Port #566 historical native runtime evidence - #668

Scope

Preserve #566's historical OpenHands/DeepAgents evidence and supported example sources on the current integration base. The role packet moves unchanged into evidence; the receipt pointer and its convergence-record hash are rebound. The stale guidance document is replaced by a dated port record: historical; sets no default; runtime-job selection is owned by the runtime-jobs lane (#633).

• ported from Qualify native runtime roles beyond OpenHands through OmniRoute #566 @e9bd476
• Source head: e9bd476cdab11751ee7be00ac2c954ce65ff3c60; source branch left unrewritten.
• Base commit: 1f5a791b02a230aced670c88bab3d3d0ebcf401a (origin/main at rebase, 2026-10-03T19:42Z). Head: a5b1cc434089c5304b929160a35712c7f50cd425 (review-repair round; the independently reviewed head was 29ecc3c5fc9eb842d07eeffb97c8f54df083c537). The sandboxed content build ran on cac8700ba914950266272347468bff7ad630a4bf; its content commit was rebased onto current main unchanged, and the registry was rebuilt there from main's manifests/evidence.json under the hot-file protocol.
• Commits: 5d5175d66 Port PR Qualify native runtime roles beyond OpenHands through OmniRoute #566 historical native runtime evidence (33 content paths); 29ecc3c5f Register PR Qualify native runtime roles beyond OpenHands through OmniRoute #566 port evidence on current main (manifests/evidence.json only); 9092bbc8b Port Qualify native runtime roles beyond OpenHands through OmniRoute #566: correct token and base wording in the port record (review minors) (port record only); a5b1cc434 Register the corrected port record (manifests/evidence.json only, last).
• Lane: lane:foundation; runtime-jobs lane acknowledgement on this PR is required before merge.
• Owned paths touched: evidence/artifacts/native-runtime-role-resolution-20260930/, evidence/receipts/native-runtime-role-resolution-20260930.json, examples/deepagents-omniroute/, examples/openhands-native-capabilities/, .github/osv-scanner-lockfiles.json, and manifests/evidence.json.
• The 30 carried paths receive only the two prescribed content edits. The relocated packet and dated port record add two paths. Five OSV inventory entries are added. Evidence registration is the last commit.

SOTA sources

The port preserves these frozen upstream revisions and their supported interfaces. Source pins and execution evidence remain historical; this publication makes no new framework selection.

SOTA sources

• OpenHands CLI 1.16.0 @2963442dacc7cea44e39b7c4e73724295c853465: OpenHands/OpenHands-CLI, native CLI and selected original tests named in openhands-install.json.
• OpenHands SDK/tools 1.50.1 @1e1390acc8788346ba4804c34323284009bf3f5e: OpenHands/software-agent-sdk, LLM.generate, native configuration and original SDK tests.
• LiteLLM 1.93.2 @cd1bd0f4b8af865f8d05fbd938392fdd4703babc: BerriAI/litellm, provider-prefix resolver.
• DeepAgents 0.7.21 @4394bcd00b8eb46e7c423939643a0dfcfb5d8773: langchain-ai/deepagents, create_deep_agent, FilesystemBackend and the original selected test suite.
• LangGraph checkpoint-sqlite 3.1.1 @b2926a0ff9589c28c7e01fe7cdbb337b86d5a4b4: langchain-ai/langgraph, matched released checkpointer source.
• LangChain Core 1.6.6 @04ac76c07ec173a44e3e57de54861d0b637e3299: langchain-ai/langchain, native secret-aware dumps used in the preserved CodeQL fix.
• OpenAI Agents SDK 0.22.3 @fdf21db62c303a3db54b0dfbee82de2141fa2799: openai/openai-agents-python, source-reviewed SandboxAgent alternate in the unchanged role packet.

Port procedure: docs/lanes.md, Hot-file protocol; scripts/host_receipts.py:register_file; docs/acceptance-evidence-policy.md. The native generators ran in their prescribed order on the rebased tree and produced unchanged reports.

Evidence-class table

Local commands run

Committed-head acceptance on head 29ecc3c5f (base 1f5a791b0), run 2026-10-03T19:44Z in the publishing worktree. Every check used nice -n 19 with TMPDIR=/var/tmp/566-publish PYTHONDONTWRITEBYTECODE=1. Exit codes are verbatim. Acceptance at the review-repair head a5b1cc434 is recorded in this PR's review-record comment.

$ python3 scripts/validate.py
exit 0
{"components": 69, "hashed_files": 9573, "profiles": 4, "receipts": 194, "status": "passed"}
Integrity and scope checks only; no live provider or GPU execution.

$ python3 scripts/validate_convergence.py evidence/artifacts/native-runtime-role-resolution-20260930/experiment.json
exit 0
record[0]: valid

$ python3 -m pytest tests/test_osv_lockfile_coverage.py
exit 1
No module named pytest; no tests executed by this invocation (pytest is not installed on this host; nothing was installed).

$ python3 tests/test_osv_lockfile_coverage.py
exit 0
Ran 54 tests in 2.358s
OK
(The unchanged module's own unittest entry point, run in place of the missing pytest. It checks inventory policy; it is not an advisory scan.)

$ git diff --check origin/main...HEAD
exit 0
(no diagnostics)

$ git diff --quiet e9bd476cdab11751ee7be00ac2c954ce65ff3c60 HEAD -- evidence/artifacts/native-runtime-role-resolution-20260930 examples/deepagents-omniroute examples/openhands-native-capabilities ':!evidence/artifacts/native-runtime-role-resolution-20260930/experiment.json' ':!evidence/artifacts/native-runtime-role-resolution-20260930/role-packet-20260930.json' ':!evidence/artifacts/native-runtime-role-resolution-20260930/port-record-20261003.md'
exit 0

$ sha256sum evidence/artifacts/native-runtime-role-resolution-20260930/role-packet-20260930.json
exit 0
d8c9922b0e85feba8002dd8df51db0d38eae895b9072bb8d8007ae8df0176a63

$ git diff --numstat e9bd476cdab11751ee7be00ac2c954ce65ff3c60 HEAD -- evidence/receipts/native-runtime-role-resolution-20260930.json evidence/artifacts/native-runtime-role-resolution-20260930/experiment.json
exit 0
1 1 evidence/artifacts/native-runtime-role-resolution-20260930/experiment.json
1 1 evidence/receipts/native-runtime-role-resolution-20260930.json

$ python3 scripts/evidence_manifest.py --check
exit 0
{"files": 9573, "status": "passed"}

$ merge-tree landing check (disjoint-drift rule, #608) on main 1f5a791b0 and head 29ecc3c5f
exit 0
main 1f5a791b02a230aced670c88bab3d3d0ebcf401a head 29ecc3c5fc9eb842d07eeffb97c8f54df083c537 base 1f5a791b02a230aced670c88bab3d3d0ebcf401a merged-tree 9168eacbc9534978f047bd6e5e78c50d4a5e6ee7 merge-tree-exit 0
ok   1: clean three-way merge
ok   2: merged-vs-main paths 34, outside PR-owned 0 []
ok   3: main drift 0 paths, overlap with PR-owned inputs (registry excluded) []
ok   4: registry foreign rows equal True, order preserved True, PR-owned rows 33, unowned top-level keys differing []
ok   5: merged files[] sorted by path with no duplicates: True (9573 rows)
LANDABLE

$ git commit, both commits (pre-commit hook: gitleaks 8.30.1 git --pre-commit --staged --redact)
exit 0
no leaks found (content commit ~703 KB scanned; registry commit ~9 KB scanned)

$ git push -u origin foundation/pr566-runtime-evidence-port-20261003 (pre-push hook, no bypass)
exit 0
pre-push: running the registry tests on 29ecc3c5fc9eb842d07eeffb97c8f54df083c537
test_every_tracked_lockfile_and_manifest_is_listed ... ok
test_every_blueprint_value_under_a_label_key_is_classified ... ok
test_all_published_workflows_are_listed_and_covered ... ok
Ran 3 tests, OK

Registry rebuild on the rebased tree (hot-file protocol): main's manifests/evidence.json was taken unchanged, the exact source receipts[] row and the experiment's convergence_records[] path were appended once, and the contract's host_receipts.register_file command registered the 32 port files and refreshed the existing OSV inventory row. python3 scripts/component_matrix.py --write (exit 0; {"flip_rule_violations": 0, "rows": 32, "status": "written"}) and then python3 scripts/new_host_grand_list.py --write (exit 0; {"status": "written", "layers": 32, "winners": 66}) left all four generated reports unchanged. A structural comparison against main 1f5a791b0 returned exit 0: 9,573 rows, 32 added (exactly the port files), none removed, only the OSV inventory row changed, and all 9,540 foreign rows byte-equal to main's and in main's order; receipts[] and convergence_records[] equal main's plus the one source row and path. The sandboxed build's actual-file comparison, rerun on the committed tree with its base pinned to 1f5a791b0, also returned exit 0: all 28 unchanged carried files are byte-identical to the source, both edits equal the exact permitted substitutions, the relocated packet keeps its 18,535 bytes and SHA-256, the worker keeps its CodeQL-fix hash, and only the five prescribed OSV inventory rows are added.

Retained failed attempt from the sandboxed build: its first registry comparison returned exit 1 because another session advanced the shared origin/main ref during the run, and its corrected comparison pinned the build base. Both are superseded by the current-main rebuild above, which must be repeated if main moves before merge.

Pending publication gates: CI's OSV, dependency-review, secret-scan and CodeQL results for this head; eight required checks with the sole bucket pass; the runtime-jobs lane's ack; and an independent Opus/max exact-head review. The exact pytest command remains unfulfilled locally (exit 1, pytest not installed). No local osv-scanner executable is on PATH, so no local advisory scan ran.

Decision record

evidence/artifacts/native-runtime-role-resolution-20260930/port-record-20261003.md records provenance, rebindings, the historical role table, all seven pins, evidence classes, held conditions and custody alternatives. It marks Codex SDK 0.159.2 and #551's old draft reference as stale. The earlier custody snapshot described #628 as open; the read-only refresh verifies its merge on 2026-10-03. Runtime-job choices remain owned by #633.

Stop and return to the custody coordinator if a carried lockfile or secret scan produces a finding, substantive receipt fields require edits, or the runtime-jobs lane requests absorption, retirement or relocation. Pins and retained failed attempts are preserved.

Host evidence

Not applicable: this port changes no evidence/hosts/ files and claims no new host acceptance.

• python3 scripts/host_receipts.py validate passes for every new/changed host receipt (not applicable).
• Independent host-receipt review is recorded or requested (not applicable).
• No platform status change is made from a host receipt alone.

Checklist

• New/changed GitHub Actions are pinned to a full commit SHA with a version comment (not applicable; no workflow changes).
• New/changed workflows declare scoped permissions (not applicable; no workflow changes).
• No secrets are printed, logged or committed; no new required secret is added.
• No new paid hosting, subscription or billing surface is introduced.
• Peer-owned untracked files and worktrees are preserved.
• Current-main rebase, registration rebuild and committed-head acceptance commands complete. Not ticked: the exact python3 -m pytest tests/test_osv_lockfile_coverage.py exits 1 because pytest is not installed on this host (nothing was installed), so pytest has not passed. Disclosed substitute: python3 tests/test_osv_lockfile_coverage.py (the module's own unittest entry point) ran 54 tests OK, and CI runs the module's LockfileInventoryTests and FrozenScanTests classes under python3 -m unittest (.github/workflows/security-scan.yml L75-77). Accepting the substitute in place of pytest is a coordinator decision.
• Eight required checks report 8 required checks, buckets: pass.
• No new CodeQL alert.
• Runtime-jobs lane acknowledgement exists on this PR.
• Independent Opus/max exact-head review is clear.

🤖 Generated with Claude Code