations #6645. Each reproduced twice today and fails the same way at BASE; their dates are refreshed.S04.goal-asked, plusS04.goal-asked-branch= known failure Bug: run_goal returns an unanswered question from an abandoned branch #6907).add_mcp_servers, the startup budget, andS09.runtime-add-resume= known failure [Bug]: Runtime-added MCP server tools lose their server prefix after a restart or resume #6927.programs/sdk_verify.py: keeps tool observations up to 2000 characters. A long run path cutsrc/calc.pyout of S14's grep check; this was a harness gap.https://github.com/enyst/openhands-evidence/blob/8f33a10f673cae03559f03fd95f40a3bd65cc658/software-agent-sdk/map-maintenance/2026-11-17/sdk-report-edited.md)).All 23 are the known failures above.sdk-map checkok (496 IDs, 34 untested).tests/cross/test_verify_sdk_skill.py,test_control_sdk.pyandtest_verify_agent_server_skill.py: 98 passed.Both maps
Same promise:orServer:links yet. The twin promises driven on both surfaces today agree:F19.conversation-mcp-add-restartxfail andS09.runtime-add-resumefail.F40.goal-abandoned-branchxfail andS04.goal-asked-branchfail.F19.conversation-mcp-add-start-failure). The SDK (S09.runtime-add-servers) raises an MCP error by design, so the gap is the route's error mapping.POST /api/conversations/{id}/mcpanswers 500 when an added server cannot start ([evidence](https://github.com/enyst/openhands-evidence/blob/d7e013006c972cafff76c39a95c9356d9baedc6d/software-agent-sdk/bugs/2026-11-17/mcp-add-start-failure-500/README.md)).https://github.com/enyst/openhands-evidence/blob/d7e013006c972cafff76c39a95c9356d9baedc6d/software-agent-sdk/bugs/2026-11-17/mcp-runtime-prefix-lost/README.md))./goal. Openbugissues since BASE:LocalConversation.add_mcp_servers, so whichever merges second may need a small conflict resolution.ask_user(feat: ask_user tool for decision cards, behind an experimental feature #6852): Canvas has no decision-card UI. Follow-up Show the agent's ask_user questions as decision cards OpenHands#19773, open PR feat(chat): render experimental decision cards OpenHands#19789.MessageEditEventyet. Follow-up Show the agent's message edits in Canvas (experimental message_edits) OpenHands#19793.[Title](conversation:<id>)as a link with an emptyhref. Filed [Feature]: Render conversation mentions (conversation:<id> links) as links to the conversation OpenHands#19814, with the screenshot below.disabledcatalogs are not used by Canvas yet. Users still get correct behavior; there is no follow-up.Issue Number
#5751
How to Test
Evidence, pinned to enyst/openhands-evidence@8f33a10:
https://github.com/enyst/openhands-evidence/blob/8f33a10f673cae03559f03fd95f40a3bd65cc658/software-agent-sdk/map-maintenance/2026-11-17/agent-server-report-main.md)and [sdk-report-main.md](https://github.com/enyst/openhands-evidence/blob/8f33a10f673cae03559f03fd95f40a3bd65cc658/software-agent-sdk/map-maintenance/2026-11-17/sdk-report-main.md): every family at 0275e71 before the edits.https://github.com/enyst/openhands-evidence/blob/8f33a10f673cae03559f03fd95f40a3bd65cc658/software-agent-sdk/map-maintenance/2026-11-17/agent-server-report-edited.md)and [sdk-report-edited.md](https://github.com/enyst/openhands-evidence/blob/8f33a10f673cae03559f03fd95f40a3bd65cc658/software-agent-sdk/map-maintenance/2026-11-17/sdk-report-edited.md):edited families re-run independently.https://github.com/enyst/openhands-evidence/tree/8f33a10f673cae03559f03fd95f40a3bd65cc658/software-agent-sdk/map-maintenance/2026-11-17/edited/agent-server): transcripts of every changed bullet. For example, F40 is under20261117-091608-5350-map-f40, and F19 (with the [Bug]: POST /api/conversations/{id}/mcp answers 500 when an added MCP server fails to start #6926 and [Bug]: Runtime-added MCP server tools lose their server prefix after a restart or resume #6927 xfails) is under20261117-091608-5142-map-f19.https://github.com/enyst/openhands-evidence/tree/8f33a10f673cae03559f03fd95f40a3bd65cc658/software-agent-sdk/map-maintenance/2026-11-17/main/agent-server-fails)and [main/sdk-fails/](https://github.com/enyst/openhands-evidence/tree/8f33a10f673cae03559f03fd95f40a3bd65cc658/software-agent-sdk/map-maintenance/2026-11-17/main/sdk-fails): every failing bullet of the main replay.Video/Screenshots
Agent Canvas (OpenHands/OpenHands 5ed4ffa), launched on an agent server built from 0275e71:
POST /api/skills.[Skills modal](
https://raw.githubusercontent.com/enyst/openhands-evidence/d02df825d6b1ea00bcb23ca8e8be84b1ec340d7b/software-agent-sdk/map-maintenance/2026-11-17/ui/F07.skills-modal.png)href([Feature]: Render conversation mentions (conversation:<id> links) as links to the conversation OpenHands#19814).[Mention link](
https://raw.githubusercontent.com/enyst/openhands-evidence/d02df825d6b1ea00bcb23ca8e8be84b1ec340d7b/software-agent-sdk/map-maintenance/2026-11-17/ui/F07.mention-link.png)[Skills page](
https://raw.githubusercontent.com/enyst/openhands-evidence/d02df825d6b1ea00bcb23ca8e8be84b1ec340d7b/software-agent-sdk/map-maintenance/2026-11-17/ui/F18.page.png)[MCP page](
https://raw.githubusercontent.com/enyst/openhands-evidence/d02df825d6b1ea00bcb23ca8e8be84b1ec340d7b/software-agent-sdk/map-maintenance/2026-11-17/ui/F17.library-catalog.png)Type
Notes
Evidence lives in enyst/openhands-evidence, not under
.pr/.Generated by Claude Code
🐳 Agent Server images for this PR — GHCR package, pull/run commands, and all pushed tags (click to expand)
• GHCR package: https://github.com/OpenHands/agent-sdk/pkgs/container/agent-server
Variants & Base Images
eclipse-temurin:17-jdkpython-node-runtimepython-node-runtimepython-node-runtimegolang:1.21-bookwormPull (multi-arch manifest)
# Each variant is a multi-arch manifest supporting both amd64 and arm64 docker pull ghcr.io/openhands/agent-server:128290f-pythonRun
All tags pushed for this build
About Multi-Architecture Support
128290f-python) is a multi-arch manifest supporting both amd64 and arm64128290f-python-amd64) are also available if needed