Record accepted SDK preparation and bind OpenHands fact workspace

[seathatflowsinourveins]

Scope

The merged SDK guide still said preparation review was pending after independent acceptance and publication. This update records the accepted source state, identifies the selected OpenHands SDK checkout as the fact task's project, and binds its complete pinned tree as the grading source set. Actual target execution and lifecycle gates remain open.

• Base commit: 73ca1e8b53383f5bd5e1a59a06eade50a0086370.
• Lane: lane:foundation.
• Owned paths: examples/native-sdk-acceptance/README.md; blueprints/convergence-practice/clean-resolution-20261002/open-work.json.

SOTA sources

• OpenHands/software-agent-sdk v1.50.1, commit 1e1390acc8788346ba4804c34323284009bf3f5e: hello-world workspace/task, example harness exclusions, mock cross-test's actual task, and metrics example's distinct task.
• The selected caller at native-agent-stack@b5079894cb08508b09214613dac5774e0fb017ef: install plan and acceptance launch. The upstream commit API binds root tree 64e12e9ee91c3ec48e9f7c516ca4a8c178539d4d; its complete versioned tree is the explicit local grading corpus.
• cli/cli v2.102.0, commit fc4b137cdef0a6bd28fd461b7cf9c84a5812a8cd: native CI command and scoped retained qualification.
• Original source-preparation state: final review binding at45ffc7d4e, publication as73ca1e8b5, and catalog source review carryover.

Evidence-class table

Claim	Evidence class	Command / receipt
Source preparation was accepted and published	source_review	Exact-head acceptance and native merge readback linked above
Selected caller assigns the OpenHands SDK checkout; the pinned tree grades facts and supporting tests have different scopes	source_review	Pinned caller/upstream files above; independent Astra source inspection
Recorded gh suite/build/local-smoke results	native_proven	Existing retained qualification and independent output/artifact review; no replay or fresh run in this documentation change
Documentation and frozen convergence record remain consistent	local_integration	Convergence/publication validators and pre-push registry tests

Local commands run

rtk python3 scripts/validate_convergence.py blueprints/convergence-practice/clean-resolution-20261002/oracle-experiment.json --root . --json
exit0; three valid observations
rtk python3 scripts/validate.py
exit0; 69 components, 9,315 hashed files, four profiles, 186 receipts
rtk git diff --check
exit0

The native pre-push registry tests also passed. No unchanged oracle suite, model trial or upstream suite was repeated for this documentation update. Fixture/oracle/control/output artifacts and task semantics are unchanged. Independent source review and the caller-binding correction are recorded in PR comments; actual target readiness and source binding, useful SDK execution, native result/usage and lifecycle/role acceptance remain open.

Decision record

Existing docs/decisions/2026-10-02-clean-resolution-goal.md and the scoped resolution envelope remain the governing records. This updates current status and explains pinned source behavior; it makes no new runtime selection or comparison claim.

Host evidence

No host receipt or platform status changes.

Checklist

• No workflow or action-pin changes.
• No secrets printed, logged or committed; no new required secret.
• No new paid hosting, subscription or billing surface.
• Peer-owned files and worktrees preserved.

[seathatflowsinourveins]

Independent read-only Sol/max review ACCEPTS this documentation update, with no overclaim or incorrect closure. The reviewer checked the original SDK acceptance/publication records, GitHub's actual merge and required contexts, the bounded GitHub CLI qualification, catalog review carryover, and all four pinned OpenHands source files. Target readiness, runtime execution, task grading, usage, resume/recovery, selected MCP and role gates remain open.

The review's only suggestion was to point the workspace citation at upstream line24; that is applied. Root bound the reviewed README bytes (prior SHA prefix46e7bad9a19f) to the committed README with only that requested anchor replacement; the requirement-map bytes remain unchanged (SHA-256 902389a0bf561505ced931426ccb7451a5cca963b5f39a6676212be4d407967f). Final README SHA-256 is 534192f3efd0133895a44de337a1a6b48a853895776a41d9e61edc89c09d350c. Both files bind clean head ce2dd050bde3ab8c990b90a9f62e2718544bc046.

The source review reused prior matching evidence and ran no test or provider/model trial. Coordinator checks returned: scoped convergence validator0 with three valid observations, repository validator0 with69components/9,315hashed files/four profiles/186receipts, diff check0, and all three native pre-push registry tests passed. The final anchor-only correction reran only required pre-push checks. Full hosted checks on this head remain the publication gate.

The OpenHands inspection also prevents an upstream-comment trap: the mock cross-test says it sends the same message, but its actual task creates hello.py; the selected hello-world example asks for project facts. The linked actual pinned source governs the scope. No arbitrary three-answer oracle, new task prompt, model runner or target selection was introduced.

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.
To continue using code reviews, add credits to your account and enable them for code reviews in your settings.

[seathatflowsinourveins]

Current head is 7db28359dfd4e2b071a3c61b77d51434fefdaeec, incorporating current main b5079894cb08508b09214613dac5774e0fb017ef. Both reviewed documentation files remain byte-identical to accepted ce2dd050bde3ab8c990b90a9f62e2718544bc046; the branch's registry exactly equals main's registry. Repository validation returned0 and all three required pre-push tests passed. The clean local and remote heads match.

The independent source review therefore carries over to this head. Current-head CI remains open; the previous commit's checks are not publication evidence for the new commit. SDK target execution and the full goal remain open.

[seathatflowsinourveins]

Correction to my previous OpenHands selector handoff and PR612 wording: I inspected the example's os.getcwd() behavior but missed its selected caller. The accepted install plan already selects OpenHands/software-agent-sdk v1.50.1 in $tool_root/openhands-source; its after-sign-in command changes to that directory. The generated installer and acceptance caller confirm the contract.

The claim that the planned project and launch-directory selection were missing was too broad and is withdrawn. The remaining requirements are verifying the actual target checkout/files against the selected source, freezing the grading source bindings/allow-list, retaining the facts artifact, and selected provider/model/permission readiness. The existing plan does not need an arbitrary replacement project.

PR612's corresponding paragraph is held for correction before publication. I requested Astra/max primary-source review because the accepted source wording omitted consequential caller evidence. The accepted any-three-supported-facts rubric, distinct upstream test scopes, original failures and target ownership remain intact. No target, provider, configuration or task-prompt change was made. This correction will be retained in the owned guide and durable handoff, with a proposed harness anti-pattern row for its existing owner.

[seathatflowsinourveins]

The OpenHands caller correction is complete at cf2101651bf353ce4c9764733fc00334a09855f3. This supersedes the earlier missing-project-selector claim and the temporary hold in this thread.

The selected install plan, installer and acceptance caller already bind the task to the OpenHands SDK checkout. The unchanged upstream example supplies the current-directory workspace and asks for three project facts. Actual target directory, source bytes, resolved packages, route/model/permission readiness and returned artifact grading still need acceptance.

Astra/max was invoked because the caller evidence contradicted an accepted but incomplete source review. It accepted the corrected binding and the complete pinned Git tree as the local grading corpus: commit 1e1390acc8788346ba4804c34323284009bf3f5e, root tree 64e12e9ee91c3ec48e9f7c516ca4a8c178539d4d. This preserves any three supported distinct facts; it does not impose arbitrary answers or establish sandbox permissions. Native upstream identity reads returned exit 0. The independent review bound the actual README SHA-256 8dc2fa946547c71b1ab312637020cd45784975b30bf194d1b15b0c774bef33ba, which the coordinator matched before committing. The requirement-map bytes remain unchanged from their previous accepted review, SHA-256 902389a0bf561505ced931426ccb7451a5cca963b5f39a6676212be4d407967f.

Scoped convergence validation returned exit 0 with three valid observations; repository validation returned exit 0; diff checks and three required pre-push tests passed. These are local integration checks. No fixture, provider or target run was repeated for this prose correction. Hosted required checks remain pending and will be freshly verified as the complete eight-context set before any guarded merge.

Correction recorded: reviewing the callee without its selected caller caused the erroneous missing-selector claim. Both public relay locations have been corrected; the private owner handoff also contains the verification path and proposed anti-pattern entry for the harness owner.

[seathatflowsinourveins]

PR612 now binds head f77e433c1946b5dd3f800412990c74f8ccd82b78, incorporating merged catalog metadata main 5803017249ad2a5e90ace4ab4e33e443ae00b66e. The merge was clean. Native git diff --exit-code returned 0 for both accepted SDK source files against cf2101651bf353ce4c9764733fc00334a09855f3; the evidence registry is byte-identical to current main. The Astra correction acceptance therefore still binds the actual source bytes.

Scoped convergence validation returned exit 0, repository validation returned exit 0 (69 components, 9,315 hashed files, four profiles, 186 receipts), diff checking returned 0, and the protected native push ran all three required pre-push tests successfully. No unchanged fixture, model or upstream source suite was repeated for this base integration.

New-head hosted CI is pending; the complete eight required contexts will be checked freshly before a guarded merge. The existing native watcher follows the current PR head. Target/provider/useful-task/role/lifecycle gates remain open under the existing ownership handoff.

[seathatflowsinourveins]

The complete eight required contexts passed for f77e433c1946b5dd3f800412990c74f8ccd82b78. Before merging, the fresh main read detected the concurrently merged configuration PR608 at 99d5b122c4d696f818b9be0a9514c154f9902adc, including a shared evidence-registry change. The retained prior-head check JSON is 1,370 bytes/SHA-256 2f4914a00149c7ca1f17aba8f2dce156aa103ae0ba501b0df1c532e3c86bf246; it is historical passing evidence, not acceptance of the new head.

Following the hot-file rule, PR612 now integrates that main as 74ce20f02ae04b52fe8ac3f94294ca897920671c. The merge was clean. Both accepted SDK source files are byte-identical to the reviewed prior head; manifests/evidence.json is byte-identical to current main. Native equality checks returned 0. Scoped convergence and repository validation returned 0; native push returned 0 after all three required pre-push tests passed. No SDK fixture, provider or upstream source suite was replayed for this integration.

New-head hosted checks remain pending. The old watcher completed with native exit 0; a new watcher is needed for the pushed head. This PR remains source-status/caller documentation only; target/provider/useful-task/role/lifecycle gates remain separate. The next unrelated non-hot-file main advance alone does not mandate another CI replay, but a new shared-hot-file change still does under the written policy.

[seathatflowsinourveins]

PR612 now integrates current main 0eb47187925bf8134660b7a305aa32b0d44b9e17 at head c3cba27bc3b720afdc196f6698b5cdbd14cfdd0b. The preceding 74ce20f head completed all eight required checks, but PR613 concurrently changed the shared registry; the repository's hot-file integration rule therefore required this new base before merge.

Both reviewed SDK source files remain byte-identical to the accepted correction: README SHA256 8dc2fa946547c71b1ab312637020cd45784975b30bf194d1b15b0c774bef33ba, open-work SHA256 902389a0bf561505ced931426ccb7451a5cca963b5f39a6676212be4d407967f. This branch's registry equals current main exactly. Scoped convergence validation, repository validation, diff check and the three native pre-push checks passed. Fresh required checks are running on the new head; merge still requires the complete eight-context set and a matching reviewed SHA.

Actual SDK execution is now a separate ongoing lane under the builder's existing bounded handoff. It does not change this PR's source-preparation scope or retrospectively turn fixture controls into provider acceptance. No unchanged fixture or upstream source suite was replayed for this base integration.