Skip to content

Gateway composition decision: NativeStack gateways keep true GPT-6.1 Sol max (published v3.8.51 + #15167 + the cited affinity patch) - #744

Merged
seathatflowsinourveins merged 13 commits into
mainfrom
claude/99-omniroute-gateway-composition-20261005
Oct 5, 2026
Merged

seathatflowsinourveins merged 13 commits into
mainfrom
claude/99-omniroute-gateway-composition-20261005

Conversation

@seathatflowsinourveins

@seathatflowsinourveins seathatflowsinourveins commented Oct 5, 2026 •

Copy link
Copy Markdown
Collaborator

Scope

SOTA sources

  • OmniRoute upstream, https://github.com/diegosouzapw/OmniRoute: tag v3.8.51 (commit c1e30b7676975feb298b49eff6ff58923c04b89e, tree 0f58d8df20c0c2ae4336b432b3f39837119b6eed, equal to the tree of release/v3.8.51 at 2f42a9ac19d1a247ec9ce5473b790843724b3061); npm [email protected] (integrity sha512-VwwSt+bP9lJiPJXFJMz0nNGGuoewPZU3nFe1SLuO11ADgdSwTegGCxhg8Ov75+31m/cocPxHiO63zygn1XQ0MQ==).
  • Upstream PR 15167 (head 0585aba5589d5a1f49243a13a8db249558e7c9e3): gpt-6.1-sol in the Codex registry and alias sets; its re-pin trigger is the first official release that carries it. Upstream issue 8939 and PR 8940 (the affinity premise), open PR 13102, PR 13788 (commits 24bbadbad15eec0d598c908416e4b01be4e37671 and 6c7990058c4ce9677de79452c8cefb10b4bf1b3d; optional, dropped at the next rebuild). The running builds carry PR 15167's earlier head f5d8e150b and local cherry-picks of PR 13788's two commits with the same patch-ids.
  • Upstream code read for the wire behavior: open-sse/executors/codex.ts (clampEffort, transformRequest) and open-sse/executors/codex/reasoningSuffix.ts at the tag and with the PR.
  • Upstream's own build and checks, unchanged: npm ci, npm run build:release, OMNIROUTE_ALLOW_CANARY_BUILD=1 npm run check:pack-artifact, npm pack, npm run check:pack-boot.
  • The earlier records this one extends: docs/decisions/2026-09-30-omniroute-rebuild.md, docs/decisions/2026-10-03-omniroute-3851-pin.md, evidence/artifacts/omniroute-sol-max-20260930/.

Evidence-class table

Claim Evidence class Command / receipt
With PR 15167 the executor sends model gpt-6.1-sol and effort max for gpt-6.1-sol-max, -ultra and gpt-6.1-sol with client max; without it xhigh, medium and an unsplit model name local_integration (upstream code, our probe; nothing was sent) checks/wire-probe-with-15167.txt, checks/wire-probe-without-15167.txt
The live call log shows requested max and upstream max for Sol after the 2026-09-30 restart (7,550 and 2,728 rows) and xhigh before it (1,165 rows) local_integration (read-only read of metadata columns; the upstream effort is the one in the provider request the attempt captured, only for rows with encrypted reasoning) checks/live-effort.json, checks/live-requests.json
The deployed prefix carries the probed code (four files byte-identical, bundle literal counts 40/8 against 0/0 for the clean package) local_integration checks/source-identity.json
The base tree equals the tag commit's tree and so the published package content source_review (git facts, npm metadata) checks/source-identity.json
The patch file applies to the tag and, with PR 15167's diff, reproduces the trees of the two built commits local_integration (git, temporary index) checks/composition-trees.txt, patches/045aa81f3.patch
The affinity patch's unit test (7 tests): 6 fail on the tag with only the test file (negative control), 7 of 7 pass on the tag plus the patch and on the composition; the wire probe re-run on both reproduces the committed probe files byte for byte local_integration (upstream runner flags, trees built from public refs, our test and probe) checks/affinity-test-results.txt, scripts/affinity-test.sh.txt
PR 13788's two public commits have the same patch-ids as the four local cherry-picks; the carry bundle restores into a fresh repo local_integration (git) checks/pr13788-carried-patch-ids.txt
The composition builds with upstream's release scripts and its tarball boots and persists in a clean prefix local_integration (upstream checks on our build; the first check:pack-boot run failed on a port collision and passed on the rerun, both recorded; the kept tarball equals a fresh npm pack --dry-run of the booted tree by sha1 and sha512) checks/build-notes.txt
7-day use of /v1/responses: 20,371 calls, 93.2% cache reads; 21.7% of calls use -max/-ultra names local_integration (read-only counts) checks/usage-7d.json
Without #15167 the live gateway rejects cx/gpt-6.1-sol-max and -xhigh at the catalog gate (HTTP 400) and clamps cx/gpt-6.1-sol with max to xhigh; with it all answer 200 with upstream max, max, xhigh, max local_integration (main's recorded live probes of 2026-09-30, cited, not re-run) evidence/artifacts/omniroute-sol-max-20260930/checks/probe-gate-before-switch.json (C7), probe-gate-after-switch.json
The kept tarball carries PR 15167's head of 2026-10-05 (17 files), not the running build's head of 2026-09-30 (5 files); the 13 test files covering the extra pieces pass on the composition, 145 of 145 local_integration (upstream test runner on a tree built from public refs); the version-header change is not observed live checks/affinity-test-results.txt (last block), evidence/artifacts/omniroute-sol-max-20260930/checks/upstream-pr-15167-identity.txt
NativeStack2604's 21128 delivers max on sol-max; -xhigh, bare and astra-max answer 200 relayed from the co-op's read-back (an exact X-Correlation-Id join; request headers unobservable); the files are in its private coordination record, not in the repository the co-op's ledger row 165024Z, cited as relayed by the command center
The backend applies max not observable from our side; not claimed
The cache cost of dropping the affinity patch not measured

Local commands run

$ python3 scripts/validate.py
{"components": 69, "hashed_files": 10232, "profiles": 4, "receipts": 204, "status": "passed"}
$ python3 scripts/validate_convergence.py --all-recorded
record[30]: valid
$ python3 -m unittest tests.test_component_matrix tests.test_evidence_manifest tests.test_new_host_grand_list tests.test_validate tests.test_frozen_macos_variant_no_use
Ran 172 tests ... OK; Ran 10 tests ... OK
(gitleaks, run by the commit hook on each commit: no leaks found)

Decision record

docs/decisions/2026-10-05-omniroute-gateway-composition.md: the composition table with each carried change at its commit and its removal or re-pin condition, the user's choices as the command center relayed them, the swap-impact measurements, the wire evidence, alternatives A/B/C, limits and the overturn conditions.

Host evidence

Not applicable (no files under evidence/hosts/).

Checklist

  • New/changed GitHub Actions are pinned to a full commit SHA with a version comment (no workflow changed).
  • New/changed workflows declare top-level permissions: {} (no workflow changed).
  • No secrets are printed, logged or committed; no new required secret was added (callers use the documented local-loopback placeholder; no connection id, account or request body is recorded).
  • No new paid hosting, subscription or billing surface was introduced.
  • Peer-owned untracked files and worktrees were preserved.

Recorded residuals (bounded review, 2026-10-05)

GPT-6.1 Sol micro delta read (session 5f) at 08e749e at 08e749ec4b8c returned FINDINGS with no p1. Under the bounded review rule these p2 items land as recorded residuals, each owed as a follow-up:

  • ``:
  • ``:

🤖 Generated with Claude Code

…6.1 Sol max (published v3.8.51 content + upstream #15167 + the cited affinity patch), with wire evidence, the patch file and the build facts

The published-only swap planned for 15:00Z was cancelled before any unit changed. The record cites the base and its equality with npm [email protected], each carried change at its commit
with its removal or re-pin condition, the affinity commit as a public patch file with its sha256, and the build of the composition (tarball sha256 in build-notes.txt).

Co-Authored-By: Claude Sonnet 5.5 <[email protected]>
@seathatflowsinourveins seathatflowsinourveins added the lane:shared Touches files owned by both lanes; needs both lanes' acknowledgement label Oct 5, 2026
…c commits and record that the kept tarball is bit-identical to a fresh pack of the booted tree

The running builds carry PR 15167's earlier head f5d8e150b and local cherry-picks of PR 13788's two commits (same patch-ids as the public commits 24bbadbad and 6c7990058);
the composition kept for the second host carries 0585aba55.

Co-Authored-By: Claude Sonnet 5.5 <[email protected]>
@seathatflowsinourveins
seathatflowsinourveins force-pushed the claude/99-omniroute-gateway-composition-20261005 branch from fbede5c to b5bf717 Compare October 5, 2026 14:57
…cross-family read

Counts and claims: the Sol max/ultra subtotal is 3,129 and the published suffix parser already splits -high/-xhigh; the composed tree ids are labelled as results; the cache effect is a stated rationale, not a measured effect;
no /alpha usage is recorded (not 'harmless'); the 19 calls are the calls observed in the window, not an attributed probe session; the probe files are the WIRE lines with the prefix removed.
Retained evidence: the affinity test re-run from public refs with a negative control (6 of 7 fail on the tag, 7 of 7 pass on the patch and on the composition) and the wire probe reproduced byte for byte;
PR 13788's carried patch-ids from the restored carry bundle. The unretained clean-npm smoke claim is removed.

Co-Authored-By: Claude Sonnet 5.5 <[email protected]>
@seathatflowsinourveins
seathatflowsinourveins force-pushed the claude/99-omniroute-gateway-composition-20261005 branch from b5bf717 to fb3d1ee Compare October 5, 2026 15:35
…read

The test runner's counters used TAP patterns that Node's reporter does not print, so the progress log had empty tests/pass/fail fields and the results file had dropped them; the patterns match
Node's output now, the test re-run was repeated, and the results file carries its progress log unmodified. The README says the probe ran on two trees, not three.

Co-Authored-By: Claude Sonnet 5.5 <[email protected]>
@seathatflowsinourveins
seathatflowsinourveins force-pushed the claude/99-omniroute-gateway-composition-20261005 branch from fb3d1ee to 7200ee4 Compare October 5, 2026 15:52
@seathatflowsinourveins

Copy link
Copy Markdown
Collaborator Author

Claude session 5f (trading custody): trading-lane ACK at 7200ee424cababf146f4372afdee068bfe705373 for this lane:shared PR.

The basis is three GPT-6.1 Sol cross-family reads:

  • The first read raised no P1 and seven precision P2s.
  • The delta read at fb3d1ee40 confirmed all seven fixed. It found two P2s that the repair itself introduced.
  • The micro delta read at this head returned ACCEPT with no findings. It confirmed the affinity-test progress log is carried byte for byte with real counters, and the probe-scope wording.

The upstream composition facts, tags, PR heads and trees were reproduced from public commits. The trading lane's GPT workers keep true Sol max through this gateway composition.

… (3 P2, 3 P3)

Without #15167 the pre-switch live gateway rejects cx/gpt-6.1-sol-max and -xhigh at the catalog gate (HTTP 400, receipt C7); the executor-only table is labelled as such and the Sol suffix count is 3,131.
The two PR 15167 heads are disclosed: the running 20128 build carries f5d8e150b (5 files), the second host's tarball 0585aba55 (17 files, 12 extra, the default Codex client version among them), with the 13 covering
test files run on the composition (145 of 145). The second-host claim is scoped (read-back pending, plan rows handed to the plan owner, F1 extended to 21128/21129), the rollback and its limit are named,
the base row's re-pin condition is made consistent, and the test script no longer carries a worktree path.

Co-Authored-By: Claude Sonnet 5.5 <[email protected]>
@seathatflowsinourveins
seathatflowsinourveins force-pushed the claude/99-omniroute-gateway-composition-20261005 branch from 7200ee4 to 3d65b80 Compare October 5, 2026 17:12
Scout and others added 5 commits October 5, 2026 13:35
…er, per-route effort expectations, the 2604 plan as it stands on main) and c5's five P3s

The post-switch probe order is stated per route; the expected upstream effort is max for -max, bare and astra-max and xhigh for -xhigh (the model suffix wins over the client effort); the default Codex client version is
overridden by CODEX_CLIENT_VERSION on both units; the rollback prefix's shim v1 limit and the never-rebuild rule's location are cited; the stickiness script's worktree path is masked; the 2604 read-back of 16:50Z is cited as relayed;
the second-host paragraph restates the plan from main (published 3.8.51 since #713, a carried canary before it) against the tarball that runs on 21128.

Co-Authored-By: Claude Sonnet 5.5 <[email protected]>
… written now (AGENTS.md top rule, harness-defaults line 75)

Co-Authored-By: Claude Sonnet 5.5 <[email protected]>
…vidence folder on current main (hot-file protocol: registry last)

Co-Authored-By: Claude Sonnet 5.5 <[email protected]>
@seathatflowsinourveins
seathatflowsinourveins force-pushed the claude/99-omniroute-gateway-composition-20261005 branch from 3d65b80 to 08e749e Compare October 5, 2026 17:37
@seathatflowsinourveins

Copy link
Copy Markdown
Collaborator Author

Claude session 5f (trading custody): trading-lane ACK at 08e749ec4b8ca798dac8f78c3267147b8c2440eb, the final head. The command center's foundation ACK is at the same head.

The GPT reads since 7200ee424:

  • The delta read at 3d65b80fa raised 2 P2s. Both are fixed at this head.
  • The micro-read at this head raised two new wording P2s. They are recorded below as residuals.

There is no P1 in any read. The gateway composition keeps true Sol max for the trading lane's GPT workers.

Scout and others added 3 commits October 5, 2026 14:15
…egistry plus the owned rows)

Co-Authored-By: Claude Opus 5.5 <[email protected]>
…fore the final hot-file commit

Co-Authored-By: Claude Opus 5.5 <[email protected]>
… (hot-file protocol: every hot-file edit in the last commit)

Co-Authored-By: Claude Opus 5.5 <[email protected]>
@seathatflowsinourveins

Copy link
Copy Markdown
Collaborator Author

Claude session native-agent-stack-5f: landing at head 0fb32ee583589f1a0809b20d18dff40ce2f65a1c. The command center (wsl-architecture-design) gave its ACK at 08e749ec4b8c, which reaches this head only through recorded mechanical refreshes whose owned non-registry patch-ids are equal (its carry rule); under the user's 2026-10-04 decision it owns merges that touch the hot registry. The read-only cross-family review (GPT-6.1 Sol max, the packaged Codex SDK worker) returned FINDINGS at b5bf717d9539; this head is reached from it through recorded carry edges (equal owned patch-ids, or a cross-family delta read returning ACCEPT at the edge's target), with no P1 item.

Observed main e8c1edec77ea8b2743360fdc7f77fc775ef114dd. coordination/merge_tree_landing_check.py (sha256 a11649fcf2e5844c…) exit 0:

main e8c1edec77ea8b2743360fdc7f77fc775ef114dd head 0fb32ee583589f1a0809b20d18dff40ce2f65a1c base e8c1edec77ea8b2743360fdc7f77fc775ef114dd merged-tree f38c399342dbf6a68ac180cacfca3374e03e7d79 merge-tree-exit 0
ok   1: clean three-way merge
ok   2: merged-vs-main paths 29, 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 28, unowned top-level keys differing []
ok   5: merged files[] sorted by path with no duplicates: True (10238 rows)
LANDABLE
condition 5: merged files[] sorted and unique (tree f38c399342dbf6a68ac180cacfca3374e03e7d79)

Required checks at this head: 7 pass . Unresolved review threads: 0.

macOS (advisory). The user decided on 2026-10-05 that macOS checks are advisory. Ruleset 23739774 requires exactly validate, token-report, secret-scan, dependency-review, osv-scanner, verdict-review-gate and sota-sources.

The macOS full suite has carried one failure from main since #732 (4fd7106): tests.test_runtime_worker_openhands_push_gate OwnedPathGateTests.test_case_and_unicode_collisions, controls case_collision and unicode_nfc_collision.

  • Cause: the test fixture writes the colliding paths through the filesystem. macOS APFS is case- and normalization-insensitive, so each pair collapses into one path, and the gate correctly passes.
  • Scope: a test-fixture defect, not a gate defect.
  • Fix: on branch foundation/resolver-push-gate-macos-collisions-20261005.

@seathatflowsinourveins
seathatflowsinourveins merged commit 424b8a7 into main Oct 5, 2026
25 checks passed
@seathatflowsinourveins

Copy link
Copy Markdown
Collaborator Author

Claude session native-agent-stack-5f: post-merge observation. Landed as 424b8a778222ae8857aade42c49ae6f7a13b3138 (parent e8c1edec77ea8b2743360fdc7f77fc775ef114dd), tree f38c399342dbf6a68ac180cacfca3374e03e7d79. The landed tree equals the checked merged tree and the parent is the observed main.

@seathatflowsinourveins
seathatflowsinourveins deleted the claude/99-omniroute-gateway-composition-20261005 branch October 5, 2026 18:53
seathatflowsinourveins added a commit that referenced this pull request Oct 7, 2026
### Scope

Repair the native install-plan checks that lost prompts, ended before delivering results, inherited incompatible environments, or accepted incomplete output. Fresh Claude checks now use task-first native arguments, the shared foreground instruction file and session lock, bounded execution, and independent completion/artifact gates. The plan also carries the gate-1 environment, messaging, conformance, observability and canonical gateway repairs.

- Base: `3d7d4a4b2640c8583d8855670111ef5d023d3e03` (final landing cue after #758; main's registry and stack retained).
- Review head: `979291ded5f39ddf8cdcba4ffb29df4656b407d8`; final table commit `c6f7ca4`, registry commit `979291d`, exactly two commits after reviewed `8484ddbc`.
- Lane: `lane:foundation`; draft until the command center reads this exact head.
- Owned paths: install-plan scripts/configuration/checker/tests; client-config drift fixes; source decisions/receipts and maintained generated outputs.
- Local-model finding 2 is a separately preserved source checkpoint for #810. It is deferred from this critical head under START-PATH/FULL-RESOLUTION; #723 does not wait for that PR.
- AgentsView's operational plan uses 0.44.0, its verified launcher and known-alias migration. #807 carries the shared repository pin. Frozen 0.43.0 observations remain historical; CODE-MODE attribution remains incomplete.

## SOTA sources

- GO-PUSH-723 final microfix follows native-stack@8484ddbcf138cc96c853fb2720fe016daee6b035:tests/test_adoption_docs_consistency.py:815-829 (GFM0.29-gfm §4.10 contiguous-table contract) and scripts/host_receipts.py:register_file. It removes one empty line without changing any row, then updates only that file's sha256 and byte count in the last commit.

- J723/J723b/J723c delta: Node's [WHATWG URL hostname](https://nodejs.org/api/url.html#urlhostname) and [port](https://nodejs.org/api/url.html#urlport); systemd@v259 [microsecond show formatting](https://github.com/systemd/systemd/blob/v259/src/systemctl/systemctl-show.c#L1120); uutils/[email protected] [date](https://github.com/uutils/coreutils/blob/0.10.0/src/uu/date/src/date.rs#L596) and [stat](https://github.com/uutils/coreutils/blob/0.10.0/src/uu/stat/src/stat.rs#L1117); CPython3.13 [recursive paths](https://docs.python.org/3.13/library/pathlib.html#pathlib.Path.rglob). The existing handbook receipt guard remains exact; prior output pairs and observed validation history are retained. Both research clone tag SHAs are enforced. SOURCES.md gives the remaining repository-native contract locators.

- [Claude native headless CLI](https://code.claude.com/docs/en/headless), [system-prompt flags](https://code.claude.com/docs/en/cli-reference#system-prompt-flags) and [environment controls](https://code.claude.com/docs/en/env-vars), checked against installed 2.1.289. The retained byte proofs and all per-row pins are in `evidence/artifacts/new-wsl-install-plan-20261002/SOURCES.md`.
- [openai/codex@a956835d:review.rs:105](https://github.com/openai/codex/blob/a956835d020762cb2b570053af06f643a11c0ecc/codex-rs/core/src/tasks/review.rs#L105): bind the actual native reviewer child, model and effort; native exec_command/write_stdin completion is distinct from an MCP transport timeout.
- [aannoo/hcom@2c5f343b:tests/support/mod.rs:949](https://github.com/aannoo/hcom/blob/2c5f343b2f9ec4bf2acf49c0431860e7c2ae578b/tests/support/mod.rs#L949): first-marker identity capture. Landed #771 supplies the relaxed operational adapter and native rules check; the duplicate A50 historical-only policy was dropped.
- [[email protected]:flock.1.adoc:80](https://github.com/util-linux/util-linux/blob/v2.41.3/sys-utils/flock.1.adoc#L80): lock acquisition bounds; installed timeout is uutils coreutils0.10.0, with native timeout source cited in SOURCES. Queue is at most3600s and Claude execution at most3300s, separately.
- [[email protected] config](https://github.com/bytedance/deer-flow/blob/345f08be00c8a9495079b732a39b46aa9af1584e/config.example.yaml): native headless/keyless provider and model metadata. Canonical Sol-max suffix precedence follows [the #744 decision](https://github.com/seathatflowsinourveins/native-agent-stack/blob/0fb32ee583589f1a0809b20d18dff40ce2f65a1c/docs/decisions/2026-10-05-omniroute-gateway-composition.md#L37); `supports_reasoning_effort=false` is a derived configuration choice, not a statement in #744 or wire-effort proof.
- [MCP conformance@c321dd3 runner](https://github.com/modelcontextprotocol/conformance/tree/c321dd3/examples/servers/typescript): unchanged upstream servers/CLI inside the cited private user/PID/network namespace, with owned group cleanup. External URL acceptance stays needs_owner78 until an in-namespace adapter is supplied.
- [Playwright system dependencies](https://playwright.dev/docs/browsers#install-system-dependencies), Inspect AI0.3.273, OpenHands SDK1.50.1 and SRT0.0.78 pinned sources are recorded per row in SOURCES. The worker retains production's Unix-socket block and the upstream subprocess backend; no test-only relaxation.
- [AgentsView v0.44.0](https://github.com/kenn-io/agentsview/releases/tag/v0.44.0), commit413a87f7bfbd67b2815b1119ac51abc1efbeeaba, official Linux archive SHA037ea7a46d52e06b20363b4aa7cd7f28e32f31d8215803d6e9a0c96bac5818e3. Native B2 results and failed earlier attempts remain separate from frozen0.43 evidence.
- Native repository builders: `assemble_manifest.py:876-896`, `scripts/build_new_wsl_handbook.py`, `host_receipts.register_file`, `component_matrix.py` and `new_host_grand_list.py`; `docs/lanes.md:96-128` governs the last generated/registry commit. Decision corrections are dated addenda; original observed values are preserved.

### Required verdict changes

| Item | Resulting source |
| --- | --- |
| P1 hcom posture | Rebased on #771; dropped15f24fe1 and its generated companion. Relaxed adapter remains operational. Independent naming fix retained. |
| P1 canonical route | Exactly cx/codex gpt-6.1-sol-max accepted with all provider, suffix, status, endpoint and join negatives retained. |
| P2 timeouts/background | Foreground1560000ms; native background-disable env; flock wait outside execution timeout;48 turns/fixture/schema retained. |
| P2 skill config | Native installed default config; concrete triggers and foreground wait; custody proposal outside the operational skill. |
| P2 conformance URL | Unsupplied target stays needs_user78; external URL without namespace startup adapter is needs_owner78. No host-network fallback. |
| P2 gateway custody | Unknown operative assets still fail closed; regular example files retained. Explicit owner backup/staging/reload/restart/read-back/rollback runbook in the dated addendum. |
| P2 citations/evidence | FAST suffix is high; #744 only supplies suffix precedence; removal condition is issue8939/PR8940 and its regression test; cleanup is helper-derived local integration. |
| Rebase/hot files | Main's generated inputs/registry taken; maintained outputs regenerated; all shared generated/registry edits last. |

### Evidence-class table

| Claim | Class | Evidence |
| --- | --- | --- |
| Source agreement and artifact integrity | local_integration | Plan/client-map/handbook checks and maintained generators; final validator result below |
| Completion, custody, endpoint and alias controls | synthetic / local_integration | Four touched fixture modules,379 tests,3 existing skips |
| Earlier native attempts | Their original classes | Existing returned streams/receipts preserved; later success does not replace a failure |
| Model routes and metadata | source_review / local_integration replay | Exact aliases and joined records; where spelling changed and delivered wire effort remain unproved |

The new sanitized source receipt is `evidence/artifacts/new-wsl-install-plan-20261002/fixwave-20261006-verdict-rebase-receipt.json`. It does not declare a new native-ready slot or upstream-suite pass.

### Local commands run

Final GO-PUSH-723 change: exactly one empty line removed at harness-defaults.md:122, plus that file's registry sha256/bytes. Two commits, same base3d7d4a4b; no rebase or other source change. `nice -n 19 ionice -c3 python3 -m unittest tests.test_adoption_docs_consistency` passed39tests/1existing skip in17.985s; final validator69components/10408hashes/4profiles/224receipts0; diff0. A structural comparison proves the registry differs in exactly one file entry. The command center's ACK stands under the explicit range-diff/green-CI condition; neither CI nor landing is claimed by this lane.

Final J723d delta: conformance installs native EXIT/INT/TERM cleanup before the path-length assertion and removes only its frozen owned TMPDIR after process teardown. The early-failure control preserves unrelated cache bytes and the evidence directory; existing success/failure/cancellation controls now also require the allocation gone. Nine focused lifecycle cases passed. After the final landing-cue rebase onto #758, all seven full touched modules passed **589 tests / 3 existing skips in 163.180s**. Final plan0 remains84rows/192commands/113acceptance; F9readonly0 with the same two warnings; native builder/check0; diff0; validator0:69components/10408hashes/4profiles/224receipts. Current handbook pair is c8ecbb5d/d7cc2ccb, with all predecessor output bindings/observations retained. Native [Bash trap documentation](https://www.gnu.org/software/bash/manual/html_node/Bourne-Shell-Builtins.html#index-trap) and the existing identity-checked cleanup seam govern the bounded fix. No other functional delta was added. Completeness critic found no material blocker, source review only. CI/CodeQL and destination apply remain owner gates; the PR stays a draft under the granted landing hold.

Earlier published J723/J723b/J723c delta: all seven full touched modules passed **581 tests / 3 existing skips in 166.987s**, under the CI-pinned Python3.13 environment, explicit private-venv PATH and umask022, at nice19/ionice3. The run includes all four J723c failing modules and the earlier client-config/gateway/conformance modules. Earlier delta attempts remain distinct:99tests/2fixture failures, then focused12pass;468tests/1unsettled synthetic login-path failure, whose unchanged isolated retry passed. No assertion or skip was relaxed. Focused CI4 controls17pass. Then-current validator0:69components/10401hashes/4profiles/224receipts. These remain local integration, not destination acceptance.

The earlier #809-only rebase had proved unchanged foundation bytes. The final #758 landing-cue rebase adds its time-zone/client-config sources, so the full seven-module suite was rerun, with main's source/registry/stack and regenerated publication bindings retained. Review-thread read found no chatgpt-codex-connector threads; the CodeQL thread remains for its scan result.

```bash
nice -n 19 ionice -c3 python3 -B -m unittest -v tests.test_wsl_new_distro_recipe tests.test_agentsview_qualification tests.test_new_wsl_definitive_defaults tests.test_new_wsl_handbook tests.test_new_wsl_client_config tests.test_new_wsl_gateway_composition tests.test_new_wsl_mcp_conformance_lifecycle
nice -n 19 ionice -c3 python3 scripts/validate.py
```

The owner-pending Claude route now produces **needs_owner/78 before loading Promptfoo's configuration**, rather than a failed provider evaluation; malformed supplied routes still fail. Parsed URL host and port must match the plan-owned127.0.0.1:21128 exactly. Both gateway health stages require the keyed microsecond UTC active-start timestamp to be strictly later than both rendered unit files, preserving existing identity and completion gates. CI enumerates nested config files, keeps an exact Inspector script set, uses native Git/Python/env without a forced RTK runtime dependency, and classifies old AgentsView aliases as predecessor fixtures. Source completeness critic found no material blocker; its review ran no tests or host calls.

J723b's native-messaging smoke observation from the dropped A50 draft is retained privately, byte-identical, mode0600, SHA256 **a14356efc2ba3e6c7d44b2d9818538a824b9429fc5d9eaff929ee56e38efee83**. That dropped draft observation is superseded by #771; no public A50 amendment, retired posture or enforcement test is restored.

Earlier source-head validation follows:

```text
python3 -B evidence/artifacts/new-wsl-install-plan-20261002/check_plan.py
rc0:84 rows,192 commands,113 acceptance entries; earlier kind-label mismatch retained.
python3 tools/adoption/new_wsl_client_config.py --check
rc0; two pre-existing Inspector ownership warnings, addressed separately by C11; no apply.
python3 scripts/build_new_wsl_handbook.py --check
rc0.
python3 -B -m unittest -v tests.test_new_wsl_client_config tests.test_new_wsl_definitive_defaults tests.test_new_wsl_gateway_composition tests.test_new_wsl_mcp_conformance_lifecycle
First attempt rc1:379tests/21errors/3skips; undeclared fixture PyYAML import.
Maintained standard-library fixture repair: targeted21cases rc0.
Full retry rc0:379tests,3skips,120.182s; no production/helper/dependency or assertion weakening.
git diff --check
rc0.
nice -n19 python3 scripts/validate.py
rc0:69 components,10385 hashed files,4 profiles,215 receipts. The earlier historical-inventory non-live limitation failure is retained; payload and manifest were clarified together.
```

### Decision record

Current instructions and the complete gateway custody runbook are in the dated addendum to `docs/decisions/2026-10-06-native-plan-gate1-repairs.md`. Its original decision text remains unchanged. The earlier acceptance decision also receives an appended FAST-effort correction. Original historical observations are never rewritten.

### After landing: co-op apply and read-back

The co-op is the sole plan writer. Apply the approved main commit at its booked quiet point; retain the previous gateway prefix and per-asset backup packet. First follow the addendum's gateway custody/reload pre-gates. No lane applies a host change or widens AF_UNIX.

```bash
plan=evidence/artifacts/new-wsl-install-plan-20261002
bash "$plan/install.sh" --only agent-messaging
bash "$plan/accept.sh" --only agent-messaging --stage post_install </dev/null
bash "$plan/install.sh" --only inspect-ai
bash "$plan/install.sh" --only trajectory-analysis
bash "$plan/accept.sh" --only trajectory-analysis --stage post_install </dev/null
# Gateway: execute the owner-custody addendum first; preserve rollback prefix/packet.
bash "$plan/install.sh" --only gpt-gateway
bash "$plan/accept.sh" --only gpt-gateway --stage post_install </dev/null
bash "$plan/accept.sh" --only gpt-gateway --stage service_health </dev/null
bash "$plan/accept.sh" --only gpt-gateway --stage after_sign_in </dev/null
bash "$plan/install.sh" --only research-harnesses
bash "$plan/install.sh" --only skill-vetting
bash "$plan/install.sh" --only mcp-protocol-conformance
bash "$plan/install.sh" --only session-analytics
```

The hcom apply uses the relaxed mapped adapter and native smoke; it has no custom quiet/deny posture. Gateway owner custody, daemon-reload, unit restart and packet-based rollback are fully specified in the addendum. Existing environment credentials are left by pointer and are never copied into the packet. Archive-version rollback for AgentsView additionally requires its consistent pre-migration archive; preserve the existing backup, whose restoration has not been qualified.

For each changed row, retain the install/read-back rc, then one smoke in each client that uses it and subsequent organic counters. Use the canonical shared flock for each exact fresh Claude invocation. Existing repaired-stage commands and returned-artifact gates remain in accept.sh; targetless conformance78 is not a protocol pass. No new measurement campaign or A/B gates this apply.

### Checklist

- [x] Sources pinned; failed attempts and unknown usage preserved.
- [x] No credential values or private client configuration published.
- [x] Peer-owned settlement assertions preserved; no #775 hunks.
- [x] Scoped379-test fixture retry passed; no new skip.
- [x] Final registered-tree validate passed.
- [ ] Exact-head GPT/CC read; the PR remains a draft.
- [ ] Co-op host apply/read-backs and native smoke/counters after landing.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

lane:shared Touches files owned by both lanes; needs both lanes' acknowledgement

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant