Record the paired-run results (planning, orchestration) and pin the RTK receipt's locators - #1005

…ng for review micros)

Codex review in GitHub at bf864c3, four P2s, each reproduced before the change:
- background-workflows: the route now says a saved Workflow in a headless run fenced with --setting-sources local is launched by scriptPath (its name was not found in 6 of 6 runs of paired run 1, arm B) and that
  a long verifier is cut by the wait ceiling; Overturn when says which measured failures the route carries, and the slots.md index cell follows (a new check compares every row's Overturn when across the three surfaces).
- agent-teams note: "won no fixture" is replaced by the counted result against the single session: 1 of the 5 fixtures where its diff was captured (F2), none of F3 to F6.
- tests/test_rtk_receipt_locators.py: a pin whose commit is present but whose tree is missing (a partial clone) is skipped; a bad path with the tree present still fails.
- usage-cost-monitoring: the metering sentence is rewritten from a committed artifact (meter-windows-20261011.json) with a test that recomputes it. The "about 1.9 times" figure was one instant's (2.56 and 1.64 over two
  fixed 24-hour windows) and the clause that the events leave the advisor out contradicted the advisor run's Stage 0; both are gone.

Paired run 5, as the command center directed: routing_5 in paired-runs.json (20 runs, the 12 decisive findings with the runs that found each, costs, hashes), RoutingRunTests recomputing the strict verdict from the rows
(Opus stays), the record's addendum paragraph, run 4 marked stopped and run 5 removed from the owed table. No slot row, default or skill text changes.

Co-Authored-By: Claude Sonnet 5.5 <[email protected]>

…ule for retired slots

Answers the Codex review in GitHub of #1001 at ba6ab3a (four P2s) and the co-op's GPT first read at the same head (three P2s, the first three below); each was reproduced before the change.
- Sentence units split at a full stop or a semicolon whatever follows, and at a line break: "This route was retired; the current default is claude-code-action." was one unit that said retired.
- The generic phrases "in Actions", "through Actions" and "under WIF" count as the retired route only with Claude context, judged per leaf string; the named workflows, "through #909" and the managed app count
  everywhere. The routes the decision retains (every bullet of its "What remains" section, a catalog job in GitHub Actions, a non-model gate under WIF) are accepted.
- The skill's entry point SKILL.md is scanned, and a test requires every file of the skill folder to be read by a check.
- The one-default rule names the retired exception in slots.json, in the slots.md copy and in step 3 of SKILL.md (a retired row has none and says what replaced the route).
- The ruled default text is pinned exactly (the read's measured boundary: the date changed on all three surfaces passed), and a check compares every row's Overturn when across slots.json, the index and the
  layer pages (one step wider than the review of #1005, where an edit left the index cell behind).
Each finding fails its own test when its fix is undone alone, and 43 single-edit mutations are caught; the old guard accepts the four bad scenarios of the read and rejects its two legitimate ones.

Co-Authored-By: Claude Sonnet 5.5 <[email protected]>

…uling, count the totals in the tolerant floor, state the order of the second trial

Answers the command center's Claude first read of #1005 at 5c5eeca (three P2s, five P3s) and Codex's re-review in GitHub of the same head (one P1, four P2s); each was reproduced before the change.
- The ruling file the record cites was edited by the command center after the record cited it (CC P2-1): routing_5.ruling and papers.ruling carry the current sha256 and size, the record's paragraph is regenerated from them, and a test that is
  enabled by PAIRED_PAPER_ROOTS resolves every private paper of the artifact against its recorded hash, so a paper that changes after it is cited is noticed on a host that has it.
- The ruling's condition for Sonnet at medium (the change is mechanical, or a second Claude-family read exists) was dropped from the artifact and the paragraph (CC P2-2); it is in both now, with the clause that the allowance is the command center's
  routing discretion and not something run 5 measured (CC P3-2).
- The slot-manifest label named an older slots.json and its list lacked three pairs, so the exact-difference check skipped (CC P2-3): the label is the sha256 of the shipped file and the list is recomputed (background-workflows route and overturn_when,
  usage-cost-monitoring evidence, and the agent-teams overturn condition edited below).
- Codex P1: the second trial was added by the command center after trial 1 had met the registered rule (addendum 3 of the run's papers says so). The paragraph and the decision now state that the two-trial verdict is a ruled extension and not a preregistered
  outcome, and that an extension after a pass can only move the verdict toward Opus.
- Codex P2 and CC P3-1: the tolerant reading also bounds the totals, and the floor printed (0.62) modelled the per-finding clause only; the full reading's floor is 0.56 (exact convolution, checked against enumeration). Both figures are kept and labelled, the
  correction is recorded in a new private note (hash in papers.correction), and the strict rule's chance with each finding's own pooled rate (0.73, a plug-in calculation) is added.
- Codex P2 (workers.md:29): the agent-teams overturn condition names the class the default is scoped to (work where workers must message each other); paired run 1 measured bounded fixes, which fall outside it.
- Codex P2 (workers.md:76): the task-list alternative states that it has no completion condition of its own and needs the external done check of loops-completion, as the measured F5 runs used all 8 sessions without declaring done.
- Codex P2 and CC P3-3: a blob that is absent in a blobless partial clone skips the locator test and a path that is not in the tree at the pin still fails.
- CC P3-4: the meter test's tolerance is 0.01. CC P3-5: the addendum's heading names both dates.
Each correction fails its own test when undone alone: 31 single-edit mutants of the run-5 data and text and 7 of the locator check are caught.

Co-Authored-By: Claude Sonnet 5.5 <[email protected]>

…uard's other gaps

Answers the command center's Claude micro of #1001 at ef3f3e7 (P2-1, P3-1 to P3-5) and Codex's four threads of the same head (:43, :63, :122, :211). Each case was reproduced before the change:
compare-guards-ef3f3e7b-vs-third-commit.txt runs 19 scenarios through the guard of ef3f3e7 and this one; the old guard is wrong on 13 (it accepts a retired route that is reintroduced), this one on none.
- P2-1 (correction #160): joining wrapped lines into one paragraph let "Retired 2026-10-10" followed by a line that says the current default is claude-code-action pass on four prose surfaces (SKILL.md, slots.md,
  cross-client.md, layer-page preambles), the scenario Codex raised at ba6ab3a line 41. Prose is judged one physical line at a time again, a line break ends a unit, and the message of every line-judged surface
  states the rule (RULE; the legend of slots.md states it too and a test pins the two to each other).
- P3-1 and Codex :43: three more names of the same route, all in the vendor's own text: the page's title "Claude Code GitHub Actions" and its "installs the Claude GitHub App"
  (code.claude.com/docs/en/github-actions.md, capture sha256 87b99aef, lines 5, 9, 33, 63-64), the mirror claude-code-base-action (anthropics/claude-code-action@2dca132ff0e0:
  .github/scripts/check_workflow_hardening.py:18, .github/workflows/sync-base-action.yml:1 and 43, CLAUDE.md:20) and the carrier named first ("Use the GitHub Action for Claude reviews", generic
  like "in Actions": it counts when the field or unit is about Claude). The families "in Actions" and "through Actions" take on, with, inside and via.
- Codex :122 and the command center's P3-3 and P3-5: a unit that negates the retirement outright ("not retired", "never retired", "no longer retired", "isn't retired", "unretired") no longer counts as saying
  retired. The other half of that scenario, "was retired, but is now current", is the boundary the command center ruled (the guard is a tripwire for stale text; review reads what it accepts); the module names it.
- Codex :211 and P3-3: a layer-page section with a label twice (a second Route or Overturn when bullet) is a violation, because the page-to-slots.json comparison keeps the first bullet only.
- Codex :63 and P3-4: the target of every slots.md row is checked against <page>#<id>, from the layer of the slot.
- The description line of SKILL.md names the topic that invokes the skill; it is pinned by sha256 to the text the routing was measured with and read as a topic, not as a route.
- Scheduled-tasks (Codex on #1005, scheduling-supervision.md:9, and the command center's ruling): the route no longer bans every model-backed job. The retirement record
  (docs/decisions/2026-10-10-retire-in-github-claude-review.md:5-6, :43) retires Claude review inside GitHub and keeps Codex's GitHub review, so the route says GitHub Actions schedules repository jobs and a
  scheduled Claude task runs locally; a test pins the wording on slots.json, the layer page and the alternatives.
The label of claude-native-slot-manifest.json equals the sha256 of the shipped slots.json again (relabel last); test_the_label_names_the_shipped_slots_file and test_no_other_exported_field_differs
(no skip) are unchanged from ef3f3e7. 72 of 72 single-edit mutants of the guard and its data are caught and 5 controls pass (mutations-v4-forward-commit-3.md).

Co-Authored-By: Claude Sonnet 5.5 <[email protected]>

… the registered trigger, the time bounds, the basis of the tolerant floor

Answers the command center's Claude micro of 425240f (PASS_WITH_P3: P3-1 to P3-7) and Codex's three new threads of the same head (workers.md:61, test_paired_runs_results.py:91, scheduling-supervision.md:9). Each was
reproduced before the change: five tests of this commit fail on the data of 425240f (retained output fail-before-micro3.txt).
- P3-1: decision.two_trial_verdict_basis no longer says that the decision to run two trials was not fixed before any run. Addendum 1 (paired-routing/addendum-01-ruling-and-extension-20261011.md, sha256 1917091f, line 13,
  written before any run) fixed the trigger too: after a passing trial 1 the run ends (Sonnet takes review micros, no second trial). The command center ran trial 2 after seeing that result (addendum 3, pinned beside the
  basis with its path, size and hash), so the two-trial verdict stays a ruled extension and not a preregistered outcome. The paragraph of the record is regenerated from the artifact; the ruling's sidecar correction
  (sha256 338600a9) is pinned as papers.ruling_correction and the ruling's own bytes are unchanged; ruling.routing says "read or micro" like the ruling.
- Codex P1 (test_paired_runs_results.py:91): the tolerant floor's reading and null model are the run's own, registered before any run (addendum 1, lines 17-18: the reading with the clause on the totals, and the noise floors of
  two identical reviewers who each find a finding with probability p in each run); no third-party source defines them, so the reference is the brute-force enumeration that already fails before the correction. The
  convolution of the per-finding differences is the one numpy 2.5.3 documents for numpy.convolve ("the sum of two independent random variables is distributed according to the convolution of their individual
  distributions"); a test cross-checks tolerant_floor against numpy.convolve where numpy is installed, the artifact records both bases (decision.tolerant_floor_basis), and the gated papers test reads lines 17-18 of the
  retained addendum.
- Codex P2 (workers.md:61) and P3-3: the orchestration-frameworks row and the record said "a 45-minute ceiling per run". The registered design has two bounds, a 50-minute process cap for a headless session and a 45-minute
  cap for the team (paired-runs.json design.caps; preregistration.md line 156, sha256 531758c4), and the Workflow's verifier was cut by the client's own 45-minute wait for background tasks: the stderr of the runs F1-B, F3-B,
  F4-B and F5-B reads "Background tasks still running 45m after the last turn", after 2779 to 2794 seconds, inside the 50-minute cap. The row (slots.json and workers.md) and the record say so, and
  test_the_time_bounds_the_row_and_the_record_state_are_the_registered_ones computes the counts from the run rows (4 of 6 Workflow runs, 4 of 6 team runs of which 3 at the team cap).
- Codex P2 (scheduling-supervision.md:9): that text is #1001's (commit ba6ab3a). This commit carries the same wording as #1001's third commit (efaa077): GitHub Actions schedules repository jobs, Claude runs inside GitHub
  are retired (2026-10-10, docs/decisions/2026-10-10-retire-in-github-claude-review.md:5-6, :43), so a scheduled Claude task runs locally. The test of the wording is that commit's.
- P3-7: the RTK locator test cites git 2.53.0 git-ls-tree(1), OUTPUT FORMAT, for the entry it splits.
The label of claude-native-slot-manifest.json equals the sha256 of slots.json again (relabel last). Its strict test arrives with #1001.

Co-Authored-By: Claude Sonnet 5.5 <[email protected]>

…the runner's setting, not the client's

The command center's Claude micro of #1005 at 49bb649 (P2-1, P3-2, P3-3); each item was reproduced on that head before the change.
- P2-1: the previous commit said the Workflow's verifier was cut by "claude -p's own 45-minute wait for background tasks". The runner set CLAUDE_CODE_PRINT_BG_WAIT_CEILING_MS=2700000 for every headless run
  (paired/orch_run.py line 137); the client's default is 10 minutes (the retained env-vars.md line 357: "Default: `600000`, or 10 minutes"; headless.md line 77: "after 10 minutes of continuous idle waiting by default"). The decision
  is unaffected: the runs show exactly what a 45-minute ceiling produces (the four Workflow runs that ended there ran background work for 2,702 to 2,703 seconds, the ceiling being 2,700). One rewording pass over the
  orchestration-frameworks and background-workflows rows (slots.json), their pages (workers.md) and the record, with the setting in design.caps and the three sources as papers (the runner file, retained and hashed; the two
  documentation lines, hashed and checked on their lines by the gated test).
- The test pins of the old wording are inverted and fail on the data of 49bb649 (fail-before-wave.txt); test_the_background_wait_ceiling_is_the_runners_setting_and_the_clients_default_is_ten_minutes computes the 2,700 seconds
  from the run rows; 13 single-edit mutants are caught and the control passes (mutations-wave.md).
- P3-3: the git-ls-tree quote of the RTK locator test is exact (git 2.53.0, OUTPUT FORMAT and -z).
The label of claude-native-slot-manifest.json equals the sha256 of slots.json again (relabel last).

Co-Authored-By: Claude Sonnet 5.5 <[email protected]>

…ew (#1001)

* Retire the skill's Claude-in-GitHub rows and point them to local review

The claude-native-practice skill still selected the route that
docs/decisions/2026-10-10-retire-in-github-claude-review.md retired:
the github-action row (anthropics/claude-code-action at a pin) and the
Actions part of pr-review-pre-cue.

- slots.json, slots.md and the layer pages: `retired` joins the status
  vocabulary; github-action becomes retired with a dated block (record,
  correction #141, scope, the three workflows, the former default, route,
  alternatives, evidence and notes as a snapshot, the revisit triggers);
  pr-review-pre-cue becomes default (scoped) and lists the routes
  available now (local Claude reads, GPT reads, Codex's GitHub review,
  the required verdict-review-gate) apart from the open #953 worker.
- The scope is the owner's repositories (the command center's ruling of
  2026-10-10), not "this repository": a hermetic A/B showed a session in
  another repository concluding that the action was still fine there.
  The same ruling retires the managed Claude Code Review GitHub App.
- Stale texts the independent pre-read and the stricter check found: the
  release-tracking note (Dependabot as the action's tracker), the
  scheduled-tasks route (GitHub Actions schedules only for jobs that run
  no model) and its source mapping, the usage-cost-monitoring note
  (Actions under WIF), and the pre-cue source mapping (in Actions through
  #909). The pre-cue route no longer lists a read the record does not name.
- tests/test_native_practice_retired_routes.py: no current row selects a
  retired route; a retired route is named elsewhere only in a sentence or
  list item that says it is retired; every line of slots.json, slots.md,
  the layer pages and cross-client.md is read (duplicates kept, the source
  mappings of live slots included); only the dated snapshots are exempt.
  Positive controls are the rows as they stood; the mutation table is in
  the pull request description.
- The slot manifest of #923 is relabeled against the new slots.json
  bytes, so the exact-difference check of the provenance test runs again
  instead of skipping.
- docs/decisions/2026-10-09-claude-code-native-practice.md: a dated
  pointer to the retirement, as #981 did for the older records.

Source read for this change: anthropics/claude-code-action at
2dca132ff0e0c4094ce6048b422c6915a071210b, base-action/src/
parse-sdk-options.ts:283-284 (CLAUDE_CODE_ENTRYPOINT is set to
claude-code-github-action).

Co-Authored-By: Claude Sonnet 5.5 <[email protected]>

* Close the holes in the retirement guard and qualify the one-default rule for retired slots

Answers the Codex review in GitHub of #1001 at ba6ab3a (four P2s) and the co-op's GPT first read at the same head (three P2s, the first three below); each was reproduced before the change.
- Sentence units split at a full stop or a semicolon whatever follows, and at a line break: "This route was retired; the current default is claude-code-action." was one unit that said retired.
- The generic phrases "in Actions", "through Actions" and "under WIF" count as the retired route only with Claude context, judged per leaf string; the named workflows, "through #909" and the managed app count
  everywhere. The routes the decision retains (every bullet of its "What remains" section, a catalog job in GitHub Actions, a non-model gate under WIF) are accepted.
- The skill's entry point SKILL.md is scanned, and a test requires every file of the skill folder to be read by a check.
- The one-default rule names the retired exception in slots.json, in the slots.md copy and in step 3 of SKILL.md (a retired row has none and says what replaced the route).
- The ruled default text is pinned exactly (the read's measured boundary: the date changed on all three surfaces passed), and a check compares every row's Overturn when across slots.json, the index and the
  layer pages (one step wider than the review of #1005, where an edit left the index cell behind).
Each finding fails its own test when its fix is undone alone, and 43 single-edit mutations are caught; the old guard accepts the four bad scenarios of the read and rejects its two legitimate ones.

Co-Authored-By: Claude Sonnet 5.5 <[email protected]>

* Bind the manifest label to the shipped slots file and close the remaining holes of the retirement guard

Answers the CC's Claude first read of #1001 at 6e97620 (two P2s, six P3s) and Codex's second review in GitHub of the same head (two P2 threads); each was reproduced before the change.
- P2-1: the slot manifest was labeled against the slots.json of ba6ab3a, so the exact-difference check of the provenance test skipped at the head (OK, skipped=1). The label is the sha256 of the shipped file again,
  the check no longer skips, and a new test fails when a change to slots.json does not carry its label (the CC's ruling: a silent skip is this defect's class, so it fails loud).
- P2-2: a live source mapping (skills-library, primary_sources[5]) still ran `claude -p` "in a GitHub workflow ... plus a nightly cron"; the guard had no such form. It is reworded to say the run is retired,
  and "in a GitHub workflow" (also from, via, through, using, as) counts as the retired route when the sentence is about Claude.
- P3-1: the Claude Code GitHub Action, /install-github-app and an @claude mention are the retired route under other names (claude-code-security-review stays out: a documented rejection).
- P3-4: the Former default exemption of a layer page applies to a retired slot only.
- P3-5: a retired slot's own source mapping is read, and now says retired.
- P3-2: "e.g." and "i.e." no longer end a sentence, and wrapped lines of prose are one paragraph (the slots.md prose and SKILL.md were judged line by line).
- P3-6: the manifest prose lists the fields beyond the exported ones that differ from the snapshot.
- P3-7: the entry point directs ("a retired row has no default, so follow the replacement it names"), and the step-3 pin no longer fails with an IndexError on a renumbering.
Each correction fails its own test when it is undone alone, and the mutation table is extended.

Co-Authored-By: Claude Sonnet 5.5 <[email protected]>

* Judge retired-route prose one physical line at a time and close the guard's other gaps

Answers the command center's Claude micro of #1001 at ef3f3e7 (P2-1, P3-1 to P3-5) and Codex's four threads of the same head (:43, :63, :122, :211). Each case was reproduced before the change:
compare-guards-ef3f3e7b-vs-third-commit.txt runs 19 scenarios through the guard of ef3f3e7 and this one; the old guard is wrong on 13 (it accepts a retired route that is reintroduced), this one on none.
- P2-1 (correction #160): joining wrapped lines into one paragraph let "Retired 2026-10-10" followed by a line that says the current default is claude-code-action pass on four prose surfaces (SKILL.md, slots.md,
  cross-client.md, layer-page preambles), the scenario Codex raised at ba6ab3a line 41. Prose is judged one physical line at a time again, a line break ends a unit, and the message of every line-judged surface
  states the rule (RULE; the legend of slots.md states it too and a test pins the two to each other).
- P3-1 and Codex :43: three more names of the same route, all in the vendor's own text: the page's title "Claude Code GitHub Actions" and its "installs the Claude GitHub App"
  (code.claude.com/docs/en/github-actions.md, capture sha256 87b99aef, lines 5, 9, 33, 63-64), the mirror claude-code-base-action (anthropics/claude-code-action@2dca132ff0e0:
  .github/scripts/check_workflow_hardening.py:18, .github/workflows/sync-base-action.yml:1 and 43, CLAUDE.md:20) and the carrier named first ("Use the GitHub Action for Claude reviews", generic
  like "in Actions": it counts when the field or unit is about Claude). The families "in Actions" and "through Actions" take on, with, inside and via.
- Codex :122 and the command center's P3-3 and P3-5: a unit that negates the retirement outright ("not retired", "never retired", "no longer retired", "isn't retired", "unretired") no longer counts as saying
  retired. The other half of that scenario, "was retired, but is now current", is the boundary the command center ruled (the guard is a tripwire for stale text; review reads what it accepts); the module names it.
- Codex :211 and P3-3: a layer-page section with a label twice (a second Route or Overturn when bullet) is a violation, because the page-to-slots.json comparison keeps the first bullet only.
- Codex :63 and P3-4: the target of every slots.md row is checked against <page>#<id>, from the layer of the slot.
- The description line of SKILL.md names the topic that invokes the skill; it is pinned by sha256 to the text the routing was measured with and read as a topic, not as a route.
- Scheduled-tasks (Codex on #1005, scheduling-supervision.md:9, and the command center's ruling): the route no longer bans every model-backed job. The retirement record
  (docs/decisions/2026-10-10-retire-in-github-claude-review.md:5-6, :43) retires Claude review inside GitHub and keeps Codex's GitHub review, so the route says GitHub Actions schedules repository jobs and a
  scheduled Claude task runs locally; a test pins the wording on slots.json, the layer page and the alternatives.
The label of claude-native-slot-manifest.json equals the sha256 of the shipped slots.json again (relabel last); test_the_label_names_the_shipped_slots_file and test_no_other_exported_field_differs
(no skip) are unchanged from ef3f3e7. 72 of 72 single-edit mutants of the guard and its data are caught and 5 controls pass (mutations-v4-forward-commit-3.md).

Co-Authored-By: Claude Sonnet 5.5 <[email protected]>

* Align the manifest's scheduled-tasks note, state the rule in every explaining unit, split units after a closing delimiter and pin the scope of the name patterns

The final fix round of #1001 under the command center's convergence ruling: its Claude micro at efaa077 (blockers B1 to B3, P3-2) and its adjudication of 2026-10-11T13:50Z of Codex's third review. Each item was reproduced on that
head before the change.
- claude-native-slot-manifest.json:47 (Codex P2, B1, a text of this PR that is false at the head): the note on the differences from the snapshot still said the route of scheduled-tasks was narrowed "to jobs that run no model",
  while the shipped route says that GitHub Actions schedules repository jobs, that Claude runs inside GitHub are retired (2026-10-10, docs/decisions/2026-10-10-retire-in-github-claude-review.md:5-6, :43) and that a scheduled
  Claude task runs locally. The note says so now, and test_the_scheduled_tasks_route_retires_the_claude_routes_and_not_every_model_backed_job reads the manifest as well as the route (fails with the note undone: M79).
- B2: the module claims that the message of every line-judged surface states the rule, and it did not for the labelled bullets of a layer page or for the Overturn when cell of an index row. Every explaining unit states it now
  (a selecting field names no unit and states nothing); the three expected messages carry the rule and test_every_explaining_unit_of_a_layer_page_and_of_the_index_states_the_rule_and_a_selecting_field_does_not pins all three
  kinds (mutants M83 to M85).
- tests/test_native_practice_retired_routes.py:83 (Codex P2, P3-2): a closing quote, parenthesis or bracket between the full stop and the white space kept `The route was "retired." The current default is claude-code-action.` in one
  unit that said retired. The sentence pattern takes closing delimiters (the curly quotes are built with chr(), so the file stays ASCII); test_a_closing_quote_parenthesis_or_bracket_after_the_full_stop_does_not_join_two_claims fails
  without it (M78, M82) and keeps `claude-code-action was retired (see the record.) No row selects it.` accepted (control C6).
- :48 (Codex P2, workflow-first phrasing) is a follow-up and the class closes (rule 4 of the convergence ruling): the module states that the guard pins the named retired artifacts and the retired slot rows, that generic phrasing of a
  Claude run in a GitHub workflow cannot be enumerated and is out of its scope (the retirement record governs it), and test_generic_workflow_phrasing_is_out_of_scope_and_the_module_says_so pins the boundary. No pattern is added.
- B3 is the description: the two earlier command blocks name the 37 modules they ran (the 46 of the first block without the nine tests.test_local_pages_* modules).
80 single-edit mutants of the guard and its data are caught and 6 controls pass (mutations_retire_v6.py).

Co-Authored-By: Claude Sonnet 5.5 <[email protected]>

* Close the sentence-splitter class by documentation, pin the module's ASCII and the scope list

The command center's confirmation micro of #1001 at bdf60fc (research-cc-role-20261008/pr1001-claude-micro-bdf60fc6.md: blockers 1 and 2, P3s) and its correction #163 of 2026-10-11T16:18Z on Codex's three threads of 15:39:21Z.
No code of the guard changes in this commit.
- Blocker 2: the module said that it stays ASCII and held one U+2026 in a comment (added by efaa077). It is three full stops now, and test_the_source_of_this_module_is_printable_ascii pins the claim; it fails on the module
  of bdf60fc (['U+2026'] != []).
- Blocker 1 (the pull request description, and this message corrects the message of bdf60fc): the earlier statement that the 37 modules ran "without its nine tests.test_local_pages_* modules" was false. They are the first 37
  modules of the 46-module list, and the nine left out are its last nine (tests.test_local_pages_fleet_execution to tests.test_local_pages_workstation, 144 tests, run and counted); 29 of the 46 are local-pages modules.
- Codex :92 (an exclamation mark or a question mark does not end a unit) is not patched: the sentence splitter has drawn findings in five consecutive rounds (the lowercase clause, "e.g.", the joined lines, the closers,
  the exclamation mark and the question mark) and was patched four times, so rule 4 of the convergence ruling closes the class the way line 48 closed. The module docstring states exactly what the guard accepts as a unit end
  and that review reads everything else; test_the_unit_ends_and_the_scope_of_the_name_patterns_are_documented pins the sentence and the behavior, and fails on the module of bdf60fc, where the sentence is missing.
  A pinned upstream sentence segmenter is a research-queue item (rule 8, layer instructions-skills).
- Codex :51 (a carrier-first name of the managed Code Review app) is confirmed within the closed class: the same scope paragraph says that reordered phrasings of a pinned artifact are out of the guard's scope.
- P3-1: the scope list named 11 of the 17 patterns; it names all 17 now, and test_the_scope_of_the_name_patterns_names_every_pattern compares the list itself with PATTERNS (it fails on the module of bdf60fc).
  P3-2: the right single quotation mark closer is tested (its mutant survived). P3-3: the comment at the sentence pattern says what the code does.
- Codex :170 (retired.on is checked for its shape only) is a follow-up under rule 5: the shipped value is the date of the record, so 2099-12-31 is a mutant and not a reproduced failure of the shipped state.
88 single-edit mutants of the guard and its data are caught and 6 controls pass (mutations_retire_v9.py).

Co-Authored-By: Claude Sonnet 5.5 <[email protected]>

---------

Co-authored-by: Claude Sonnet 5.5 <[email protected]>

, #1005 and #986 (#1018)

The row of adoption/skills/manifest.json and this repository's source pin in catalogs/landscape/skills-lifecycle.json follow the skill to c59655c, the commit of main that holds its final state after the retirement of the Claude-in-GitHub rows (#1001), the paired-run results (#1005) and the cross-client page (#986).

Only the measured pin fields change, from the commit's own bytes: url, ref, tree_sha (3a485d2... to f08d0d2...), skill_md_sha256 and skill_md_bytes (2,224 to 2,290), and the manifest's checked_at. The description stays 349 characters, so the listing sums and the ceiling are unchanged. A dated amendment in the row's decision record names the move; the three files are registered in manifests/evidence.json. The skill is not installed by this change.

Co-authored-by: Claude Sonnet 5.5 <[email protected]>

…oint the Cloud routines locator

Fold 3 on 450608c, at the command center's request after its read of fold 2.

Regression tests: tests/test_native_practice_owed_check_corrections.py, in the style of
tests/test_native_practice_doc_corrections.py, pins the two corrections fold 2 made from the
command center's owed-checks report (command-center/claude-native/OWED-CHECKS-2.1.302-20261017.md).

- F1.1, settings-permissions (SettingsPermissionsRmPromptTests): [email protected] L245 (2.1.299)
  fixes rm prompts on shell -c scripts; it does not change rule evaluation, which L373 and L280
  (2.1.299) do. The slots.json row cites L245 only as "L245 (2.1.299, rm prompts on shell `-c`
  scripts)", its rule-evaluation sentence cites L373 and L280 without L245, and the native-clients.md
  section carries the corrected Evidence and Notes lines and neither of the two wrong phrases.
- F2.1, subagent-definitions (SubagentDefinitionsDriftNoteTests): the note has none of the three
  stale drift phrases and states that the copies match; the workers.md section mirrors it. The
  count of definitions in the corrected sentence equals the files in adoption/agents/claude and in
  .claude/agents, which are byte-identical. Its locator into
  docs/decisions/2026-10-13-claude-agent-drift-check.md names the line that records the old note as
  stale. The host folder ~/.claude/agents is not readable from the repository.

Fail first: with only the test file overlaid on 728141f, each of the 5 tests fails with an
assertion (FAILED (failures=1) each; the module FAILED (failures=5)). At 450608c with the file added,
each passes (module OK).

Locators: /usr/bin/grep -rn over the whole repository, not capped, for the basename of slots.json and
of each of the 12 reference pages this PR changed, followed by :<digits>, finds 103 locators in text
files and one compiled .pyc that git ignores. Only slots.json moved lines in this PR: every hunk in the
pages replaces lines one for one. One live locator into it was stale. The gap-resolution record at
line 174 cited reference/slots.json:1625 for the "Cloud routines" alternative of scheduled-tasks: the
right line at 8c95a50, the revision its search names, and on main. The text is at line 1649 now.
Because the sentence reports that search, the locator keeps :1625 at 8c95a50 and adds "current
locator `:1649`", the form the ecosystem-roadmap record uses (2026-09-28-ecosystem-roadmap.md:357).
Line 1649 is "Cloud routines" in the alternatives of scheduled-tasks. The edit stays on its line, so
the record keeps its 245 lines and its lines 179-181 keep the numbers that the usage-cost-monitoring
note and tests/test_native_practice_ledger_obs_sched.py:85 cite. Its slots.json:3 locator
(client_version) still resolves. The other hits stay as they are:
- 64 sit in evidence/artifacts/claude-native-practice-20261009/primary-source-reading.json. They
  resolve at a30c218, and tests.test_native_practice_snapshot_provenance checks them there. The
  record that describes them (2026-10-09-claude-code-native-practice.md:344) gives slots.md:35 as
  an example and says it resolves at that revision.
- The test comments that cite a review finding of #1001 or #1005 are history.
- The rest name other files whose names end the same way: 2026-09-30-omniroute-runtime-workers.md,
  2026-09-28-openhands-resolver-isolation.md, a docs/token-efficiency.md that this repository has
  never tracked (agent-lab's), and the dated slots.json snapshots under evidence/artifacts.

Evidence rows: scripts/host_receipts.register_file added rows for the two changed files; the
gap-resolution record had none. scripts/evidence_manifest.py --write left the file unchanged.

Checks, run with fold 3 in the tree before this commit: python3 scripts/validate.py rc 0;
python3 scripts/evidence_manifest.py --check rc 0. The test modules that name a changed path (grep
-rl over tests/ for each changed file's basename, claude-native-practice and slot-manifest, not
capped) are now 74. Two are new to the list: the new module, and tests.test_upstream_surface_watch,
which reads the gap-resolution record. unittest counted 4,220 tests. 64 modules pass, among them the
new module (5 tests), tests.test_native_practice_snapshot_provenance (9) and
tests.test_native_practice_fork_toggle (2). The other 10 fail because of this host, not the fold. CI
passed every validate shard at 450608c.
- These fail the same way with the tracked tree at 450608c: tests.test_upstream_surface_watch (4
  CachedArtifactTests, which read this host's surface-watch cache; its record checks pass);
  tests.test_local_pages_overview (the 2 ComposerTests); tests.test_local_pages_architecture_read_boundary
  and tests.test_local_pages_architecture_sources (26 errors each, no /usr/bin/zstd); and one test
  of tests.test_token_e2e_grader (git clone --local across filesystems).
- tests.test_component_matrix, tests.test_final_admission, tests.test_final_stack and
  tests.test_gap_closeouts need jsonschema, which this host's python3 lacks. With fold 3 they pass
  in a Python 3.12 venv built from .github/requirements-ci.txt, as do tests.test_lane_packets and
  tests.test_paired_runs_results.
- tests.test_saturation_ledger lost one V2PendingCliTests test, with "native publication scanner
  unavailable", in each of two runs made while three test streams shared the host. Run alone it
  passes under python3 (OK, skipped=1) and in the venv (OK).

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

…ude Code 2.1.302 (#1234)

* docs(claude-native-practice): bring the practice pages current to Claude Code 2.1.302

Source for every edit: anthropics/claude-code CHANGELOG.md at v2.1.302 (fetched 2026-10-16T22:15Z
with `gh api repos/anthropics/claude-code/contents/CHANGELOG.md -H "Accept: application/vnd.github.raw"`,
sha256 prefix 57fde96b3318543c), cited by line as [email protected] L<n>; each cited line was
checked against the fetched file. The fetched file has no 2.1.297 or 2.1.298 section.

- Headers: slots.md, slots.json and the eleven layer pages move to client 2.1.302, checked
  2026-10-16; cross-client.md is left alone (hash-tested against cross-client-lines.json).
- native-clients.md: settings-permissions records the 2.1.299 rule-evaluation change (L373,
  L280, L245, L270) with an owed audit of prefix-matched mcp__ entries; checkpoints-rewind
  records that 2.1.300 transcripts carry each request (L186) and owes a re-measured footprint;
  model-effort-routing keeps CLAUDE_CODE_WORKFLOW_SUBAGENT_MODEL unset too (L411, L224, L371).
- isolation.md: sandbox cites the 2.1.299 and 2.1.300 sandbox changes and has M1 run on
  2.1.302 (L249, L190); the M49 worktree probe moves to 2.1.302 (L153, L148, L77, L324, L157).
- token-efficiency.md: context-compaction records idle compaction, the /config row and
  subagent autoCompactWindow at their defaults (L217-L218, L6, L410, L110); output-compression
  cites Read allow_large, CLAUDE_CODE_DISABLE_FILE_MENTIONS and updatedInput validation (L414,
  L216, L189) and owes an rtk-hook check.
- workers.md: subagent-definitions records the tools: list change and fork subagents on in -p
  (L374, L410, L35, L36); background-workflows routes the 2.1.300 resume through the Workflow
  tool's approval, to be re-observed (L185, L365, L81, L411).
- mcp-surfaces.md: the 4,096-character up-front limit (L469, L377, L142).
- agent-sdks.md: headless-sdk routes the 2.1.302 -p start mode and fork subagents (L34, L35,
  L36, L291).
- observation-inference.md: usage-cost-monitoring lists the new OTel fields for the collector
  allowlists (L126, L130, L75, L76, L18, L467, L186).

slots.json mirrors every page edit (Evidence, Notes and Route match the pages exactly); the
slot manifest's sha256_when_labeled follows slots.json and differs_from_this_snapshot_in gains
the newly differing evidence and route slots; manifests/evidence.json re-registers the changed
files.

Checks: python3 scripts/validate.py rc 0; scripts/evidence_manifest.py --check rc 0; the 13
native-practice test modules ran 347 tests, with the same two ComposerTests failures as on the
base 90471d6 (the overview composer's layer pages hit a ModuleNotFoundError in this
environment) and nothing else.

Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01QcCViiYWLxs4igpJR3yws6

* Correct the Workflow overturn citation and four wording slips in the 2.1.302 practice pages

The CC's Claude-family read of 0b361d4 found one false citation: workers.md and slots.json gave
[email protected] L411 as a Workflow resume or limit change, but L411 (2.1.296) only adds
CLAUDE_CODE_WORKFLOW_SUBAGENT_MODEL. The overturn still fired through L365 (2.1.299, the nesting
limit) and L185 (2.1.300, resume through the Workflow tool's approval), so the parenthetical now cites
those two.

Wording now follows the cited lines: the /config Idle compaction row turns idle compaction off or back
on (L6); the 4,096-character figure is the default limit (L469); L34's comma is restored. The
settings-permissions (L373) and headless-sdk (L34, L35) notes now say that their overturn fired, as the
other updated slots do.

slots.json mirrors every edit; the slot manifest's sha256_when_labeled and the seven changed files'
manifests/evidence.json rows were refreshed with scripts/host_receipts.py register_file.
Checks: python3 scripts/validate.py rc 0; python3 scripts/evidence_manifest.py --check rc 0; the 13
test modules that read these pages ran 347 tests with the same two ComposerTests failures as on the
untouched base 90471d6 and nothing else.

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

* docs(claude-native-practice): the fork toggle is fork-only (GPT read P1 at 887f929)

The headless-sdk row (agent-sdks.md:9 and its slots.json mirror) offered CLAUDE_CODE_FORK_SUBAGENT=0
as an alternative to keeping Agent out of --tools for a worker that must not fan out. [email protected]
L35 says the variable turns off fork subagents (subagent_type "fork") only, so a named subagent can still
fan out. The row now keeps Agent out of --tools as the control and describes the toggle as fork-only.

tests/test_native_practice_fork_toggle.py pins the corrected text (it fails on 887f929 with only the
test overlaid) and the page's mirror of the row. The two files' evidence rows are refreshed through
scripts/host_receipts.py register_file.

Source: anthropics/claude-code CHANGELOG.md at tag v2.1.302, line 35.

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

* docs(claude-native-practice): record the 2.1.302 owed checks and relabel the slot manifest

A fold on 728141f that applies the PR #1234 diffs of the command center's owed-checks report
(command-center/claude-native/OWED-CHECKS-2.1.302-20261017.md, written 2026-10-17T02:43:07Z). Each
page edit is mirrored in its slots.json row. Rule source: anthropics/claude-code CHANGELOG.md at tag
v2.1.302, cited as [email protected] L<n> with the release each line sits under.

- native-clients.md:53-54, settings-permissions (report F1.1, F1.2): L245 (2.1.299) fixes rm
  prompts on shell -c scripts; it does not change rule evaluation. The evidence and the note now
  cite L373 and L280 for rule evaluation and L245 for the built-in prompt. The owed audit of mcp__
  rules against L373, L280 and L270 (2.1.299) replaces the owed sentence: no rule relied on prefix
  matching, so none changes.
- workers.md:13, subagent-definitions (F2.1, F2.2): the note said four host copies differ from
  adoption/agents/claude and four host-only definitions exist. diff -rq of adoption/agents/claude
  and of .claude/agents against ~/.claude/agents exits 0, 11 files each, so the claim was false at
  728141f. The owed audit of the tools: lists against L374 (2.1.299) is recorded: 25 whole-name
  grants from six servers, and no list changes.
- token-efficiency.md:50-51, output-compression (F3.1): the owed check of the rtk hook against the
  updatedInput validation of L189 (2.1.300) is recorded (rtk-ai/[email protected]
  src/hooks/hook_cmd.rs:648-651 replaces only command), and L916 (2.1.290) joins the evidence.
- mcp-surfaces.md:12-13, mcp-config-tool-search (F4.1): the owed measurement of the up-front
  context at the 4,096-character default of L469 (2.1.296) is recorded, beside L377 (2.1.299);
  L602 (2.1.295) and L1926 (2.1.280) join the evidence.
- observation-inference.md:13, usage-cost-monitoring (F5.2): the note said each new field is to be
  added to the collector's allowlists. It now says neither collector profile keeps the five event
  fields of L126 (2.1.300), L75 and L76 (2.1.301) and L18 (2.1.302), and that the span attributes
  of L130 (2.1.300) need no allowlist entry because the templates export no Claude traces.

The six diffs were applied with git apply after git apply --check (exit 0 each). Each replaced
evidence or notes line occurs once at 728141f; the "]," lines that the slots.json hunks also rewrite
are placed by their context. Each result has the sha256 the report proposes.

Label fix: current_decision_source.sha256_when_labeled in claude-native-slot-manifest.json now names
the sha256 of this commit's slots.json, taken from sha256sum. This corrects 728141f, which changed
slots.json without moving the label, so tests.test_native_practice_snapshot_provenance
(test_the_label_names_the_shipped_slots_file) fails there; the CC had capped its test selection for
that commit. differs_from_this_snapshot_in is unchanged. Of the fields the manifest exports, the fold
changes only evidence, in settings-permissions, output-compression and mcp-config-tool-search, and
the manifest already lists all three. Notes are not exported.

The manifests/evidence.json rows of the seven changed files were refreshed with
scripts/host_receipts.register_file; scripts/evidence_manifest.py --write left the file unchanged.

Checks, run with the fold applied before this commit: python3 scripts/validate.py rc 0;
python3 scripts/evidence_manifest.py --check rc 0. The test modules that name a changed path (grep
-rl over tests/ for each changed file's basename, claude-native-practice and slot-manifest, not
capped) are 72; each ran on its own (unittest counted 4,099 tests). 64 pass, among them
tests.test_native_practice_snapshot_provenance (9 tests) and tests.test_native_practice_fork_toggle
(2). The other 8 fail from this host's environment, not the fold; at 728141f the only test that
failed in CI was the label test. Four fail the same way on the untouched 728141f:
tests.test_local_pages_overview (the 2 ComposerTests failures);
tests.test_local_pages_architecture_read_boundary and tests.test_local_pages_architecture_sources
(26 errors each, no /usr/bin/zstd); and one test of tests.test_token_e2e_grader (git clone --local
cannot hardlink across filesystems). The other four need jsonschema, which this host's python3
lacks: tests.test_component_matrix, tests.test_final_admission, tests.test_final_stack and
tests.test_gap_closeouts. With the fold they pass in a Python 3.12 venv built from
.github/requirements-ci.txt, as do tests.test_lane_packets, tests.test_paired_runs_results and
tests.test_saturation_ledger, whose jsonschema tests had skipped.

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

* test(claude-native-practice): pin the F1.1 and F2.1 corrections; re-point the Cloud routines locator

Fold 3 on 450608c, at the command center's request after its read of fold 2.

Regression tests: tests/test_native_practice_owed_check_corrections.py, in the style of
tests/test_native_practice_doc_corrections.py, pins the two corrections fold 2 made from the
command center's owed-checks report (command-center/claude-native/OWED-CHECKS-2.1.302-20261017.md).

- F1.1, settings-permissions (SettingsPermissionsRmPromptTests): [email protected] L245 (2.1.299)
  fixes rm prompts on shell -c scripts; it does not change rule evaluation, which L373 and L280
  (2.1.299) do. The slots.json row cites L245 only as "L245 (2.1.299, rm prompts on shell `-c`
  scripts)", its rule-evaluation sentence cites L373 and L280 without L245, and the native-clients.md
  section carries the corrected Evidence and Notes lines and neither of the two wrong phrases.
- F2.1, subagent-definitions (SubagentDefinitionsDriftNoteTests): the note has none of the three
  stale drift phrases and states that the copies match; the workers.md section mirrors it. The
  count of definitions in the corrected sentence equals the files in adoption/agents/claude and in
  .claude/agents, which are byte-identical. Its locator into
  docs/decisions/2026-10-13-claude-agent-drift-check.md names the line that records the old note as
  stale. The host folder ~/.claude/agents is not readable from the repository.

Fail first: with only the test file overlaid on 728141f, each of the 5 tests fails with an
assertion (FAILED (failures=1) each; the module FAILED (failures=5)). At 450608c with the file added,
each passes (module OK).

Locators: /usr/bin/grep -rn over the whole repository, not capped, for the basename of slots.json and
of each of the 12 reference pages this PR changed, followed by :<digits>, finds 103 locators in text
files and one compiled .pyc that git ignores. Only slots.json moved lines in this PR: every hunk in the
pages replaces lines one for one. One live locator into it was stale. The gap-resolution record at
line 174 cited reference/slots.json:1625 for the "Cloud routines" alternative of scheduled-tasks: the
right line at 8c95a50, the revision its search names, and on main. The text is at line 1649 now.
Because the sentence reports that search, the locator keeps :1625 at 8c95a50 and adds "current
locator `:1649`", the form the ecosystem-roadmap record uses (2026-09-28-ecosystem-roadmap.md:357).
Line 1649 is "Cloud routines" in the alternatives of scheduled-tasks. The edit stays on its line, so
the record keeps its 245 lines and its lines 179-181 keep the numbers that the usage-cost-monitoring
note and tests/test_native_practice_ledger_obs_sched.py:85 cite. Its slots.json:3 locator
(client_version) still resolves. The other hits stay as they are:
- 64 sit in evidence/artifacts/claude-native-practice-20261009/primary-source-reading.json. They
  resolve at a30c218, and tests.test_native_practice_snapshot_provenance checks them there. The
  record that describes them (2026-10-09-claude-code-native-practice.md:344) gives slots.md:35 as
  an example and says it resolves at that revision.
- The test comments that cite a review finding of #1001 or #1005 are history.
- The rest name other files whose names end the same way: 2026-09-30-omniroute-runtime-workers.md,
  2026-09-28-openhands-resolver-isolation.md, a docs/token-efficiency.md that this repository has
  never tracked (agent-lab's), and the dated slots.json snapshots under evidence/artifacts.

Evidence rows: scripts/host_receipts.register_file added rows for the two changed files; the
gap-resolution record had none. scripts/evidence_manifest.py --write left the file unchanged.

Checks, run with fold 3 in the tree before this commit: python3 scripts/validate.py rc 0;
python3 scripts/evidence_manifest.py --check rc 0. The test modules that name a changed path (grep
-rl over tests/ for each changed file's basename, claude-native-practice and slot-manifest, not
capped) are now 74. Two are new to the list: the new module, and tests.test_upstream_surface_watch,
which reads the gap-resolution record. unittest counted 4,220 tests. 64 modules pass, among them the
new module (5 tests), tests.test_native_practice_snapshot_provenance (9) and
tests.test_native_practice_fork_toggle (2). The other 10 fail because of this host, not the fold. CI
passed every validate shard at 450608c.
- These fail the same way with the tracked tree at 450608c: tests.test_upstream_surface_watch (4
  CachedArtifactTests, which read this host's surface-watch cache; its record checks pass);
  tests.test_local_pages_overview (the 2 ComposerTests); tests.test_local_pages_architecture_read_boundary
  and tests.test_local_pages_architecture_sources (26 errors each, no /usr/bin/zstd); and one test
  of tests.test_token_e2e_grader (git clone --local across filesystems).
- tests.test_component_matrix, tests.test_final_admission, tests.test_final_stack and
  tests.test_gap_closeouts need jsonschema, which this host's python3 lacks. With fold 3 they pass
  in a Python 3.12 venv built from .github/requirements-ci.txt, as do tests.test_lane_packets and
  tests.test_paired_runs_results.
- tests.test_saturation_ledger lost one V2PendingCliTests test, with "native publication scanner
  unavailable", in each of two runs made while three test streams shared the host. Run alone it
  passes under python3 (OK, skipped=1) and in the venv (OK).

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

* docs(claude-native-practice): close the four read findings on 1af2052, each pinned by a test

Fold 4 on 1af2052. Both reads of 1af2052 are done; this commit fixes the four findings the command
center listed from them. Each correction has a regression test that fails, with an AssertionError,
on the text it replaces.

1. P2: F5.2 had no regression test (observation-inference.md:13, slots.json:2107; F5.2 of the
   command center's owed-checks report, OWED-CHECKS-2.1.302-20261017.md). The new class
   UsageCostMonitoringSpanAttributeTests in tests/test_native_practice_owed_check_corrections.py
   checks that the row and the page:
   - lack "is to be added to the collector's tool-name and privacy allowlists";
   - carry "The span attributes need no allowlist entry" and "To be tested, with Claude traces on".
   It also checks that the test line the note cites (tests/test_observability_tool_names.py:178)
   still asserts OTEL_TRACES_EXPORTER "none". Run with only the test file overlaid on 728141f,
   each of its 3 tests fails. They pass at 1af2052, where fold 2 made the correction, and at this
   commit.

2. P3: the L916 sentence ([email protected] L916, 2.1.290) left out F3.2 (token-efficiency.md:51,
   slots.json:785). One sentence is added to the output-compression note, written to the row and
   the page from one string. It says three things:
   - rtk 0.51.0 rewrites every npx command (rtk-ai/[email protected] src/discover/rules.rs:120);
   - none of the 19 `npx *skills` deny rules in the user settings has an `rtk npx` twin;
   - PR #1238 adds the 19 twins to the template.
   The counts come from an exact-key read of permissions.deny at 2026-10-17T04:41:38Z (19 rules, 0
   twins). The repository's own matcher (ProfileTemplateSettingsTests.bash_rule_matches) finds no
   deny rule for six rtk spellings of the skills CLI there, and one for the plain spelling. PR
   #1238's template holds 19 rules and 19 twins at its head 6bf2eb5.
   OutputCompressionRtkNpxGapTests checks that the row and the page carry the sentence after the
   force-push one; both checks fail on 728141f and on 1af2052. It also checks that the template
   still holds 19 `Bash(npx *skills` rules, which stays true after #1238. The slot manifest's
   sha256_when_labeled is the sha256 of this commit's slots.json, taken from sha256sum.
   differs_from_this_snapshot_in is unchanged because notes are not exported
   (test_no_other_exported_field_differs passes).

3. P3: fold 3's "current locator `:1649`" in docs/decisions/2026-10-09-claude-code-gap-resolution.md:174
   would go stale. As the command center ruled, the record is back to main's bytes (git checkout
   origin/main; git diff --quiet origin/main exits 0), because its citation is already pinned to the
   search revision 8c95a50. The evidence row fold 3 added for it is removed; the record has none on
   main. DispositionCitationTests in tests/test_upstream_surface_watch.py already checks this record
   against references that move. It gains two tests:
   - the routines sentence keeps `reference/slots.json:1625` and has no "current locator" (fails on
     1af2052; passes on 728141f and here);
   - line 1625 is "Cloud routines" at 8c95a50 (skipped when the history lacks that revision).

4. P3: the slot manifest's note (claude-native-slot-manifest.json:69) named six slots whose notes
   differ from the snapshot when 21 do. It also named three of the five top-level keys that differ:
   checked_at and client_version differ too, since this PR's first commit. The sentence now names
   fields, not slots: "Slot fields this export does not carry also differ (notes, primary_sources,
   primary_sources_total and github-action's retired block), and so do the top-level checked_at,
   client_version, cross_client, selection_rule and status_values." A new test in
   tests.test_native_practice_snapshot_provenance's BaselineHistoryTests,
   test_the_note_names_what_differs_beyond_the_exported_fields, builds that sentence from the
   baseline and current slots files. On failure it prints the sentence to write; it fails on
   728141f and on 1af2052. The note's other counts hold at this commit: the 2.1.302 update changed
   the evidence of eleven slots and four routes.

Evidence rows: manifests/evidence.json is 450608c's copy, with rows registered through
scripts/host_receipts.register_file for the six files this fold changes and for the fold-3 test.
Against 1af2052 it drops the record's row and refreshes those six. scripts/evidence_manifest.py
--write left it unchanged.

Checks, run with fold 4 in the tree before this commit:
- python3 scripts/validate.py rc 0; python3 scripts/evidence_manifest.py --check rc 0.
- The five acceptance modules, each by name, all OK:
  tests.test_native_practice_ledger_obs_sched (8 tests), tests.test_native_practice_ledger_rows_1 (3),
  tests.test_native_practice_ledger_rows_tokeff_new (7), tests.test_native_practice_snapshot_provenance
  (10) and tests.test_skills_manifest (38).
- The test selection is not capped: grep -rl over tests/ for the basename of every path this PR or
  this fold changes, plus claude-native-practice and slot-manifest, gives 76 modules.
  - 75 ran one at a time and unittest counted 4,324 tests; 66 modules pass.
  - tests.test_runtime_worker_openhands_push_gate, which reads the committed HEAD, runs after this
    commit.
  - The other 9 fail because of this host, not the fold; CI passed every validate shard at 450608c.
- These fail the same way with the tracked tree at 450608c, as measured for fold 3:
  - tests.test_upstream_surface_watch: the same 4 CachedArtifactTests, which read this host's
    surface-watch cache. Its new record tests pass.
  - tests.test_local_pages_overview: the 2 ComposerTests.
  - tests.test_local_pages_architecture_read_boundary and tests.test_local_pages_architecture_sources:
    26 errors each, no /usr/bin/zstd.
  - one test of tests.test_token_e2e_grader: git clone --local across filesystems.
- tests.test_component_matrix, tests.test_final_admission, tests.test_final_stack and
  tests.test_gap_closeouts need jsonschema, which this host's python3 lacks. With fold 4 they pass in
  a Python 3.12 venv built from .github/requirements-ci.txt, as do tests.test_lane_packets and
  tests.test_paired_runs_results.

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

* fix(claude-native-practice): keep the gap-resolution record's evidence row; scope the rtk npx sentence

Fold 5 on 609d3e1. Both reads of 609d3e1 requested changes, and the higher rating governs. This
commit fixes the two findings the command center listed.

1. P1 (Muse): the command center corrects ruling 3 of fold 4.
   - docs/decisions/2026-10-09-claude-code-gap-resolution.md stays at main's bytes, but its
     manifests/evidence.json row must not be dropped.
   - Fold 3 (1af2052) registered the record. The PR-registration check requires a row for every
     path a PR commit registered (scripts/validate.py:776 and :786-788). CI runs it with the PR
     base and the previous heads (.github/workflows/validate.yml:173-178). Only a deleted file may
     lose its entry (docs/lanes.md:119-127; the fixtures at tests/test_registry_rows.py:68-84 and
     :112).
   - At 609d3e1, python3 scripts/validate.py --registration-base 1af2052 and
     --registration-base f00d478 each exited 1 with "docs/decisions/2026-10-09-claude-code-gap-resolution.md:
     required registration row missing", and CI's validate-shard-0 failed.
   - scripts/host_receipts.register_file restores the row with the reverted file's current sha256
     and bytes (15899 bytes; the file is unchanged against main).

2. P3 (Claude and Muse): "Every npx command is rewritten too" is false at rtk v0.51.0
   (rtk-ai/rtk@e001f773) for an npx command that a deny rule matches.
   - rtk's Claude hook checks the original command against the Bash deny rules first
     (src/hooks/permissions.rs:84-90) and skips the rewrite on a match
     (src/hooks/hook_cmd.rs:688-691); only the remaining npx commands reach the npx rule
     (src/discover/rules.rs:120).
   - The sentence now limits the rewrite to npx commands that no deny rule matches, and says that
     the twins in #1238 cover the `rtk npx` spelling, which rtk's own deny check does not cover.
   - The edit script reads the sentence from the test's CLAUSE and writes it to
     token-efficiency.md:51 and slots.json:785. The module docstring is scoped the same way.
   - With only the test file overlaid on 609d3e1,
     OutputCompressionRtkNpxGapTests.test_the_row_names_the_gap_and_its_fix and
     test_the_page_carries_the_row each fail with an AssertionError; both pass here.
   - The slot manifest's sha256_when_labeled is the sha256 of this commit's slots.json, taken from
     sha256sum. differs_from_this_snapshot_in is unchanged, because notes are not exported.

Evidence rows: register_file refreshed the rows of slots.json, token-efficiency.md, the slot
manifest and the test module, and restored the record's. scripts/evidence_manifest.py --write left
the file unchanged.

Checks, run with fold 5 in the tree before this commit:
- python3 scripts/validate.py rc 0; python3 scripts/evidence_manifest.py --check rc 0. The
  registration checks read the committed head, so validate.py with --registration-base origin/main
  and with --registration-base 1af2052 runs after this commit; the PR body records the exits.
- The five acceptance modules all pass: tests.test_native_practice_ledger_obs_sched (8 tests),
  tests.test_native_practice_ledger_rows_1 (3), tests.test_native_practice_ledger_rows_tokeff_new (7),
  tests.test_native_practice_snapshot_provenance (10) and tests.test_skills_manifest (38).
- The uncapped selection is the same 76 modules as for fold 4. 75 ran before the commit, in two
  streams, with tests.test_saturation_ledger alone afterwards; unittest counted 4,324 tests.
  - 66 pass, among them tests.test_registry_rows, tests.test_validate and tests.test_host_receipts.
  - tests.test_runtime_worker_openhands_push_gate runs on the committed head.
  - The other 9 are the environment failures recorded for fold 4. Each fails the same way at the
    previous head, or passes in the CI-requirements venv: four CachedArtifactTests, two
    ComposerTests, no /usr/bin/zstd (two modules), one cross-filesystem clone, and four modules
    that need jsonschema.

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

---------

Co-authored-by: Claude Opus 5.5 <[email protected]>