Port #435: keep jCodeMunch execute off across all four carriers

[seathatflowsinourveins]

Scope

• Port #435 at head 319af22db969234f98596ed0e515194e26c1be00: the default, researcher, builder and reviewer carrier clauses now leave jCodeMunch route execution off and require the caller's own order arguments. Default and researcher retain menu(query?). The largest block measures 4,099 bytes against the unchanged 4,100-byte bound.
• Base commit: 463a57b983eec540ae90eb45c2b1a7c6fc469aed (main at publication). Head: f95d4e3842179554487b3c50d820205ddf299ba5, after one review repair round on the reviewed head e0c6b38d3257a43d3e7c7974413ce744d1045b0c (see "Review repair round"). The builder prepared the content on 1f5a791b02a230aced670c88bab3d3d0ebcf401a. At publication the content commit was rebased onto main, manifests/evidence.json was rebuilt from main's copy in the last commit, and acceptance was rerun on the committed head.
• Lane: lane:foundation.
• Owned paths touched: four adoption/hooks/claude/token-lanes-block*.md files and their four SHA256SUMS rows; docs/token-session-handbook.md; append-only docs/decisions/2026-09-27-token-lanes-subagent-start.md; one anti-pattern row in docs/harness-defaults.md; two comments in adoption/templates/project.codex.config.template.toml; tests/test_token_lanes_subagent_start.py; the two byte-identical probe artifacts; final hash registration in manifests/evidence.json. The two generators ran and left all four reports unchanged.

route(execute=true) shapes the chosen search action's query from the whole task sentence. The corrected clauses use route for recommendations, then leave the order arguments to the caller. The handbook also gives order("search_symbols", {repo, query: NAME}) for a known name. Every other carrier line and tool grant remains byte-identical; protected owner paths are unchanged and the template's parsed TOML is unchanged.

SOTA sources

The port follows the pinned jCodeMunch argument-shaping implementation, verified against installed 1.108.319 and the identical source blob at v1.108.327.

SOTA sources

• jgravelle/jcodemunch-mcp, 8f7b34abe16fb459e0bf1c04747d584216dfe32e (1.108.319): counter.py L584-590, counter.py L616-630, server.py L5535-5580.
• v1.108.327 counter.py L616-630, peeled commit 6d5ae86c130f96624e2ca2d797fa3b853c210b9d: the same counter blob 0605443d87ed6ec035b14831e99fb7c6540f7b12 as the pinned source and upstream main at the October 3 check. Release published October 2. No component pin changes here.
• #433, merged as 55fc8d17: historical promptfoo 0.123.1 Codex-lane smoke, 6/6 missed route targets and 6/6 passing explicit-order gate rows, including the repaired scoring and its recorded boundaries.
• Historical native probe record and returned probes, copied byte for byte from Token lanes: jCodeMunch route without execute, then order with written arguments #435's pinned head: 5,147 and 3,031 bytes, SHA-256 prefixes 3691f77945bc13b0 and b7d415ab6119eeb6.
• #435's 2026-09-27 cross-family review at 4b8c18ff returned agree. It reviewed the original change; it is not a review of this port.

Evidence-class table

Claim	Evidence class	Command / receipt
Installed 1.108.319 and upstream pinned/current counter bytes agree; route shapes queries from task text	source_review	Installed version/source hashes; pinned counter.py and server.py; identical counter blob at v1.108.327 and upstream main
Original coordinator probes returned the named symbol via explicit order, misranked the bare identifier, and kept the whole sentence in route's template	native_proven (historical local probes; not rerun)	Byte-identical probes.json, three calls on 2026-09-27
Original Codex smoke missed targets 6/6 with route; the explicit-order gate passed 6/6	native_proven (historical workstation smoke; not new acceptance)	#433, including its superseded pre-repair scoring and repaired rows
Hook delivery now carries the corrected clause for all five required role contexts; carrier/handbook old-option exclusions reject old text	local_integration	Failing-first (builder): 19 tests, 10 expected failures; corrected module: 19 tests, OK, by the builder and again at the committed head
Role routing, grant agreement, installation fixtures, owner generation and adjacent contracts pass	local_integration / synthetic	Targeted suite: builder 739 tests, OK, 12 skips (baseline also 12 skips); committed head 739 tests, OK, 12 skips; repair head f95d4e38 739 tests, OK, 12 skips
Carrier sizes/checksums, comment-only TOML, append-only decision, pinned probe identity and registration consistency	local_integration	Independent byte/TOML comparisons; sha256sum; wc; owner diff; diff --check; validate.py; evidence_manifest.py --check
The head lands on current main without touching foreign content	local_integration	Merge-tree landing check (disjoint-drift rule, #608) on main 463a57b9 and head e0c6b38d: LANDABLE; repair round on main 6112d14d and head f95d4e38: LANDABLE

No subagent or Codex-worker run with the new text, new native probe execution, token saving, or host rollout is claimed. The historical README remains unchanged, including its description of the original Codex scope; this port's residuals below define the narrower carried scope.

Local commands run

Builder lines ran on the prepared base 1f5a791b with uncommitted content. Committed-head lines ran at publication on head e0c6b38d, rebased onto main 463a57b9. The published command excerpts omit private temporary/cache setup and log redirections. All acceptance checks ran under nice -n 19, with temporary files under /var/tmp.

$ nice -n 19 python3 -m unittest tests.test_token_lanes_subagent_start
Failing-first (builder, before the carrier edits) exit 1: Ran 19 tests in 2.625s; FAILED (failures=10).
Corrected text (builder) exit 0: Ran 19 tests in 2.490s; OK.
Committed head exit 0: Ran 19 tests in 2.299s; OK.

$ nice -n 19 uv run --no-project --with pyyaml==6.0.1 python -m unittest tests.test_token_lanes_subagent_start tests.test_install_claude_profile tests.test_task_model_routing tests.test_freeze_snapshot tests.test_codex_agents tests.test_codex_worker_lane tests.test_render_config tests.test_adoption_docs_consistency tests.test_token_e2e_preregistration tests.test_new_wsl_client_config tests.test_wsl_new_distro_recipe
Baseline (builder) exit 0: Ran 737 tests in 226.568s; OK (skipped=12).
Corrected content (builder) exit 0: Ran 739 tests in 230.261s; OK (skipped=12).
Committed head exit 0: Ran 739 tests in 228.906s; OK (skipped=12).

$ cd adoption/hooks/claude && nice -n 19 sha256sum -c SHA256SUMS
Exit 0, builder and committed head: all 10 entries OK.

$ nice -n 19 wc -c adoption/hooks/claude/token-lanes-block*.md
Exit 0, committed head: default 4099, researcher 3086, builder 2888, reviewer 2099, scout 1089, verifier 2186 bytes.

$ nice -n 19 python3 tools/adoption/new_wsl_client_config.py --check
Exit 0: check passed at the builder baseline, with the corrected content and at the committed head.

$ nice -n 19 git diff --exit-code origin/main -- adoption/new-wsl tools/adoption docs/decisions/2026-10-02-new-wsl-client-configuration.md docs/decisions/2026-10-01-new-wsl-distro-recipe.md adoption/platforms/linux-wsl2-new-distro.md tests/test_new_wsl_client_config.py tests/test_wsl_new_distro_recipe.py adoption/templates/codex.AGENTS.template.md examples/claude-native/CLAUDE.md adoption/agents blueprints/us-equities
Exit 0, committed head: empty.

$ nice -n 19 git grep -n -F 'execute?' -- adoption/hooks/claude docs/token-session-handbook.md
Baseline (builder) exit 0: seven expected old-text matches.
Committed head exit 1: no matches, as required.

$ nice -n 19 git diff --check origin/main
$ nice -n 19 git diff --check origin/main...HEAD
Exit 0, committed head: clean (both forms).

$ nice -n 19 python3 scripts/component_matrix.py --write
Exit 0, registration on main 463a57b9: {"flip_rule_violations": 0, "rows": 32, "status": "written"}; report bytes unchanged.

$ nice -n 19 python3 scripts/new_host_grand_list.py --write
Exit 0, registration on main 463a57b9: {"status": "written", "layers": 32, "winners": 66}; report bytes unchanged.

$ nice -n 19 python3 scripts/component_matrix.py --check
Exit 0, committed head: {"rows": 32, "status": "checked"}

$ nice -n 19 python3 scripts/new_host_grand_list.py --check
Exit 0, committed head: {"status": "passed", "layers": 32, "winners": 66}

$ nice -n 19 python3 scripts/evidence_manifest.py --check
Exit 0, committed head: {"files": 9564, "status": "passed"}

$ nice -n 19 python3 scripts/validate.py
Baseline (builder) exit 0: status passed, 9541 hashed files.
Before registration (builder) exit 1: expected stale content hashes and two unregistered evidence files, retained as an intermediate result.
Committed head exit 0: {"components": 69, "hashed_files": 9564, "profiles": 4, "receipts": 193, "status": "passed"}
Integrity and scope checks only; no live provider or GPU execution.

$ merge-tree landing check (disjoint-drift rule, #608) on main 463a57b9 and head e0c6b38d
Exit 0: LANDABLE.
ok   1: clean three-way merge
ok   2: merged-vs-main paths 13, 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 12, unowned top-level keys differing []
ok   5: merged files[] sorted by path with no duplicates: True (9564 rows)

$ git push -u origin foundation/jcodemunch-route-carriers-20261003
Exit 0, no hook bypass: the pre-push hook ran its registry tests on e0c6b38d in a detached checkout: Ran 3 tests in 0.218s; OK. The new remote branch reads e0c6b38d.

The registry was rebuilt from main's manifests/evidence.json at 463a57b9: ten changed rows for the port's existing files and two inserted rows for the probe artifacts. Every foreign row equals main's, in main's order, and the twelve port rows equal the builder's registration.

Required CI runs on this head after publication; this description claims no CI result. The builder base's hosted validate job at 1f5a791b failed tests.test_secret_path_guard.K4GuardTests.test_k4_timing, helper k4_f_header: Ran 9951 tests in 1181.107s, FAILED (failures=1, skipped=967). That failure is separate from this port: the secret-path guard files are untouched, and its timing cause was not diagnosed or repaired here. Main's later validate run at cb7dabd7 completed successfully. Every required check must pass on this head before merge.

Review repair round

An independent read-only review by Claude Opus 5.5 at max effort, run as a separate headless session, read head e0c6b38d and returned clear (0 blocking, 0 should-fix, 3 minor). One repair round fixed all three findings, each a wording or link fix:

• Decision record, size section: it now names 1f5a791b as the measurement commit and records that every port path is byte-identical at the PR base 463a57b9.
• Decision record, references to the unpublished work contract: the failing-first paragraph now records the observed results at e0c6b38d (module 19 tests, OK; targeted suite 739 tests, OK, 12 skips), and the residual locator reads (L2828-2838 at dcae68bd).
• This description: the probe and decision-record links are now absolute blob links at f95d4e38.

Commits: ba2f89f7 changes only the addendum (main's 56,858-byte record remains an exact prefix of the 69,053-byte head). f95d4e38, the last commit under the hot-file protocol, changes only that record's sha256 and bytes row in manifests/evidence.json. Both generator checks passed, so no report changed. Main moved to 6112d14d (#649) after publication. It touches no port path, and this round made no rebase or merge.

Repair-round results on head f95d4e38, all under nice -n 19 with temporary files under /var/tmp:

$ python3 -m unittest tests.test_token_lanes_subagent_start
Exit 0 on the edited tree before the commits (record bytes equal the head's): Ran 19 tests in 2.143s; OK. The targeted suite below reruns the module at the head.

$ python3 scripts/component_matrix.py --check
Exit 0: {"rows": 32, "status": "checked"}

$ python3 scripts/new_host_grand_list.py --check
Exit 0: {"status": "passed", "layers": 32, "winners": 66}

$ python3 scripts/validate.py
Exit 0: {"components": 69, "hashed_files": 9564, "profiles": 4, "receipts": 193, "status": "passed"}

$ python3 scripts/evidence_manifest.py --check
Exit 0: {"files": 9564, "status": "passed"}

$ git diff --check origin/main...HEAD
Exit 0: clean.

$ cd adoption/hooks/claude && sha256sum -c SHA256SUMS
Exit 0: all 10 entries OK.

$ git grep -n -F 'execute?' -- adoption/hooks/claude docs/token-session-handbook.md
Exit 1: no matches, as required.

$ git diff --exit-code origin/main...HEAD -- <owner and protected paths>
Exit 0: empty.

$ python3 tools/adoption/new_wsl_client_config.py --check
Exit 0: check passed.

$ uv run --no-project --with pyyaml==6.0.1 python -m unittest <the 11 contract modules>
Exit 0: Ran 739 tests in 214.199s; OK (skipped=12).

$ merge-tree landing check (disjoint-drift rule, #608) on main 6112d14d and head f95d4e38
Exit 0: LANDABLE.
ok   1: clean three-way merge
ok   2: merged-vs-main paths 13, outside PR-owned 0 []
ok   3: main drift 7 paths, overlap with PR-owned inputs (registry excluded) []
ok   4: registry foreign rows equal True, order preserved True, PR-owned rows 12, unowned top-level keys differing []
ok   5: merged files[] sorted by path with no duplicates: True (9565 rows)

$ git push origin foundation/jcodemunch-route-carriers-20261003
Exit 0, no hook bypass: the pre-push hook ran its registry tests: Ran 3 tests in 0.202s; OK. The remote branch reads f95d4e38.

Decision record

docs/decisions/2026-09-27-token-lanes-subagent-start.md#addendum-2026-09-27-jcodemunch-route-arguments preserves the need, historical facts, evidence classes, erratum, alternatives and comparison that would overturn the change. All four block and alternative sizes were recomputed at the builder's base 1f5a791b; main's five later commits change none of the port's paths.

Residuals:

• The Codex managed-block section and its two original test edits are not ported. The configuration owner's new-WSL dropped-list assertion now lies at tests/test_new_wsl_client_config.py L2861-2869; that decision belongs to the configuration owner or token lane.
• The three custom Codex agents and host rollout are out of scope. Installed carrier text remains old until the host owner runs install_claude_profile.py --only guard.
• The current-distro Gate A freeze was lifted by the October 1 program; the new-distribution map leaves the carriers unwired. No owner map, record, generated block or test changes here.
• A merged decision removing jCodeMunch from carriers, or an upstream shape_execute_args change, overturns this port. The wave-2 context-mode-only sentence remains a coordination draft, not an accepted decision. At publication, main's Token stack winner: one full stack chosen on recorded evidence, provisional until Gate A (decision record) #508 retirement record states that it changes no carrier, and the Add macOS host receipts for 9 token-efficiency components + jcodemunch-mcp (issue #276) #392 retirement changes no carrier path.
• Currency wave W1: eight foundation pins to their latest clean upstream releases (mcporter held) #642 was OPEN and unmerged at publication. If its pin change lands before this port merges, refresh the base, cite the merged pin and redo registration and acceptance.
• An independent read-only review of this port by Claude Opus 5.5 at max effort, run as a separate headless session, read head e0c6b38d and returned clear: 0 blocking, 0 should-fix, 3 minor. All three minor findings were wording or link fixes and are fixed at f95d4e38; no review residual remains. The original review at 4b8c18ff does not substitute for it.
• Token lanes: jCodeMunch route without execute, then order with written arguments #435 stays open until this port merges; it is then closed unmerged with a comment citing this PR and its merge commit.

Host evidence

Not applicable: no file under evidence/hosts/ changes and no platform-status change is made.

• python3 scripts/host_receipts.py validate passes for every new/changed host receipt (not applicable).
• Independent review is recorded (scripts/host_receipts.py review) or explicitly requested in this PR (the coordinator's review ran at e0c6b38d, verdict clear; its record is a PR comment, and its three minor findings are fixed at f95d4e38).
• 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 (no workflow change).
• New/changed workflows declare top-level permissions: contents: read (no workflow change).
• No secrets are printed, logged or committed; no new required secret was added without an owner.
• No new paid hosting, subscription or billing surface was introduced.
• Peer-owned untracked files and worktrees were preserved.

🤖 Generated with Claude Code

[seathatflowsinourveins]

Claude session native-agent-stack-0c: review record and repair round for this port.

Review. An independent read-only review by Claude Opus 5.5 at max effort, run as a separate headless session, read head e0c6b38d3257 and returned verdict clear: 0 blocking, 0 should-fix, 3 minor.

Its checks

1. Contract fidelity: PASS. 13 files, all on the allowed list; a content commit plus a last registry-only commit; each block equals its base with only the jCodeMunch clause replaced; blocks measure 4099/3086/2888/2099 bytes within 4,100; scout and verifier unchanged; the record is append-only; the probes are byte-identical to Token lanes: jCodeMunch route without execute, then order with written arguments #435's head; protected paths are unchanged.
2. Evidence policy: PASS. The probes and the Capability gate: promptfoo codex-sdk harness for the Codex jcodemunch, ai-memory and M13 gates, reconciled with native OTel #433 smoke stay labelled historical; no local test, summary or replay is presented as upstream acceptance or a new run; the cited upstream lines say what the records state. Boundary: the reviewer did not re-fetch v1.108.327 or upstream main.
3. Registry: PASS. Only the 10 port rows change and 2 probe rows are inserted, all 12 recomputed; evidence_manifest.py --check, validate.py, component_matrix.py --check and new_host_grand_list.py --check exit 0.
4. Privacy: PASS. A case-insensitive grep of every added line and the PR body for host user names, home and state paths, private file names, the user's email and token patterns returned 0 matches, with a matching positive control.
5. PR body claims: PASS. Base, head and label match, and the stated committed-head results reproduce (module 19 tests OK; the 11 contract modules 739 tests OK). Not verifiable with its allowed commands: the base CI failure run, main's earlier validate run and the pre-push hook claim.
6. Other: no blocking or should-fix issue. No test is weakened, the two new tests fail on the old text and pass on the new, Token lanes: jCodeMunch route without execute, then order with written arguments #435's objection window was clean, and Currency wave W1: eight foundation pins to their latest clean upstream releases (mcporter held) #642 is open and unmerged.

Findings and dispositions

• Minor, decision record size section (it called 1f5a791b "the worktree's base"): fixed in ba2f89f7. It now names 1f5a791b as the measurement commit and records that every port path is byte-identical at the PR base 463a57b9.
• Minor, decision record references to the unpublished work contract (the failing-first paragraph and a residual locator): fixed in ba2f89f7. The paragraph now records the observed results at e0c6b38d (module 19 tests, OK; targeted suite 739 tests, OK, 12 skips), and the locator reads (L2828-2838 at dcae68bd).
• Minor, relative links in the PR description: fixed with gh pr edit. The probe and decision-record links are now absolute blob links at f95d4e38.

No finding became a residual, and no test or carried byte changed.

New head f95d4e3842179554487b3c50d820205ddf299ba5. ba2f89f7 changes only the addendum's wording; f95d4e38, the last commit under the hot-file protocol, changes only that record's sha256 and bytes row in manifests/evidence.json. Exit codes on the new head, under nice -n 19:

• python3 scripts/validate.py: 0, status passed (9564 hashed files)
• python3 scripts/evidence_manifest.py --check: 0 (9564 files)
• git diff --check origin/main...HEAD: 0
• component_matrix.py --check: 0; new_host_grand_list.py --check: 0 (no report rewritten)
• the 11 contract test modules (with PyYAML 6.0.1): 0, Ran 739 tests, OK (skipped=12)
• sha256sum -c SHA256SUMS: 0 (10 OK); git grep -F 'execute?' over the carriers and handbook: 1, no match, as required
• owner and protected path diff: 0; new_wsl_client_config.py --check: 0
• merge-tree landing check (disjoint-drift rule, Client configuration for the new distribution: wire only what the definitive manifest installs (map, tool, F9) #608) on main 6112d14d and head f95d4e38: 0, LANDABLE
• push: 0, pre-push hook 3 tests OK, no bypass

Main moved to 6112d14d (#649) after publication. It touches no port path; this round made no rebase or merge.

Before merge: the work contract requires no acknowledgement from another lane (this is lane:foundation only); the Codex root lane's exact-head read of f95d4e38 is still needed; and all 8 of 8 required checks must be green on f95d4e38.

[seathatflowsinourveins]

Claude session native-agent-stack-5f: merged current main into this PR with one merge commit (no rebase, no force-push). The PR is not merged and #435 is untouched.

New head: a529f72214d9981feca3d3be8d499555609fca5b

Commits on top of the previous head 10e6e1f3:

1. 7ec6ba3c: merge of main at 1a64e8f3ec104207f7659d03d07d112d12588278 (parents 10e6e1f3 and 1a64e8f3).
2. 87764f8b: hot-file protocol, manifests/evidence.json reset to main's copy.
3. a529f722: final commit, the branch's registry reapplied on main's copy.

The tree of a529f722 equals the merge commit's tree (037cdd99). Against main, the head changes exactly this PR's 13 files. The ten owned files outside the conflicts are byte-identical to 10e6e1f3, and the two probe files still match b7d415ab6119eeb6 (3,031 bytes) and 3691f77945bc13b0 (5,147 bytes).

Conflicts and resolutions (diffs against main)

1. adoption/hooks/claude/SHA256SUMS (generated, not hand-merged). Main changed the secret_path_guard.py row and this branch changed the four carrier rows, in adjacent hunks. No generator script exists for this file; its rows come from sha256sum, and install_claude_profile.py and tests/test_install_claude_profile.py read it. I regenerated it with sha256sum over the merged tree, keeping the file's row order. The result is byte-identical to fresh sha256sum output (cmp exit 0), and its rows cover the directory's nine hook files plus the guard. Main's new guard row is kept. Diff against main:

diff --git a/adoption/hooks/claude/SHA256SUMS b/adoption/hooks/claude/SHA256SUMS
index 2dc78e38b..4ae78fba1 100644
--- a/adoption/hooks/claude/SHA256SUMS
+++ b/adoption/hooks/claude/SHA256SUMS
@@ -4,4 +4,4 @@ c0979c22a701538a267e8f6b980dd2228e79db0c0b2b6bcf8d91dd2681735fb7  currency-due-n
-d6c3c19d6e94d3f48d2788963c8c9be45ec57e6bb81cab92b2c30d9fd5c8e8d6  token-lanes-block.md
-c81a91c4ffb2fa906feb403d5a11bd294b6cf2db45795d80f9acd662c5004973  token-lanes-block.builder.md
-47cc80b262468557b8cfe5c243598030757270674bc7bf286fba8c0c573abc28  token-lanes-block.researcher.md
-19857a44a201948aa1d1fc9be3e24d5f3f7cc385f333ab3893411889258af4ed  token-lanes-block.reviewer.md
+dcd6996e9247f3c8f3d543545eb958518236b21449492ddf92721590aa3f60c8  token-lanes-block.md
+2f7a53031e6eabf6b9d08989b619b74204fa36dac2559e520fb72b06225a81d3  token-lanes-block.builder.md
+149ac69352a788ad67c84b4d545cab461015c4e73a9ba4c8675259402c955e4a  token-lanes-block.researcher.md
+ec7c6b2a80c58e0156daab25b7cfd4eea6f2e3eac77137067051601045f9d2f8  token-lanes-block.reviewer.md

2. docs/harness-defaults.md. Main added three 2026-10-03 anti-pattern rows at the same anchor where this branch added its 2026-09-27 row. I kept main's rows unchanged and in order. This branch's row is byte-identical to 10e6e1f3 and sits directly after main's three rows, before the first 2026-10-02 row, which it also preceded on the branch. Main has no row for the same mistake, so no row was dropped. Diff against main (one added line, no removed lines):

docs/harness-defaults.md diff against main

diff --git a/docs/harness-defaults.md b/docs/harness-defaults.md
index fad5a26aa..d1e008334 100644
--- a/docs/harness-defaults.md
+++ b/docs/harness-defaults.md
@@ -91,0 +92 @@ When a claim or action proves wrong, correct it where it was relayed and record
+| 2026-09-27 | Recommending a tool's execute mode without reading how it builds the call | The token-lanes carrier offered `jcodemunch route(task, repo?, execute?)`. At jcodemunch-mcp 1.108.319, `route(execute: true)` passes the whole task as the chosen action's query ([`counter.py` L616-630](https://github.com/jgravelle/jcodemunch-mcp/blob/8f7b34abe16fb459e0bf1c04747d584216dfe32e/src/jcodemunch_mcp/counter.py#L616-L630), identical blob at [v1.108.327](https://github.com/jgravelle/jcodemunch-mcp/blob/v1.108.327/src/jcodemunch_mcp/counter.py#L616-L630)). On the Codex lane it returned neither target symbol in 6 of 6 runs ([#433](https://github.com/seathatflowsinourveins/native-agent-stack/pull/433)), and a native probe routed a bare identifier to a state-changing action ([probes](../evidence/artifacts/jcodemunch-route-args-20260927/probes.json)) | Read the installed version's argument shaping and probe it natively before an instruction recommends an execute or auto mode. Otherwise recommend the explicit call with written arguments | `test_injected_jcodemunch_rule_leaves_execute_off` and `test_jcodemunch_carriers_and_handbook_omit_execute_option` in `tests/test_token_lanes_subagent_start.py` |

3. manifests/evidence.json (merge path in docs/lanes.md). I took main's copy (git checkout --theirs) and re-registered the branch's 12 files with host_receipts.register_file: the ten modified files that main's manifest lists and the two new evidence files. component_matrix.py --write and new_host_grand_list.py --write changed no report. Against main, the registry differs in 12 files[] rows, and each row matches its file's sha256 and byte count. No top-level key differs, and the 9,690 rows are sorted with no duplicates.

Registry-last proof:

• git diff origin/main HEAD~1 -- manifests/evidence.json is empty (exit 0).
• git patch-id --stable gives 355e699abf94eb7c53a085495cd0e4bd2597fc29 for both git diff origin/main HEAD -- manifests/evidence.json and git diff HEAD~1 HEAD -- manifests/evidence.json.
• The reset commit and the final commit each touch only manifests/evidence.json.

manifests/evidence.json diff against main (114 lines)

diff --git a/manifests/evidence.json b/manifests/evidence.json
index e0f02907d..8e3432a85 100644
--- a/manifests/evidence.json
+++ b/manifests/evidence.json
@@ -4338,28 +4338,28 @@
     },
     {
       "path": "adoption/hooks/claude/SHA256SUMS",
-      "sha256": "9c5ebc195a0b3aae6d7732eb7eb12ad2fddb68c318383c7a3bbf988e4e846528",
+      "sha256": "cb5b70d1bc86d6a8b5049e82e3136301b1e30a5b9a50ceb254e276c789db1ecf",
       "bytes": 950
     },
     {
       "path": "adoption/hooks/claude/token-lanes-block.builder.md",
-      "sha256": "c81a91c4ffb2fa906feb403d5a11bd294b6cf2db45795d80f9acd662c5004973",
-      "bytes": 2877
+      "sha256": "2f7a53031e6eabf6b9d08989b619b74204fa36dac2559e520fb72b06225a81d3",
+      "bytes": 2888
     },
     {
       "path": "adoption/hooks/claude/token-lanes-block.md",
-      "sha256": "d6c3c19d6e94d3f48d2788963c8c9be45ec57e6bb81cab92b2c30d9fd5c8e8d6",
-      "bytes": 4088
+      "sha256": "dcd6996e9247f3c8f3d543545eb958518236b21449492ddf92721590aa3f60c8",
+      "bytes": 4099
     },
     {
       "path": "adoption/hooks/claude/token-lanes-block.researcher.md",
-      "sha256": "47cc80b262468557b8cfe5c243598030757270674bc7bf286fba8c0c573abc28",
-      "bytes": 3075
+      "sha256": "149ac69352a788ad67c84b4d545cab461015c4e73a9ba4c8675259402c955e4a",
+      "bytes": 3086
     },
     {
       "path": "adoption/hooks/claude/token-lanes-block.reviewer.md",
-      "sha256": "19857a44a201948aa1d1fc9be3e24d5f3f7cc385f333ab3893411889258af4ed",
-      "bytes": 2088
+      "sha256": "ec7c6b2a80c58e0156daab25b7cfd4eea6f2e3eac77137067051601045f9d2f8",
+      "bytes": 2099
     },
     {
       "path": "adoption/hooks/claude/token-lanes-block.scout.md",
@@ -4573,8 +4573,8 @@
     },
     {
       "path": "adoption/templates/project.codex.config.template.toml",
-      "sha256": "cf8adbabcd3db19ed32f53d9868e98bbb3b774cb6d1a7f2ab3e709a930ae227e",
-      "bytes": 2701
+      "sha256": "a6822a173f78b007478480f9e681c012ef60e75f17f36091ad1d1470fd6258af",
+      "bytes": 2941
     },
     {
       "path": "adoption/templates/systemd/omniroute-fw.service",
@@ -15929,8 +15929,8 @@
     },
     {
       "path": "docs/decisions/2026-09-27-token-lanes-subagent-start.md",
-      "sha256": "9a300299c01e31908b754bf4d003fe80e8fd279b5ca881c0e86d57d61b8f9762",
-      "bytes": 56858
+      "sha256": "423b886bef0774ecc4a2835a55e6a0d6d8f50ed55c0a53ed77d855dd241289d6",
+      "bytes": 69053
     },
     {
       "path": "docs/decisions/2026-09-28-an13-m4-m5.md",
@@ -16229,8 +16229,8 @@
     },
     {
       "path": "docs/harness-defaults.md",
-      "sha256": "3ec6b6592eb4deaeffef6780f47f30512c2d012c1215324b22c97c37af3200b7",
-      "bytes": 146676
+      "sha256": "e181324961a3fa2f8801e2f5874b9ebce4367c4c88c114a3f38816d2ea47af82",
+      "bytes": 147886
     },
     {
       "path": "docs/harness-rules-convergence-20260922.md",
@@ -16479,8 +16479,8 @@
     },
     {
       "path": "docs/token-session-handbook.md",
-      "sha256": "173fe5e0cf2121eb7d3ad76bf397843e24f52a2fd388a17a133e2691d61fc3d5",
-      "bytes": 90784
+      "sha256": "b85424755d2152313a6a7cfcbef8871c98f0f8f756e932287fa1aa1bf7e49a46",
+      "bytes": 92944
     },
     {
       "path": "docs/ultracode-token-routing-20260921.md",
@@ -32467,6 +32467,16 @@
       "sha256": "c3926e64aa7bc852dd3ab2034aafb41486192584818b7303e8f0aa2508a9d51c",
       "bytes": 2065
     },
+    {
+      "path": "evidence/artifacts/jcodemunch-route-args-20260927/README.md",
+      "sha256": "3691f77945bc13b05210623bc68c34f293a52e6736365eb3ba92f6c12f98eeb9",
+      "bytes": 5147
+    },
+    {
+      "path": "evidence/artifacts/jcodemunch-route-args-20260927/probes.json",
+      "sha256": "b7d415ab6119eeb6ed31bff86a0a805dca14dda5a2fde20389d97a112e2e241f",
+      "bytes": 3031
+    },
     {
       "path": "evidence/artifacts/jev-p1-20261003/README.md",
       "sha256": "bd0149a7003e7c05a80dd864491fbb565dc4b7aedaae5f03095a2d8236ceb032",
@@ -51824,8 +51834,8 @@
     },
     {
       "path": "tests/test_token_lanes_subagent_start.py",
-      "sha256": "13276b5068067da4786c6ac4c020562a10fb4de22c3d7e4345c4662932d0366e",
-      "bytes": 23034
+      "sha256": "8d2ed091ed91928e88b74806f38eb146c30dd5fde3faf6fb43e52381163e7ae7",
+      "bytes": 24974
     },
     {
       "path": "tests/test_token_measurement.py",

Checks on a529f722

These are local integration checks on this host, run at nice 19 with TMPDIR outside the checkout. They are not CI. This session has not observed CI on the new head.

Check	Exit	Result
Contract's targeted suite (11 modules, uv run --no-project --with pyyaml==6.0.1 python -m unittest ...)	0	Ran 753 tests in 209.671s; OK (skipped=12)
sha256sum -c SHA256SUMS in adoption/hooks/claude	0	10 of 10 OK
Carrier sizes (wc -c, limit 4,100)	n/a	default 4,099; researcher 3,086; builder 2,888; reviewer 2,099; scout 1,089; verifier 2,186
git grep -n -F 'execute?' -- adoption/hooks/claude docs/token-session-handbook.md	1	no match, as acceptance requires
python3 scripts/validate.py	0	"status": "passed", 9,690 hashed files, 193 receipts
python3 scripts/evidence_manifest.py --check	0	passed, 9,690 files
python3 scripts/component_matrix.py --check	0	checked, 32 rows
python3 scripts/new_host_grand_list.py --check	0	passed, 32 layers, 66 winners
python3 tools/adoption/new_wsl_client_config.py --check	0	check passed
Owner and forbidden paths, contract's git diff --exit-code origin/main -- ...	0	unchanged
git diff --check origin/main...HEAD	0	clean
Merge-tree landing check	0	LANDABLE

The suite ran 753 tests, against the 739 recorded before. The 14 extra tests come from main: it changed five of the eleven modules, and its edits added tests to two of them, test_new_wsl_client_config (146 to 159) and test_render_config (35 to 36). These are static def test_ counts, which total 739 at 10e6e1f3 and 753 at the new head.

Landing check output

main 1a64e8f3ec104207f7659d03d07d112d12588278 head a529f72214d9981feca3d3be8d499555609fca5b base 1a64e8f3ec104207f7659d03d07d112d12588278 merged-tree 037cdd99a59acc7ea93f5e2bcb934bfd4d379664 merge-tree-exit 0
ok   1: clean three-way merge
ok   2: merged-vs-main paths 13, 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 12, unowned top-level keys differing []
ok   5: merged files[] sorted by path with no duplicates: True (9690 rows)
LANDABLE

Notes

• Currency wave W1: eight foundation pins to their latest clean upstream releases (mcporter held) #642 is still open, and the jCodeMunch pin on main is still 1.108.319, so no re-citation is needed.
• Main's new-WSL client-configuration decision keeps the token-lane carriers unwired on the new distribution. It does not remove jCodeMunch from the carriers, and main left the four carriers unchanged, so no contract stop condition applies.
• git merge origin/main exited 1 with exactly the three expected content conflicts; there were no others.

[seathatflowsinourveins]

Claude session native-agent-stack-5f: landing at head afee0aea80a8cf7651535eba7262e154fe5c72a3. The command center (wsl-architecture-design) gave its ACK at 0d1515036220, 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 f95d4e384217; 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 its 1 P1 item(s) resolved in one repair round as recorded.

Observed main ba0e8c48f339f8327cb1c2ee9abdfc6ca9bbce9c. coordination/merge_tree_landing_check.py (sha256 a11649fcf2e5844c…) exit 0:

main ba0e8c48f339f8327cb1c2ee9abdfc6ca9bbce9c head afee0aea80a8cf7651535eba7262e154fe5c72a3 base ba0e8c48f339f8327cb1c2ee9abdfc6ca9bbce9c merged-tree 8a49e68789d7c76850694fcf2e46feded5d11f97 merge-tree-exit 0
ok   1: clean three-way merge
ok   2: merged-vs-main paths 13, 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 12, unowned top-level keys differing []
ok   5: merged files[] sorted by path with no duplicates: True (9810 rows)
LANDABLE
condition 5: merged files[] sorted and unique (tree 8a49e68789d7c76850694fcf2e46feded5d11f97)

Required checks at this head: 8 pass . Unresolved review threads: 0.

[seathatflowsinourveins]

Claude session native-agent-stack-5f: post-merge observation. Landed as b5b9c9ddbb8974f23e621d296feb748888d963ba (parent ba0e8c48f339f8327cb1c2ee9abdfc6ca9bbce9c), tree 8a49e68789d7c76850694fcf2e46feded5d11f97. The landed tree equals the checked merged tree and the parent is the observed main.