skills(ui): split the entry into SKILL.md + rules/ under #14296 item 1 (package net −58 tokens, −0.17%)

[os-litant]

Part of #14298

Ruling: "#14296 item 1 = A, comment 5507419465". Condition (a) was read at programme level by "#14568 = A, comment 5511939006" (director seat, verbatim reply 「同意」): the split PR itself must land token-neutral or negative, and the package arithmetic is carried by the programme.

The three conditions, with numbers

(a) Token-neutral or negative — measured -58.node scripts/check-skills-token-ratchet.mjs before and after, at head d72e3dcc:

	before	after
SKILL.md	22,856	3,815
rules/*.md (5 new files)	—	18,983
generated (contracts/react-blocks.contract.json, references/react-blocks.md, references/_index.md)	9,710	9,710
evals/	1,391	1,391
package	33,957	33,899

Net -58 (-0.17%). Gross deletions -248, additions +190 (five file titles, one ## Detailed Rules heading, five one-line pointers). Nothing in this PR is positive.

Programme account for the package: round 1 #14338-2,585 + this split -58 + seam card #14559-5,352 = -7,995 against the pre-programme 36,542 = -21.9%.

(b) Ceilings. Five rows added, each pinned AT its landed count so every new file starts with exactly zero headroom; the entry row re-locked to its landed count. No other row changed.

row	ceiling
skills/objectstack-ui/SKILL.md	25,445 → 3,815 (re-lock)
skills/objectstack-ui/rules/actions.md	2,108 (new)
skills/objectstack-ui/rules/dashboards.md	6,090 (new)
skills/objectstack-ui/rules/list-views.md	3,011 (new)
skills/objectstack-ui/rules/navigation.md	2,273 (new)
skills/objectstack-ui/rules/pages.md	5,501 (new)

(c) Not a channel for added prose — proved line by line, not asserted. Comparing the multiset of non-blank lines before (1,591) and after (1,586): 28 lines in the new tree are not lines of the old SKILL.md, and not one of them is a new sentence — 5 file titles, 1## Detailed Rules heading, 5 one-line pointers, 10 headings whose only change is a level shift or a deletion inside the heading text, and 7 lines whose only change is a rewritten relative link or a deleted dead clause. That is the entire set of text this PR authored. The check is reproducible from the diff.

File map

old section (spans at the merge base)	new home	tokens
## Configuring a List View · ## Configuring Kanban Views · ## Configuring Gantt Views	rules/list-views.md	3,011
## App Navigation (the App.create example) · ## Three Run Modes · ## Record Presentation	rules/navigation.md	2,273
## Dashboards · ## Dashboards (cont.) · ### Report Configuration · ## Analytics Cubes	rules/dashboards.md	6,090
## Pages — Lightning-Style Page Layouts · ## Docs — Package Documentation	rules/pages.md	5,501
## Actions (body below ### Action Types)	rules/actions.md	2,108
frontmatter · ## View Types · ## Detailed Rules · ## App Navigation head · ## Report Types · ## CRM UI Blueprint · ## Date Macros · ## Actions head · ## Common Pitfalls · ## Verify your work · ## References	stays in SKILL.md	3,815

Five files, not the audit's suggested nine — the brief says fewer is better, and two of the nine were folded deliberately: rules/filter-tokens.md was not created because ## Date Macros is cross-cutting (list views, dashboards, reports, pages) and is the target of the package's only in-file anchors, so it stays in the entry; rules/master-detail.md was not created because ### Master-Detail Forms is a routing ladder ("you almost never need a custom page or form config — prefer, in order"), which is exactly the kind of construct the entry keeps.

Dashboards now have one home. The two sections that sat 780 lines apart are adjacent in rules/dashboards.md, the second demoted to ### KPI Widgets, Filters, Drilldown under the first — which is what makes the three UI-B-02 deletions below dead text rather than a judgement call.

Deletions, per finding

finding	落点	before	after
UI-B-02	SKILL.md:554-558	"See the Production Pattern section below for the full Dashboard shape with refreshInterval, header actions, date range, global filters, widget options, and the period-over-period (compareTo) modifier; date bucketing comes from the bound dataset dimension's dateGranularity (ADR-0021)."	deleted — a forward reference to a section that is now the next thing on the page (-75)
UI-B-02	SKILL.md:1316-1317	"Dashboards (Dashboard) are first-class metadata. Beyond the basic widget layout shown above, the production-grade pattern uses:"	deleted — a second introduction to a section that no longer restarts (-33)
UI-B-02	SKILL.md:1381-1384	the "Tokens in filters" blockquote, pointing at Date Macros for the second time in the same file	deleted — the merged file carries that pointer once (-69)
UI-B-05 residue	SKILL.md:1029-1032	"Real component props/callbacks flow through — e.g. ObjectForm honors objectName / mode / recordId / formType / onSuccess / onCancel; ListView honors objectName / fields / onRowClick / navigation."	deleted — the very next blockquote says "Do not guess props — read the contract … author against it, not from memory", and this sentence is a from-memory copy of that generated contract (-57 with its trailing clause)
UI-B-02 (split-exposed)	SKILL.md## Report Types	"a report visualizes via its embedded chart: config (see the example below)"	the parenthetical deleted — the report example is now in rules/dashboards.md (-6)
UI-B-02 (split-exposed)	rules/navigation.md Record Presentation	"for bespoke layout, assign a record Page (below)"	the parenthetical deleted — Pages is now rules/pages.md (-2)
UI-B-02 (split-exposed)	rules/pages.md page type list	"configured via interfaceConfig (see the iron rule above)"	the parenthetical deleted — the iron rule is now in rules/navigation.md (-6)

Gross -248. No decision-changing construct was deleted, and nothing was deleted to reach a number — condition (a) is satisfied by the split being byte-conserving, not by these.

Every "above" / "below" in the new tree was swept for a target that left its file; those three were the only ones, and each is deleted rather than rewritten — a dead pointer is removed text, not an occasion for new prose.

The one-topic-two-files cost, declared

check:skill-identifier-liveness Leg 2 binds three sections to skills/objectstack-ui/SKILL.mdby path (report-types, action-types, navigation-item-types); a bound heading that leaves its file yields [stale-binding] … A renamed heading unbinds the assertion, so it fails here rather than quietly checking nothing, pinned by that script's own self-test. Nothing in the script was changed. So three topics are split across two files:

• ## Report Types keeps its exhaustive table in the entry; ### Report Configuration moved to rules/dashboards.md (promoted to ##).
• ### Action Types keeps its table in the entry; the rest of Actions moved to rules/actions.md (headings shifted up one level).
• ### Navigation Item Types keeps its table in the entry; the App.create example moved to rules/navigation.md.

That is a real legibility cost of the split as constrained, and it is the reason those three entry sections look thin.

Link rewrites

Three links crossed a directory boundary and were rewritten, following the skills/objectstack-platform/rules/ precedent: #date-macros--filter-placeholders became ../SKILL.md#date-macros--filter-placeholders, and ./references/react-blocks.md / ./contracts/react-blocks.contract.json became ../references/… / ../contracts/….

Not in this PR

• UI-B-04 (the duplicated generated contracts/react-blocks.contract.json, 5,352 tokens) belongs to [Decision] Skills optimization program — batch 1 (4 items): new-file ceilings for splits · planned-eval stubs · react-blocks double rendering · published pm-dispatch scope #14296 item 3 and is the spec lane's seam card Seam (spec lane, ruled): gen:react-blocks stops emitting skills/objectstack-ui/contracts/react-blocks.contract.json — the markdown rendering becomes the single AI-facing output (#14296 item 3 = A) #14559. Generated output is untouched here.
• UI-H-01 (add an eval fixture) is retired: [Decision] Skills optimization program — batch 1 (4 items): new-file ceilings for splits · planned-eval stubs · react-blocks double rendering · published pm-dispatch scope #14296 item 2 = A rules that eval stubs are deleted rather than built out, so the proposal to add a fixture no longer stands. evals/ is untouched; the real fixture stays.

premise_false: none — no audit finding was skipped this round for a false premise.

Gates

Head d72e3dcc. Family re-derived on the final commit with node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack --commands (25 commands, from the script's own merge-base changeset — not a hand-written path list), and the whole union re-run on that head under scripts/pm/os-verify-lock.sh with OS_VERIFY_LOCK_SLOT=issue-14298-r2. Every exit code captured before any pipe. 31 ran, 30 exit 0.

The six beyond the derivation, each owed for a named reason: check-skills-token-ratchet.mjs --self-test (this PR edits that gate, so its own self-test is owed on top of the derived family — 64 cases pass), check:skill-examples and check:skill-refs (the dispatch named them; skills/** changed), check:doc-anchors and check:published-readme-links (links crossed a directory), check:nul-bytes (any edit).

The verdicts that answer this PR's specific risks, quoted from the gates themselves:

• check:skill-examples — "256 prose examples type-check across 3 surface(s) — every marked block parsed, so tsc ran the SEMANTIC pass on all of them". All 15 os:check markers travelled with their fences (0 in the entry, 15 across the five rules files; 15 before, 15 after) and the marked blocks still compile from their new files.
• check:skill-identifier-liveness — "Leg 1: 465 citation(s) over 46 published file(s) … Leg 2: 8 registered exhaustive section(s), 0 ledgered gap(s)". The three path-keyed ui bindings still resolve inside SKILL.md.
• check:published-readme-links — "176 outbound link(s) across 60 published markdown file(s) … 1 anchor(s) verified, 103/103 relative target(s) found". The new ../SKILL.md#date-macros--filter-placeholders and ../references/… links resolve.
• check:doc-anchors — "295 internal #fragment link(s) across 409 source file(s) all resolve to a real heading".
• check:skill-docs / check:skill-refs — both in sync; no generated file needed regenerating, and none was touched.
• check:role-word — green after the baseline re-point (its own commit; see below).
• node scripts/check-skills-token-ratchet.mjs — "36 authored bundle file(s) within their ceilings", every objectstack-ui row at exactly zero headroom.

check:test-completeness exited 3, which is NOT MEASURED, not a failure — the gate says so itself: "Nothing was measured: this gate exited before parsing a single summary line … ⛔ It is NOT a finding". Recorded as such rather than counted green.

check:role-word needed the baseline re-pointed, and that is a pure relocation. The baseline is keyed by PATH; the two occurrences it already carried for SKILL.md ("the tab-bar role", "semantic roles") moved into rules/list-views.md and rules/navigation.md as byte-identical lines. --update therefore produced a three-line diff — one row removed, two added, total unchanged at 2 — and re-baselined no other file. It is in its own commit. The count did not move, which is what the brief asked to be shown.

Repo-wide pnpm lint was not run; the narrowing is measured, not assumed. (1) The population comes from eslint's own config, not a guess: every files glob in eslint.config.mjs is {ts,tsx,mts,cts,js,jsx,mjs,cjs}, no markdown plugin is configured, and .md appears in no glob — so the six markdown files in this diff are outside eslint's population entirely. (2) Counted from --format json: of the eight changed files eslint claims exactly one, scripts/check-skills-token-ratchet.mjs, at 0 errors / 0 warnings; role-word-baseline.json reports "File ignored because no matching configuration was supplied". (3) Invariance for untouched files: the config configures no projectService and no project, so type-aware linting is off and nothing in this diff can move the verdict of a file it does not contain. CI's Lint job runs the full sweep regardless.

🤖 Generated with Claude Code

https://claude.ai/code/session_01LraLgQVGq8egUwfYZpbYt1

---

Generated by Claude Code

[os-litant]

Part of #14298

Ruling: "#14296 item 1 = A, comment 5507419465". Condition (a) was read at programme level by "#14568 = A, comment 5511939006" (director seat, verbatim reply 「同意」): the split PR itself must land token-neutral or negative, and the package arithmetic is carried by the programme.

The three conditions, with numbers

(a) Token-neutral or negative — measured -58.node scripts/check-skills-token-ratchet.mjs before and after, at head d72e3dcc:

	before	after
SKILL.md	22,856	3,815
rules/*.md (5 new files)	—	18,983
generated (contracts/react-blocks.contract.json, references/react-blocks.md, references/_index.md)	9,710	9,710
evals/	1,391	1,391
package	33,957	33,899

Net -58 (-0.17%). Gross deletions -248, additions +190 (five file titles, one ## Detailed Rules heading, five one-line pointers). Nothing in this PR is positive.

Programme account for the package: round 1 #14338-2,585 + this split -58 + seam card #14559-5,352 = -7,995 against the pre-programme 36,542 = -21.9%.

(b) Ceilings. Five rows added, each pinned AT its landed count so every new file starts with exactly zero headroom; the entry row re-locked to its landed count. No other row changed.

row	ceiling
skills/objectstack-ui/SKILL.md	25,445 → 3,815 (re-lock)
skills/objectstack-ui/rules/actions.md	2,108 (new)
skills/objectstack-ui/rules/dashboards.md	6,090 (new)
skills/objectstack-ui/rules/list-views.md	3,011 (new)
skills/objectstack-ui/rules/navigation.md	2,273 (new)
skills/objectstack-ui/rules/pages.md	5,501 (new)

(c) Not a channel for added prose — proved line by line, not asserted. Comparing the multiset of non-blank lines before (1,591) and after (1,586): 28 lines in the new tree are not lines of the old SKILL.md, and not one of them is a new sentence — 5 file titles, 1## Detailed Rules heading, 5 one-line pointers, 10 headings whose only change is a level shift or a deletion inside the heading text, and 7 lines whose only change is a rewritten relative link or a deleted dead clause. That is the entire set of text this PR authored. The check is reproducible from the diff.

File map

old section (spans at the merge base)	new home	tokens
## Configuring a List View · ## Configuring Kanban Views · ## Configuring Gantt Views	rules/list-views.md	3,011
## App Navigation (the App.create example) · ## Three Run Modes · ## Record Presentation	rules/navigation.md	2,273
## Dashboards · ## Dashboards (cont.) · ### Report Configuration · ## Analytics Cubes	rules/dashboards.md	6,090
## Pages — Lightning-Style Page Layouts · ## Docs — Package Documentation	rules/pages.md	5,501
## Actions (body below ### Action Types)	rules/actions.md	2,108
frontmatter · ## View Types · ## Detailed Rules · ## App Navigation head · ## Report Types · ## CRM UI Blueprint · ## Date Macros · ## Actions head · ## Common Pitfalls · ## Verify your work · ## References	stays in SKILL.md	3,815

Five files, not the audit's suggested nine — the brief says fewer is better, and two of the nine were folded deliberately: rules/filter-tokens.md was not created because ## Date Macros is cross-cutting (list views, dashboards, reports, pages) and is the target of the package's only in-file anchors, so it stays in the entry; rules/master-detail.md was not created because ### Master-Detail Forms is a routing ladder ("you almost never need a custom page or form config — prefer, in order"), which is exactly the kind of construct the entry keeps.

Dashboards now have one home. The two sections that sat 780 lines apart are adjacent in rules/dashboards.md, the second demoted to ### KPI Widgets, Filters, Drilldown under the first — which is what makes the three UI-B-02 deletions below dead text rather than a judgement call.

Deletions, per finding

finding	落点	before	after
UI-B-02	SKILL.md:554-558	"See the Production Pattern section below for the full Dashboard shape with refreshInterval, header actions, date range, global filters, widget options, and the period-over-period (compareTo) modifier; date bucketing comes from the bound dataset dimension's dateGranularity (ADR-0021)."	deleted — a forward reference to a section that is now the next thing on the page (-75)
UI-B-02	SKILL.md:1316-1317	"Dashboards (Dashboard) are first-class metadata. Beyond the basic widget layout shown above, the production-grade pattern uses:"	deleted — a second introduction to a section that no longer restarts (-33)
UI-B-02	SKILL.md:1381-1384	the "Tokens in filters" blockquote, pointing at Date Macros for the second time in the same file	deleted — the merged file carries that pointer once (-69)
UI-B-05 residue	SKILL.md:1029-1032	"Real component props/callbacks flow through — e.g. ObjectForm honors objectName / mode / recordId / formType / onSuccess / onCancel; ListView honors objectName / fields / onRowClick / navigation."	deleted — the very next blockquote says "Do not guess props — read the contract … author against it, not from memory", and this sentence is a from-memory copy of that generated contract (-57 with its trailing clause)
UI-B-02 (split-exposed)	SKILL.md## Report Types	"a report visualizes via its embedded chart: config (see the example below)"	the parenthetical deleted — the report example is now in rules/dashboards.md (-6)
UI-B-02 (split-exposed)	rules/navigation.md Record Presentation	"for bespoke layout, assign a record Page (below)"	the parenthetical deleted — Pages is now rules/pages.md (-2)
UI-B-02 (split-exposed)	rules/pages.md page type list	"configured via interfaceConfig (see the iron rule above)"	the parenthetical deleted — the iron rule is now in rules/navigation.md (-6)

Gross -248. No decision-changing construct was deleted, and nothing was deleted to reach a number — condition (a) is satisfied by the split being byte-conserving, not by these.

Every "above" / "below" in the new tree was swept for a target that left its file; those three were the only ones, and each is deleted rather than rewritten — a dead pointer is removed text, not an occasion for new prose.

The one-topic-two-files cost, declared

check:skill-identifier-liveness Leg 2 binds three sections to skills/objectstack-ui/SKILL.mdby path (report-types, action-types, navigation-item-types); a bound heading that leaves its file yields [stale-binding] … A renamed heading unbinds the assertion, so it fails here rather than quietly checking nothing, pinned by that script's own self-test. Nothing in the script was changed. So three topics are split across two files:

• ## Report Types keeps its exhaustive table in the entry; ### Report Configuration moved to rules/dashboards.md (promoted to ##).
• ### Action Types keeps its table in the entry; the rest of Actions moved to rules/actions.md (headings shifted up one level).
• ### Navigation Item Types keeps its table in the entry; the App.create example moved to rules/navigation.md.

That is a real legibility cost of the split as constrained, and it is the reason those three entry sections look thin.

Link rewrites

Three links crossed a directory boundary and were rewritten, following the skills/objectstack-platform/rules/ precedent: #date-macros--filter-placeholders became ../SKILL.md#date-macros--filter-placeholders, and ./references/react-blocks.md / ./contracts/react-blocks.contract.json became ../references/… / ../contracts/….

Not in this PR

• UI-B-04 (the duplicated generated contracts/react-blocks.contract.json, 5,352 tokens) belongs to [Decision] Skills optimization program — batch 1 (4 items): new-file ceilings for splits · planned-eval stubs · react-blocks double rendering · published pm-dispatch scope #14296 item 3 and is the spec lane's seam card Seam (spec lane, ruled): gen:react-blocks stops emitting skills/objectstack-ui/contracts/react-blocks.contract.json — the markdown rendering becomes the single AI-facing output (#14296 item 3 = A) #14559. Generated output is untouched here.
• UI-H-01 (add an eval fixture) is retired: [Decision] Skills optimization program — batch 1 (4 items): new-file ceilings for splits · planned-eval stubs · react-blocks double rendering · published pm-dispatch scope #14296 item 2 = A rules that eval stubs are deleted rather than built out, so the proposal to add a fixture no longer stands. evals/ is untouched; the real fixture stays.

premise_false: none — no audit finding was skipped this round for a false premise.

Gates

Head d72e3dcc. Family re-derived on the final commit with node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack --commands (25 commands, from the script's own merge-base changeset — not a hand-written path list), and the whole union re-run on that head under scripts/pm/os-verify-lock.sh with OS_VERIFY_LOCK_SLOT=issue-14298-r2. Every exit code captured before any pipe. 31 ran, 30 exit 0.

The six beyond the derivation, each owed for a named reason: check-skills-token-ratchet.mjs --self-test (this PR edits that gate, so its own self-test is owed on top of the derived family — 64 cases pass), check:skill-examples and check:skill-refs (the dispatch named them; skills/** changed), check:doc-anchors and check:published-readme-links (links crossed a directory), check:nul-bytes (any edit).

The verdicts that answer this PR's specific risks, quoted from the gates themselves:

• check:skill-examples — "256 prose examples type-check across 3 surface(s) — every marked block parsed, so tsc ran the SEMANTIC pass on all of them". All 15 os:check markers travelled with their fences (0 in the entry, 15 across the five rules files; 15 before, 15 after) and the marked blocks still compile from their new files.
• check:skill-identifier-liveness — "Leg 1: 465 citation(s) over 46 published file(s) … Leg 2: 8 registered exhaustive section(s), 0 ledgered gap(s)". The three path-keyed ui bindings still resolve inside SKILL.md.
• check:published-readme-links — "176 outbound link(s) across 60 published markdown file(s) … 1 anchor(s) verified, 103/103 relative target(s) found". The new ../SKILL.md#date-macros--filter-placeholders and ../references/… links resolve.
• check:doc-anchors — "295 internal #fragment link(s) across 409 source file(s) all resolve to a real heading".
• check:skill-docs / check:skill-refs — both in sync; no generated file needed regenerating, and none was touched.
• check:role-word — green after the baseline re-point (its own commit; see below).
• node scripts/check-skills-token-ratchet.mjs — "36 authored bundle file(s) within their ceilings", every objectstack-ui row at exactly zero headroom.

check:test-completeness exited 3, which is NOT MEASURED, not a failure — the gate says so itself: "Nothing was measured: this gate exited before parsing a single summary line … ⛔ It is NOT a finding". Recorded as such rather than counted green.

check:role-word needed the baseline re-pointed, and that is a pure relocation. The baseline is keyed by PATH; the two occurrences it already carried for SKILL.md ("the tab-bar role", "semantic roles") moved into rules/list-views.md and rules/navigation.md as byte-identical lines. --update therefore produced a three-line diff — one row removed, two added, total unchanged at 2 — and re-baselined no other file. It is in its own commit. The count did not move, which is what the brief asked to be shown.

Repo-wide pnpm lint was not run; the narrowing is measured, not assumed. (1) The population comes from eslint's own config, not a guess: every files glob in eslint.config.mjs is {ts,tsx,mts,cts,js,jsx,mjs,cjs}, no markdown plugin is configured, and .md appears in no glob — so the six markdown files in this diff are outside eslint's population entirely. (2) Counted from --format json: of the eight changed files eslint claims exactly one, scripts/check-skills-token-ratchet.mjs, at 0 errors / 0 warnings; role-word-baseline.json reports "File ignored because no matching configuration was supplied". (3) Invariance for untouched files: the config configures no projectService and no project, so type-aware linting is off and nothing in this diff can move the verdict of a file it does not contain. CI's Lint job runs the full sweep regardless.

🤖 Generated with Claude Code

https://claude.ai/code/session_01LraLgQVGq8egUwfYZpbYt1

---

Generated by Claude Code

[os-litant]

Part of #14298

Ruling: "#14296 item 1 = A, comment 5507419465". Condition (a) was read at programme level by "#14568 = A, comment 5511939006" (director seat, verbatim reply 「同意」): the split PR itself must land token-neutral or negative, and the package arithmetic is carried by the programme.

The three conditions, with numbers

(a) Token-neutral or negative — measured -58.node scripts/check-skills-token-ratchet.mjs before and after, at head d72e3dcc:

	before	after
SKILL.md	22,856	3,815
rules/*.md (5 new files)	—	18,983
generated (contracts/react-blocks.contract.json, references/react-blocks.md, references/_index.md)	9,710	9,710
evals/	1,391	1,391
package	33,957	33,899

Net -58 (-0.17%). Gross deletions -248, additions +190 (five file titles, one ## Detailed Rules heading, five one-line pointers). Nothing in this PR is positive.

Programme account for the package: round 1 #14338-2,585 + this split -58 + seam card #14559-5,352 = -7,995 against the pre-programme 36,542 = -21.9%.

(b) Ceilings. Five rows added, each pinned AT its landed count so every new file starts with exactly zero headroom; the entry row re-locked to its landed count. No other row changed.

row	ceiling
skills/objectstack-ui/SKILL.md	25,445 → 3,815 (re-lock)
skills/objectstack-ui/rules/actions.md	2,108 (new)
skills/objectstack-ui/rules/dashboards.md	6,090 (new)
skills/objectstack-ui/rules/list-views.md	3,011 (new)
skills/objectstack-ui/rules/navigation.md	2,273 (new)
skills/objectstack-ui/rules/pages.md	5,501 (new)

(c) Not a channel for added prose — proved line by line, not asserted. Comparing the multiset of non-blank lines before (1,591) and after (1,586): 28 lines in the new tree are not lines of the old SKILL.md, and not one of them is a new sentence — 5 file titles, 1## Detailed Rules heading, 5 one-line pointers, 10 headings whose only change is a level shift or a deletion inside the heading text, and 7 lines whose only change is a rewritten relative link or a deleted dead clause. That is the entire set of text this PR authored. The check is reproducible from the diff.

File map

old section (spans at the merge base)	new home	tokens
## Configuring a List View · ## Configuring Kanban Views · ## Configuring Gantt Views	rules/list-views.md	3,011
## App Navigation (the App.create example) · ## Three Run Modes · ## Record Presentation	rules/navigation.md	2,273
## Dashboards · ## Dashboards (cont.) · ### Report Configuration · ## Analytics Cubes	rules/dashboards.md	6,090
## Pages — Lightning-Style Page Layouts · ## Docs — Package Documentation	rules/pages.md	5,501
## Actions (body below ### Action Types)	rules/actions.md	2,108
frontmatter · ## View Types · ## Detailed Rules · ## App Navigation head · ## Report Types · ## CRM UI Blueprint · ## Date Macros · ## Actions head · ## Common Pitfalls · ## Verify your work · ## References	stays in SKILL.md	3,815

Five files, not the audit's suggested nine — the brief says fewer is better, and two of the nine were folded deliberately: rules/filter-tokens.md was not created because ## Date Macros is cross-cutting (list views, dashboards, reports, pages) and is the target of the package's only in-file anchors, so it stays in the entry; rules/master-detail.md was not created because ### Master-Detail Forms is a routing ladder ("you almost never need a custom page or form config — prefer, in order"), which is exactly the kind of construct the entry keeps.

Dashboards now have one home. The two sections that sat 780 lines apart are adjacent in rules/dashboards.md, the second demoted to ### KPI Widgets, Filters, Drilldown under the first — which is what makes the three UI-B-02 deletions below dead text rather than a judgement call.

Deletions, per finding

finding	落点	before	after
UI-B-02	SKILL.md:554-558	"See the Production Pattern section below for the full Dashboard shape with refreshInterval, header actions, date range, global filters, widget options, and the period-over-period (compareTo) modifier; date bucketing comes from the bound dataset dimension's dateGranularity (ADR-0021)."	deleted — a forward reference to a section that is now the next thing on the page (-75)
UI-B-02	SKILL.md:1316-1317	"Dashboards (Dashboard) are first-class metadata. Beyond the basic widget layout shown above, the production-grade pattern uses:"	deleted — a second introduction to a section that no longer restarts (-33)
UI-B-02	SKILL.md:1381-1384	the "Tokens in filters" blockquote, pointing at Date Macros for the second time in the same file	deleted — the merged file carries that pointer once (-69)
UI-B-05 residue	SKILL.md:1029-1032	"Real component props/callbacks flow through — e.g. ObjectForm honors objectName / mode / recordId / formType / onSuccess / onCancel; ListView honors objectName / fields / onRowClick / navigation."	deleted — the very next blockquote says "Do not guess props — read the contract … author against it, not from memory", and this sentence is a from-memory copy of that generated contract (-57 with its trailing clause)
UI-B-02 (split-exposed)	SKILL.md## Report Types	"a report visualizes via its embedded chart: config (see the example below)"	the parenthetical deleted — the report example is now in rules/dashboards.md (-6)
UI-B-02 (split-exposed)	rules/navigation.md Record Presentation	"for bespoke layout, assign a record Page (below)"	the parenthetical deleted — Pages is now rules/pages.md (-2)
UI-B-02 (split-exposed)	rules/pages.md page type list	"configured via interfaceConfig (see the iron rule above)"	the parenthetical deleted — the iron rule is now in rules/navigation.md (-6)

Gross -248. No decision-changing construct was deleted, and nothing was deleted to reach a number — condition (a) is satisfied by the split being byte-conserving, not by these.

Every "above" / "below" in the new tree was swept for a target that left its file; those three were the only ones, and each is deleted rather than rewritten — a dead pointer is removed text, not an occasion for new prose.

The one-topic-two-files cost, declared

check:skill-identifier-liveness Leg 2 binds three sections to skills/objectstack-ui/SKILL.mdby path (report-types, action-types, navigation-item-types); a bound heading that leaves its file yields [stale-binding] … A renamed heading unbinds the assertion, so it fails here rather than quietly checking nothing, pinned by that script's own self-test. Nothing in the script was changed. So three topics are split across two files:

• ## Report Types keeps its exhaustive table in the entry; ### Report Configuration moved to rules/dashboards.md (promoted to ##).
• ### Action Types keeps its table in the entry; the rest of Actions moved to rules/actions.md (headings shifted up one level).
• ### Navigation Item Types keeps its table in the entry; the App.create example moved to rules/navigation.md.

That is a real legibility cost of the split as constrained, and it is the reason those three entry sections look thin.

Link rewrites

Three links crossed a directory boundary and were rewritten, following the skills/objectstack-platform/rules/ precedent: #date-macros--filter-placeholders became ../SKILL.md#date-macros--filter-placeholders, and ./references/react-blocks.md / ./contracts/react-blocks.contract.json became ../references/… / ../contracts/….

Not in this PR

• UI-B-04 (the duplicated generated contracts/react-blocks.contract.json, 5,352 tokens) belongs to [Decision] Skills optimization program — batch 1 (4 items): new-file ceilings for splits · planned-eval stubs · react-blocks double rendering · published pm-dispatch scope #14296 item 3 and is the spec lane's seam card Seam (spec lane, ruled): gen:react-blocks stops emitting skills/objectstack-ui/contracts/react-blocks.contract.json — the markdown rendering becomes the single AI-facing output (#14296 item 3 = A) #14559. Generated output is untouched here.
• UI-H-01 (add an eval fixture) is retired: [Decision] Skills optimization program — batch 1 (4 items): new-file ceilings for splits · planned-eval stubs · react-blocks double rendering · published pm-dispatch scope #14296 item 2 = A rules that eval stubs are deleted rather than built out, so the proposal to add a fixture no longer stands. evals/ is untouched; the real fixture stays.

premise_false: none — no audit finding was skipped this round for a false premise.

Gates

Head d72e3dcc. Family re-derived on the final commit with node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack --commands (25 commands, from the script's own merge-base changeset — not a hand-written path list), and the whole union re-run on that head under scripts/pm/os-verify-lock.sh with OS_VERIFY_LOCK_SLOT=issue-14298-r2. Every exit code captured before any pipe. 31 ran, 30 exit 0.

The six beyond the derivation, each owed for a named reason: check-skills-token-ratchet.mjs --self-test (this PR edits that gate, so its own self-test is owed on top of the derived family — 64 cases pass), check:skill-examples and check:skill-refs (the dispatch named them; skills/** changed), check:doc-anchors and check:published-readme-links (links crossed a directory), check:nul-bytes (any edit).

The verdicts that answer this PR's specific risks, quoted from the gates themselves:

• check:skill-examples — "256 prose examples type-check across 3 surface(s) — every marked block parsed, so tsc ran the SEMANTIC pass on all of them". All 15 os:check markers travelled with their fences (0 in the entry, 15 across the five rules files; 15 before, 15 after) and the marked blocks still compile from their new files.
• check:skill-identifier-liveness — "Leg 1: 465 citation(s) over 46 published file(s) … Leg 2: 8 registered exhaustive section(s), 0 ledgered gap(s)". The three path-keyed ui bindings still resolve inside SKILL.md.
• check:published-readme-links — "176 outbound link(s) across 60 published markdown file(s) … 1 anchor(s) verified, 103/103 relative target(s) found". The new ../SKILL.md#date-macros--filter-placeholders and ../references/… links resolve.
• check:doc-anchors — "295 internal #fragment link(s) across 409 source file(s) all resolve to a real heading".
• check:skill-docs / check:skill-refs — both in sync; no generated file needed regenerating, and none was touched.
• check:role-word — green after the baseline re-point (its own commit; see below).
• node scripts/check-skills-token-ratchet.mjs — "36 authored bundle file(s) within their ceilings", every objectstack-ui row at exactly zero headroom.

check:test-completeness exited 3, which is NOT MEASURED, not a failure — the gate says so itself: "Nothing was measured: this gate exited before parsing a single summary line … ⛔ It is NOT a finding". Recorded as such rather than counted green.

check:role-word needed the baseline re-pointed, and that is a pure relocation. The baseline is keyed by PATH; the two occurrences it already carried for SKILL.md ("the tab-bar role", "semantic roles") moved into rules/list-views.md and rules/navigation.md as byte-identical lines. --update therefore produced a three-line diff — one row removed, two added, total unchanged at 2 — and re-baselined no other file. It is in its own commit. The count did not move, which is what the brief asked to be shown.

Repo-wide pnpm lint was not run; the narrowing is measured, not assumed. (1) The population comes from eslint's own config, not a guess: every files glob in eslint.config.mjs is {ts,tsx,mts,cts,js,jsx,mjs,cjs}, no markdown plugin is configured, and .md appears in no glob — so the six markdown files in this diff are outside eslint's population entirely. (2) Counted from --format json: of the eight changed files eslint claims exactly one, scripts/check-skills-token-ratchet.mjs, at 0 errors / 0 warnings; role-word-baseline.json reports "File ignored because no matching configuration was supplied". (3) Invariance for untouched files: the config configures no projectService and no project, so type-aware linting is off and nothing in this diff can move the verdict of a file it does not contain. CI's Lint job runs the full sweep regardless.

🤖 Generated with Claude Code

https://claude.ai/code/session_01LraLgQVGq8egUwfYZpbYt1

---

Generated by Claude Code

[os-litant]

Part of #14298

Ruling: "#14296 item 1 = A, comment 5507419465". Condition (a) was read at programme level by "#14568 = A, comment 5511939006" (director seat, verbatim reply 「同意」): the split PR itself must land token-neutral or negative, and the package arithmetic is carried by the programme.

The three conditions, with numbers

(a) Token-neutral or negative — measured -58.node scripts/check-skills-token-ratchet.mjs before and after, at head d72e3dcc:

	before	after
SKILL.md	22,856	3,815
rules/*.md (5 new files)	—	18,983
generated (contracts/react-blocks.contract.json, references/react-blocks.md, references/_index.md)	9,710	9,710
evals/	1,391	1,391
package	33,957	33,899

Net -58 (-0.17%). Gross deletions -248, additions +190 (five file titles, one ## Detailed Rules heading, five one-line pointers). Nothing in this PR is positive.

Programme account for the package: round 1 #14338-2,585 + this split -58 + seam card #14559-5,352 = -7,995 against the pre-programme 36,542 = -21.9%.

(b) Ceilings. Five rows added, each pinned AT its landed count so every new file starts with exactly zero headroom; the entry row re-locked to its landed count. No other row changed.

row	ceiling
skills/objectstack-ui/SKILL.md	25,445 → 3,815 (re-lock)
skills/objectstack-ui/rules/actions.md	2,108 (new)
skills/objectstack-ui/rules/dashboards.md	6,090 (new)
skills/objectstack-ui/rules/list-views.md	3,011 (new)
skills/objectstack-ui/rules/navigation.md	2,273 (new)
skills/objectstack-ui/rules/pages.md	5,501 (new)

(c) Not a channel for added prose — proved line by line, not asserted. Comparing the multiset of non-blank lines before (1,591) and after (1,586): 28 lines in the new tree are not lines of the old SKILL.md, and not one of them is a new sentence — 5 file titles, 1## Detailed Rules heading, 5 one-line pointers, 10 headings whose only change is a level shift or a deletion inside the heading text, and 7 lines whose only change is a rewritten relative link or a deleted dead clause. That is the entire set of text this PR authored. The check is reproducible from the diff.

File map

old section (spans at the merge base)	new home	tokens
## Configuring a List View · ## Configuring Kanban Views · ## Configuring Gantt Views	rules/list-views.md	3,011
## App Navigation (the App.create example) · ## Three Run Modes · ## Record Presentation	rules/navigation.md	2,273
## Dashboards · ## Dashboards (cont.) · ### Report Configuration · ## Analytics Cubes	rules/dashboards.md	6,090
## Pages — Lightning-Style Page Layouts · ## Docs — Package Documentation	rules/pages.md	5,501
## Actions (body below ### Action Types)	rules/actions.md	2,108
frontmatter · ## View Types · ## Detailed Rules · ## App Navigation head · ## Report Types · ## CRM UI Blueprint · ## Date Macros · ## Actions head · ## Common Pitfalls · ## Verify your work · ## References	stays in SKILL.md	3,815

Five files, not the audit's suggested nine — the brief says fewer is better, and two of the nine were folded deliberately: rules/filter-tokens.md was not created because ## Date Macros is cross-cutting (list views, dashboards, reports, pages) and is the target of the package's only in-file anchors, so it stays in the entry; rules/master-detail.md was not created because ### Master-Detail Forms is a routing ladder ("you almost never need a custom page or form config — prefer, in order"), which is exactly the kind of construct the entry keeps.

Dashboards now have one home. The two sections that sat 780 lines apart are adjacent in rules/dashboards.md, the second demoted to ### KPI Widgets, Filters, Drilldown under the first — which is what makes the three UI-B-02 deletions below dead text rather than a judgement call.

Deletions, per finding

finding	落点	before	after
UI-B-02	SKILL.md:554-558	"See the Production Pattern section below for the full Dashboard shape with refreshInterval, header actions, date range, global filters, widget options, and the period-over-period (compareTo) modifier; date bucketing comes from the bound dataset dimension's dateGranularity (ADR-0021)."	deleted — a forward reference to a section that is now the next thing on the page (-75)
UI-B-02	SKILL.md:1316-1317	"Dashboards (Dashboard) are first-class metadata. Beyond the basic widget layout shown above, the production-grade pattern uses:"	deleted — a second introduction to a section that no longer restarts (-33)
UI-B-02	SKILL.md:1381-1384	the "Tokens in filters" blockquote, pointing at Date Macros for the second time in the same file	deleted — the merged file carries that pointer once (-69)
UI-B-05 residue	SKILL.md:1029-1032	"Real component props/callbacks flow through — e.g. ObjectForm honors objectName / mode / recordId / formType / onSuccess / onCancel; ListView honors objectName / fields / onRowClick / navigation."	deleted — the very next blockquote says "Do not guess props — read the contract … author against it, not from memory", and this sentence is a from-memory copy of that generated contract (-57 with its trailing clause)
UI-B-02 (split-exposed)	SKILL.md## Report Types	"a report visualizes via its embedded chart: config (see the example below)"	the parenthetical deleted — the report example is now in rules/dashboards.md (-6)
UI-B-02 (split-exposed)	rules/navigation.md Record Presentation	"for bespoke layout, assign a record Page (below)"	the parenthetical deleted — Pages is now rules/pages.md (-2)
UI-B-02 (split-exposed)	rules/pages.md page type list	"configured via interfaceConfig (see the iron rule above)"	the parenthetical deleted — the iron rule is now in rules/navigation.md (-6)

Gross -248. No decision-changing construct was deleted, and nothing was deleted to reach a number — condition (a) is satisfied by the split being byte-conserving, not by these.

Every "above" / "below" in the new tree was swept for a target that left its file; those three were the only ones, and each is deleted rather than rewritten — a dead pointer is removed text, not an occasion for new prose.

The one-topic-two-files cost, declared

check:skill-identifier-liveness Leg 2 binds three sections to skills/objectstack-ui/SKILL.mdby path (report-types, action-types, navigation-item-types); a bound heading that leaves its file yields [stale-binding] … A renamed heading unbinds the assertion, so it fails here rather than quietly checking nothing, pinned by that script's own self-test. Nothing in the script was changed. So three topics are split across two files:

• ## Report Types keeps its exhaustive table in the entry; ### Report Configuration moved to rules/dashboards.md (promoted to ##).
• ### Action Types keeps its table in the entry; the rest of Actions moved to rules/actions.md (headings shifted up one level).
• ### Navigation Item Types keeps its table in the entry; the App.create example moved to rules/navigation.md.

That is a real legibility cost of the split as constrained, and it is the reason those three entry sections look thin.

Link rewrites

Three links crossed a directory boundary and were rewritten, following the skills/objectstack-platform/rules/ precedent: #date-macros--filter-placeholders became ../SKILL.md#date-macros--filter-placeholders, and ./references/react-blocks.md / ./contracts/react-blocks.contract.json became ../references/… / ../contracts/….

Not in this PR

• UI-B-04 (the duplicated generated contracts/react-blocks.contract.json, 5,352 tokens) belongs to [Decision] Skills optimization program — batch 1 (4 items): new-file ceilings for splits · planned-eval stubs · react-blocks double rendering · published pm-dispatch scope #14296 item 3 and is the spec lane's seam card Seam (spec lane, ruled): gen:react-blocks stops emitting skills/objectstack-ui/contracts/react-blocks.contract.json — the markdown rendering becomes the single AI-facing output (#14296 item 3 = A) #14559. Generated output is untouched here.
• UI-H-01 (add an eval fixture) is retired: [Decision] Skills optimization program — batch 1 (4 items): new-file ceilings for splits · planned-eval stubs · react-blocks double rendering · published pm-dispatch scope #14296 item 2 = A rules that eval stubs are deleted rather than built out, so the proposal to add a fixture no longer stands. evals/ is untouched; the real fixture stays.

premise_false: none — no audit finding was skipped this round for a false premise.

Gates

Head d72e3dcc. Family re-derived on the final commit with node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack --commands (25 commands, from the script's own merge-base changeset — not a hand-written path list), and the whole union re-run on that head under scripts/pm/os-verify-lock.sh with OS_VERIFY_LOCK_SLOT=issue-14298-r2. Every exit code captured before any pipe. 31 ran, 30 exit 0.

The six beyond the derivation, each owed for a named reason: check-skills-token-ratchet.mjs --self-test (this PR edits that gate, so its own self-test is owed on top of the derived family — 64 cases pass), check:skill-examples and check:skill-refs (the dispatch named them; skills/** changed), check:doc-anchors and check:published-readme-links (links crossed a directory), check:nul-bytes (any edit).

The verdicts that answer this PR's specific risks, quoted from the gates themselves:

• check:skill-examples — "256 prose examples type-check across 3 surface(s) — every marked block parsed, so tsc ran the SEMANTIC pass on all of them". All 15 os:check markers travelled with their fences (0 in the entry, 15 across the five rules files; 15 before, 15 after) and the marked blocks still compile from their new files.
• check:skill-identifier-liveness — "Leg 1: 465 citation(s) over 46 published file(s) … Leg 2: 8 registered exhaustive section(s), 0 ledgered gap(s)". The three path-keyed ui bindings still resolve inside SKILL.md.
• check:published-readme-links — "176 outbound link(s) across 60 published markdown file(s) … 1 anchor(s) verified, 103/103 relative target(s) found". The new ../SKILL.md#date-macros--filter-placeholders and ../references/… links resolve.
• check:doc-anchors — "295 internal #fragment link(s) across 409 source file(s) all resolve to a real heading".
• check:skill-docs / check:skill-refs — both in sync; no generated file needed regenerating, and none was touched.
• check:role-word — green after the baseline re-point (its own commit; see below).
• node scripts/check-skills-token-ratchet.mjs — "36 authored bundle file(s) within their ceilings", every objectstack-ui row at exactly zero headroom.

check:test-completeness exited 3, which is NOT MEASURED, not a failure — the gate says so itself: "Nothing was measured: this gate exited before parsing a single summary line … ⛔ It is NOT a finding". Recorded as such rather than counted green.

check:role-word needed the baseline re-pointed, and that is a pure relocation. The baseline is keyed by PATH; the two occurrences it already carried for SKILL.md ("the tab-bar role", "semantic roles") moved into rules/list-views.md and rules/navigation.md as byte-identical lines. --update therefore produced a three-line diff — one row removed, two added, total unchanged at 2 — and re-baselined no other file. It is in its own commit. The count did not move, which is what the brief asked to be shown.

Repo-wide pnpm lint was not run; the narrowing is measured, not assumed. (1) The population comes from eslint's own config, not a guess: every files glob in eslint.config.mjs is {ts,tsx,mts,cts,js,jsx,mjs,cjs}, no markdown plugin is configured, and .md appears in no glob — so the six markdown files in this diff are outside eslint's population entirely. (2) Counted from --format json: of the eight changed files eslint claims exactly one, scripts/check-skills-token-ratchet.mjs, at 0 errors / 0 warnings; role-word-baseline.json reports "File ignored because no matching configuration was supplied". (3) Invariance for untouched files: the config configures no projectService and no project, so type-aware linting is off and nothing in this diff can move the verdict of a file it does not contain. CI's Lint job runs the full sweep regardless.

🤖 Generated with Claude Code

https://claude.ai/code/session_01LraLgQVGq8egUwfYZpbYt1

---

Generated by Claude Code

[os-litant]

Part of #14298

Ruling: "#14296 item 1 = A, comment 5507419465". Condition (a) was read at programme level by "#14568 = A, comment 5511939006" (director seat, verbatim reply 「同意」): the split PR itself must land token-neutral or negative, and the package arithmetic is carried by the programme.

The three conditions, with numbers

(a) Token-neutral or negative — measured -58.node scripts/check-skills-token-ratchet.mjs before and after, at head d72e3dcc:

	before	after
SKILL.md	22,856	3,815
rules/*.md (5 new files)	—	18,983
generated (contracts/react-blocks.contract.json, references/react-blocks.md, references/_index.md)	9,710	9,710
evals/	1,391	1,391
package	33,957	33,899

Net -58 (-0.17%). Gross deletions -248, additions +190 (five file titles, one ## Detailed Rules heading, five one-line pointers). Nothing in this PR is positive.

Programme account for the package: round 1 #14338-2,585 + this split -58 + seam card #14559-5,352 = -7,995 against the pre-programme 36,542 = -21.9%.

(b) Ceilings. Five rows added, each pinned AT its landed count so every new file starts with exactly zero headroom; the entry row re-locked to its landed count. No other row changed.

row	ceiling
skills/objectstack-ui/SKILL.md	25,445 → 3,815 (re-lock)
skills/objectstack-ui/rules/actions.md	2,108 (new)
skills/objectstack-ui/rules/dashboards.md	6,090 (new)
skills/objectstack-ui/rules/list-views.md	3,011 (new)
skills/objectstack-ui/rules/navigation.md	2,273 (new)
skills/objectstack-ui/rules/pages.md	5,501 (new)

(c) Not a channel for added prose — proved line by line, not asserted. Comparing the multiset of non-blank lines before (1,591) and after (1,586): 28 lines in the new tree are not lines of the old SKILL.md, and not one of them is a new sentence — 5 file titles, 1## Detailed Rules heading, 5 one-line pointers, 10 headings whose only change is a level shift or a deletion inside the heading text, and 7 lines whose only change is a rewritten relative link or a deleted dead clause. That is the entire set of text this PR authored. The check is reproducible from the diff.

File map

old section (spans at the merge base)	new home	tokens
## Configuring a List View · ## Configuring Kanban Views · ## Configuring Gantt Views	rules/list-views.md	3,011
## App Navigation (the App.create example) · ## Three Run Modes · ## Record Presentation	rules/navigation.md	2,273
## Dashboards · ## Dashboards (cont.) · ### Report Configuration · ## Analytics Cubes	rules/dashboards.md	6,090
## Pages — Lightning-Style Page Layouts · ## Docs — Package Documentation	rules/pages.md	5,501
## Actions (body below ### Action Types)	rules/actions.md	2,108
frontmatter · ## View Types · ## Detailed Rules · ## App Navigation head · ## Report Types · ## CRM UI Blueprint · ## Date Macros · ## Actions head · ## Common Pitfalls · ## Verify your work · ## References	stays in SKILL.md	3,815

Five files, not the audit's suggested nine — the brief says fewer is better, and two of the nine were folded deliberately: rules/filter-tokens.md was not created because ## Date Macros is cross-cutting (list views, dashboards, reports, pages) and is the target of the package's only in-file anchors, so it stays in the entry; rules/master-detail.md was not created because ### Master-Detail Forms is a routing ladder ("you almost never need a custom page or form config — prefer, in order"), which is exactly the kind of construct the entry keeps.

Dashboards now have one home. The two sections that sat 780 lines apart are adjacent in rules/dashboards.md, the second demoted to ### KPI Widgets, Filters, Drilldown under the first — which is what makes the three UI-B-02 deletions below dead text rather than a judgement call.

Deletions, per finding

finding	落点	before	after
UI-B-02	SKILL.md:554-558	"See the Production Pattern section below for the full Dashboard shape with refreshInterval, header actions, date range, global filters, widget options, and the period-over-period (compareTo) modifier; date bucketing comes from the bound dataset dimension's dateGranularity (ADR-0021)."	deleted — a forward reference to a section that is now the next thing on the page (-75)
UI-B-02	SKILL.md:1316-1317	"Dashboards (Dashboard) are first-class metadata. Beyond the basic widget layout shown above, the production-grade pattern uses:"	deleted — a second introduction to a section that no longer restarts (-33)
UI-B-02	SKILL.md:1381-1384	the "Tokens in filters" blockquote, pointing at Date Macros for the second time in the same file	deleted — the merged file carries that pointer once (-69)
UI-B-05 residue	SKILL.md:1029-1032	"Real component props/callbacks flow through — e.g. ObjectForm honors objectName / mode / recordId / formType / onSuccess / onCancel; ListView honors objectName / fields / onRowClick / navigation."	deleted — the very next blockquote says "Do not guess props — read the contract … author against it, not from memory", and this sentence is a from-memory copy of that generated contract (-57 with its trailing clause)
UI-B-02 (split-exposed)	SKILL.md## Report Types	"a report visualizes via its embedded chart: config (see the example below)"	the parenthetical deleted — the report example is now in rules/dashboards.md (-6)
UI-B-02 (split-exposed)	rules/navigation.md Record Presentation	"for bespoke layout, assign a record Page (below)"	the parenthetical deleted — Pages is now rules/pages.md (-2)
UI-B-02 (split-exposed)	rules/pages.md page type list	"configured via interfaceConfig (see the iron rule above)"	the parenthetical deleted — the iron rule is now in rules/navigation.md (-6)

Gross -248. No decision-changing construct was deleted, and nothing was deleted to reach a number — condition (a) is satisfied by the split being byte-conserving, not by these.

Every "above" / "below" in the new tree was swept for a target that left its file; those three were the only ones, and each is deleted rather than rewritten — a dead pointer is removed text, not an occasion for new prose.

The one-topic-two-files cost, declared

check:skill-identifier-liveness Leg 2 binds three sections to skills/objectstack-ui/SKILL.mdby path (report-types, action-types, navigation-item-types); a bound heading that leaves its file yields [stale-binding] … A renamed heading unbinds the assertion, so it fails here rather than quietly checking nothing, pinned by that script's own self-test. Nothing in the script was changed. So three topics are split across two files:

• ## Report Types keeps its exhaustive table in the entry; ### Report Configuration moved to rules/dashboards.md (promoted to ##).
• ### Action Types keeps its table in the entry; the rest of Actions moved to rules/actions.md (headings shifted up one level).
• ### Navigation Item Types keeps its table in the entry; the App.create example moved to rules/navigation.md.

That is a real legibility cost of the split as constrained, and it is the reason those three entry sections look thin.

Link rewrites

Three links crossed a directory boundary and were rewritten, following the skills/objectstack-platform/rules/ precedent: #date-macros--filter-placeholders became ../SKILL.md#date-macros--filter-placeholders, and ./references/react-blocks.md / ./contracts/react-blocks.contract.json became ../references/… / ../contracts/….

Not in this PR

• UI-B-04 (the duplicated generated contracts/react-blocks.contract.json, 5,352 tokens) belongs to [Decision] Skills optimization program — batch 1 (4 items): new-file ceilings for splits · planned-eval stubs · react-blocks double rendering · published pm-dispatch scope #14296 item 3 and is the spec lane's seam card Seam (spec lane, ruled): gen:react-blocks stops emitting skills/objectstack-ui/contracts/react-blocks.contract.json — the markdown rendering becomes the single AI-facing output (#14296 item 3 = A) #14559. Generated output is untouched here.
• UI-H-01 (add an eval fixture) is retired: [Decision] Skills optimization program — batch 1 (4 items): new-file ceilings for splits · planned-eval stubs · react-blocks double rendering · published pm-dispatch scope #14296 item 2 = A rules that eval stubs are deleted rather than built out, so the proposal to add a fixture no longer stands. evals/ is untouched; the real fixture stays.

premise_false: none — no audit finding was skipped this round for a false premise.

Gates

Head d72e3dcc. Family re-derived on the final commit with node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack --commands (25 commands, from the script's own merge-base changeset — not a hand-written path list), and the whole union re-run on that head under scripts/pm/os-verify-lock.sh with OS_VERIFY_LOCK_SLOT=issue-14298-r2. Every exit code captured before any pipe. 31 ran, 30 exit 0.

The six beyond the derivation, each owed for a named reason: check-skills-token-ratchet.mjs --self-test (this PR edits that gate, so its own self-test is owed on top of the derived family — 64 cases pass), check:skill-examples and check:skill-refs (the dispatch named them; skills/** changed), check:doc-anchors and check:published-readme-links (links crossed a directory), check:nul-bytes (any edit).

The verdicts that answer this PR's specific risks, quoted from the gates themselves:

• check:skill-examples — "256 prose examples type-check across 3 surface(s) — every marked block parsed, so tsc ran the SEMANTIC pass on all of them". All 15 os:check markers travelled with their fences (0 in the entry, 15 across the five rules files; 15 before, 15 after) and the marked blocks still compile from their new files.
• check:skill-identifier-liveness — "Leg 1: 465 citation(s) over 46 published file(s) … Leg 2: 8 registered exhaustive section(s), 0 ledgered gap(s)". The three path-keyed ui bindings still resolve inside SKILL.md.
• check:published-readme-links — "176 outbound link(s) across 60 published markdown file(s) … 1 anchor(s) verified, 103/103 relative target(s) found". The new ../SKILL.md#date-macros--filter-placeholders and ../references/… links resolve.
• check:doc-anchors — "295 internal #fragment link(s) across 409 source file(s) all resolve to a real heading".
• check:skill-docs / check:skill-refs — both in sync; no generated file needed regenerating, and none was touched.
• check:role-word — green after the baseline re-point (its own commit; see below).
• node scripts/check-skills-token-ratchet.mjs — "36 authored bundle file(s) within their ceilings", every objectstack-ui row at exactly zero headroom.

check:test-completeness exited 3, which is NOT MEASURED, not a failure — the gate says so itself: "Nothing was measured: this gate exited before parsing a single summary line … ⛔ It is NOT a finding". Recorded as such rather than counted green.

check:role-word needed the baseline re-pointed, and that is a pure relocation. The baseline is keyed by PATH; the two occurrences it already carried for SKILL.md ("the tab-bar role", "semantic roles") moved into rules/list-views.md and rules/navigation.md as byte-identical lines. --update therefore produced a three-line diff — one row removed, two added, total unchanged at 2 — and re-baselined no other file. It is in its own commit. The count did not move, which is what the brief asked to be shown.

Repo-wide pnpm lint was not run; the narrowing is measured, not assumed. (1) The population comes from eslint's own config, not a guess: every files glob in eslint.config.mjs is {ts,tsx,mts,cts,js,jsx,mjs,cjs}, no markdown plugin is configured, and .md appears in no glob — so the six markdown files in this diff are outside eslint's population entirely. (2) Counted from --format json: of the eight changed files eslint claims exactly one, scripts/check-skills-token-ratchet.mjs, at 0 errors / 0 warnings; role-word-baseline.json reports "File ignored because no matching configuration was supplied". (3) Invariance for untouched files: the config configures no projectService and no project, so type-aware linting is off and nothing in this diff can move the verdict of a file it does not contain. CI's Lint job runs the full sweep regardless.

🤖 Generated with Claude Code

https://claude.ai/code/session_01LraLgQVGq8egUwfYZpbYt1

---

Generated by Claude Code

[os-litant]

Part of #14298

Ruling: "#14296 item 1 = A, comment 5507419465". Condition (a) was read at programme level by "#14568 = A, comment 5511939006" (director seat, verbatim reply 「同意」): the split PR itself must land token-neutral or negative, and the package arithmetic is carried by the programme.

The three conditions, with numbers

(a) Token-neutral or negative — measured -58.node scripts/check-skills-token-ratchet.mjs before and after, at head d72e3dcc:

	before	after
SKILL.md	22,856	3,815
rules/*.md (5 new files)	—	18,983
generated (contracts/react-blocks.contract.json, references/react-blocks.md, references/_index.md)	9,710	9,710
evals/	1,391	1,391
package	33,957	33,899

Net -58 (-0.17%). Gross deletions -248, additions +190 (five file titles, one ## Detailed Rules heading, five one-line pointers). Nothing in this PR is positive.

Programme account for the package: round 1 #14338-2,585 + this split -58 + seam card #14559-5,352 = -7,995 against the pre-programme 36,542 = -21.9%.

(b) Ceilings. Five rows added, each pinned AT its landed count so every new file starts with exactly zero headroom; the entry row re-locked to its landed count. No other row changed.

row	ceiling
skills/objectstack-ui/SKILL.md	25,445 → 3,815 (re-lock)
skills/objectstack-ui/rules/actions.md	2,108 (new)
skills/objectstack-ui/rules/dashboards.md	6,090 (new)
skills/objectstack-ui/rules/list-views.md	3,011 (new)
skills/objectstack-ui/rules/navigation.md	2,273 (new)
skills/objectstack-ui/rules/pages.md	5,501 (new)

(c) Not a channel for added prose — proved line by line, not asserted. Comparing the multiset of non-blank lines before (1,591) and after (1,586): 28 lines in the new tree are not lines of the old SKILL.md, and not one of them is a new sentence — 5 file titles, 1## Detailed Rules heading, 5 one-line pointers, 10 headings whose only change is a level shift or a deletion inside the heading text, and 7 lines whose only change is a rewritten relative link or a deleted dead clause. That is the entire set of text this PR authored. The check is reproducible from the diff.

File map

old section (spans at the merge base)	new home	tokens
## Configuring a List View · ## Configuring Kanban Views · ## Configuring Gantt Views	rules/list-views.md	3,011
## App Navigation (the App.create example) · ## Three Run Modes · ## Record Presentation	rules/navigation.md	2,273
## Dashboards · ## Dashboards (cont.) · ### Report Configuration · ## Analytics Cubes	rules/dashboards.md	6,090
## Pages — Lightning-Style Page Layouts · ## Docs — Package Documentation	rules/pages.md	5,501
## Actions (body below ### Action Types)	rules/actions.md	2,108
frontmatter · ## View Types · ## Detailed Rules · ## App Navigation head · ## Report Types · ## CRM UI Blueprint · ## Date Macros · ## Actions head · ## Common Pitfalls · ## Verify your work · ## References	stays in SKILL.md	3,815

Five files, not the audit's suggested nine — the brief says fewer is better, and two of the nine were folded deliberately: rules/filter-tokens.md was not created because ## Date Macros is cross-cutting (list views, dashboards, reports, pages) and is the target of the package's only in-file anchors, so it stays in the entry; rules/master-detail.md was not created because ### Master-Detail Forms is a routing ladder ("you almost never need a custom page or form config — prefer, in order"), which is exactly the kind of construct the entry keeps.

Dashboards now have one home. The two sections that sat 780 lines apart are adjacent in rules/dashboards.md, the second demoted to ### KPI Widgets, Filters, Drilldown under the first — which is what makes the three UI-B-02 deletions below dead text rather than a judgement call.

Deletions, per finding

finding	落点	before	after
UI-B-02	SKILL.md:554-558	"See the Production Pattern section below for the full Dashboard shape with refreshInterval, header actions, date range, global filters, widget options, and the period-over-period (compareTo) modifier; date bucketing comes from the bound dataset dimension's dateGranularity (ADR-0021)."	deleted — a forward reference to a section that is now the next thing on the page (-75)
UI-B-02	SKILL.md:1316-1317	"Dashboards (Dashboard) are first-class metadata. Beyond the basic widget layout shown above, the production-grade pattern uses:"	deleted — a second introduction to a section that no longer restarts (-33)
UI-B-02	SKILL.md:1381-1384	the "Tokens in filters" blockquote, pointing at Date Macros for the second time in the same file	deleted — the merged file carries that pointer once (-69)
UI-B-05 residue	SKILL.md:1029-1032	"Real component props/callbacks flow through — e.g. ObjectForm honors objectName / mode / recordId / formType / onSuccess / onCancel; ListView honors objectName / fields / onRowClick / navigation."	deleted — the very next blockquote says "Do not guess props — read the contract … author against it, not from memory", and this sentence is a from-memory copy of that generated contract (-57 with its trailing clause)
UI-B-02 (split-exposed)	SKILL.md## Report Types	"a report visualizes via its embedded chart: config (see the example below)"	the parenthetical deleted — the report example is now in rules/dashboards.md (-6)
UI-B-02 (split-exposed)	rules/navigation.md Record Presentation	"for bespoke layout, assign a record Page (below)"	the parenthetical deleted — Pages is now rules/pages.md (-2)
UI-B-02 (split-exposed)	rules/pages.md page type list	"configured via interfaceConfig (see the iron rule above)"	the parenthetical deleted — the iron rule is now in rules/navigation.md (-6)

Gross -248. No decision-changing construct was deleted, and nothing was deleted to reach a number — condition (a) is satisfied by the split being byte-conserving, not by these.

Every "above" / "below" in the new tree was swept for a target that left its file; those three were the only ones, and each is deleted rather than rewritten — a dead pointer is removed text, not an occasion for new prose.

The one-topic-two-files cost, declared

check:skill-identifier-liveness Leg 2 binds three sections to skills/objectstack-ui/SKILL.mdby path (report-types, action-types, navigation-item-types); a bound heading that leaves its file yields [stale-binding] … A renamed heading unbinds the assertion, so it fails here rather than quietly checking nothing, pinned by that script's own self-test. Nothing in the script was changed. So three topics are split across two files:

• ## Report Types keeps its exhaustive table in the entry; ### Report Configuration moved to rules/dashboards.md (promoted to ##).
• ### Action Types keeps its table in the entry; the rest of Actions moved to rules/actions.md (headings shifted up one level).
• ### Navigation Item Types keeps its table in the entry; the App.create example moved to rules/navigation.md.

That is a real legibility cost of the split as constrained, and it is the reason those three entry sections look thin.

Link rewrites

Three links crossed a directory boundary and were rewritten, following the skills/objectstack-platform/rules/ precedent: #date-macros--filter-placeholders became ../SKILL.md#date-macros--filter-placeholders, and ./references/react-blocks.md / ./contracts/react-blocks.contract.json became ../references/… / ../contracts/….

Not in this PR

• UI-B-04 (the duplicated generated contracts/react-blocks.contract.json, 5,352 tokens) belongs to [Decision] Skills optimization program — batch 1 (4 items): new-file ceilings for splits · planned-eval stubs · react-blocks double rendering · published pm-dispatch scope #14296 item 3 and is the spec lane's seam card Seam (spec lane, ruled): gen:react-blocks stops emitting skills/objectstack-ui/contracts/react-blocks.contract.json — the markdown rendering becomes the single AI-facing output (#14296 item 3 = A) #14559. Generated output is untouched here.
• UI-H-01 (add an eval fixture) is retired: [Decision] Skills optimization program — batch 1 (4 items): new-file ceilings for splits · planned-eval stubs · react-blocks double rendering · published pm-dispatch scope #14296 item 2 = A rules that eval stubs are deleted rather than built out, so the proposal to add a fixture no longer stands. evals/ is untouched; the real fixture stays.

premise_false: none — no audit finding was skipped this round for a false premise.

Gates

Head d72e3dcc. Family re-derived on the final commit with node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack --commands (25 commands, from the script's own merge-base changeset — not a hand-written path list), and the whole union re-run on that head under scripts/pm/os-verify-lock.sh with OS_VERIFY_LOCK_SLOT=issue-14298-r2. Every exit code captured before any pipe. 31 ran, 30 exit 0.

The six beyond the derivation, each owed for a named reason: check-skills-token-ratchet.mjs --self-test (this PR edits that gate, so its own self-test is owed on top of the derived family — 64 cases pass), check:skill-examples and check:skill-refs (the dispatch named them; skills/** changed), check:doc-anchors and check:published-readme-links (links crossed a directory), check:nul-bytes (any edit).

The verdicts that answer this PR's specific risks, quoted from the gates themselves:

• check:skill-examples — "256 prose examples type-check across 3 surface(s) — every marked block parsed, so tsc ran the SEMANTIC pass on all of them". All 15 os:check markers travelled with their fences (0 in the entry, 15 across the five rules files; 15 before, 15 after) and the marked blocks still compile from their new files.
• check:skill-identifier-liveness — "Leg 1: 465 citation(s) over 46 published file(s) … Leg 2: 8 registered exhaustive section(s), 0 ledgered gap(s)". The three path-keyed ui bindings still resolve inside SKILL.md.
• check:published-readme-links — "176 outbound link(s) across 60 published markdown file(s) … 1 anchor(s) verified, 103/103 relative target(s) found". The new ../SKILL.md#date-macros--filter-placeholders and ../references/… links resolve.
• check:doc-anchors — "295 internal #fragment link(s) across 409 source file(s) all resolve to a real heading".
• check:skill-docs / check:skill-refs — both in sync; no generated file needed regenerating, and none was touched.
• check:role-word — green after the baseline re-point (its own commit; see below).
• node scripts/check-skills-token-ratchet.mjs — "36 authored bundle file(s) within their ceilings", every objectstack-ui row at exactly zero headroom.

check:test-completeness exited 3, which is NOT MEASURED, not a failure — the gate says so itself: "Nothing was measured: this gate exited before parsing a single summary line … ⛔ It is NOT a finding". Recorded as such rather than counted green.

check:role-word needed the baseline re-pointed, and that is a pure relocation. The baseline is keyed by PATH; the two occurrences it already carried for SKILL.md ("the tab-bar role", "semantic roles") moved into rules/list-views.md and rules/navigation.md as byte-identical lines. --update therefore produced a three-line diff — one row removed, two added, total unchanged at 2 — and re-baselined no other file. It is in its own commit. The count did not move, which is what the brief asked to be shown.

Repo-wide pnpm lint was not run; the narrowing is measured, not assumed. (1) The population comes from eslint's own config, not a guess: every files glob in eslint.config.mjs is {ts,tsx,mts,cts,js,jsx,mjs,cjs}, no markdown plugin is configured, and .md appears in no glob — so the six markdown files in this diff are outside eslint's population entirely. (2) Counted from --format json: of the eight changed files eslint claims exactly one, scripts/check-skills-token-ratchet.mjs, at 0 errors / 0 warnings; role-word-baseline.json reports "File ignored because no matching configuration was supplied". (3) Invariance for untouched files: the config configures no projectService and no project, so type-aware linting is off and nothing in this diff can move the verdict of a file it does not contain. CI's Lint job runs the full sweep regardless.

🤖 Generated with Claude Code

https://claude.ai/code/session_01LraLgQVGq8egUwfYZpbYt1

---

Generated by Claude Code

[os-litant]

Part of #14298

Ruling: "#14296 item 1 = A, comment 5507419465". Condition (a) was read at programme level by "#14568 = A, comment 5511939006" (director seat, verbatim reply 「同意」): the split PR itself must land token-neutral or negative, and the package arithmetic is carried by the programme.

The three conditions, with numbers

(a) Token-neutral or negative — measured -58.node scripts/check-skills-token-ratchet.mjs before and after, at head d72e3dcc:

	before	after
SKILL.md	22,856	3,815
rules/*.md (5 new files)	—	18,983
generated (contracts/react-blocks.contract.json, references/react-blocks.md, references/_index.md)	9,710	9,710
evals/	1,391	1,391
package	33,957	33,899

Net -58 (-0.17%). Gross deletions -248, additions +190 (five file titles, one ## Detailed Rules heading, five one-line pointers). Nothing in this PR is positive.

Programme account for the package: round 1 #14338-2,585 + this split -58 + seam card #14559-5,352 = -7,995 against the pre-programme 36,542 = -21.9%.

(b) Ceilings. Five rows added, each pinned AT its landed count so every new file starts with exactly zero headroom; the entry row re-locked to its landed count. No other row changed.

row	ceiling
skills/objectstack-ui/SKILL.md	25,445 → 3,815 (re-lock)
skills/objectstack-ui/rules/actions.md	2,108 (new)
skills/objectstack-ui/rules/dashboards.md	6,090 (new)
skills/objectstack-ui/rules/list-views.md	3,011 (new)
skills/objectstack-ui/rules/navigation.md	2,273 (new)
skills/objectstack-ui/rules/pages.md	5,501 (new)

(c) Not a channel for added prose — proved line by line, not asserted. Comparing the multiset of non-blank lines before (1,591) and after (1,586): 28 lines in the new tree are not lines of the old SKILL.md, and not one of them is a new sentence — 5 file titles, 1## Detailed Rules heading, 5 one-line pointers, 10 headings whose only change is a level shift or a deletion inside the heading text, and 7 lines whose only change is a rewritten relative link or a deleted dead clause. That is the entire set of text this PR authored. The check is reproducible from the diff.

File map

old section (spans at the merge base)	new home	tokens
## Configuring a List View · ## Configuring Kanban Views · ## Configuring Gantt Views	rules/list-views.md	3,011
## App Navigation (the App.create example) · ## Three Run Modes · ## Record Presentation	rules/navigation.md	2,273
## Dashboards · ## Dashboards (cont.) · ### Report Configuration · ## Analytics Cubes	rules/dashboards.md	6,090
## Pages — Lightning-Style Page Layouts · ## Docs — Package Documentation	rules/pages.md	5,501
## Actions (body below ### Action Types)	rules/actions.md	2,108
frontmatter · ## View Types · ## Detailed Rules · ## App Navigation head · ## Report Types · ## CRM UI Blueprint · ## Date Macros · ## Actions head · ## Common Pitfalls · ## Verify your work · ## References	stays in SKILL.md	3,815

Five files, not the audit's suggested nine — the brief says fewer is better, and two of the nine were folded deliberately: rules/filter-tokens.md was not created because ## Date Macros is cross-cutting (list views, dashboards, reports, pages) and is the target of the package's only in-file anchors, so it stays in the entry; rules/master-detail.md was not created because ### Master-Detail Forms is a routing ladder ("you almost never need a custom page or form config — prefer, in order"), which is exactly the kind of construct the entry keeps.

Dashboards now have one home. The two sections that sat 780 lines apart are adjacent in rules/dashboards.md, the second demoted to ### KPI Widgets, Filters, Drilldown under the first — which is what makes the three UI-B-02 deletions below dead text rather than a judgement call.

Deletions, per finding

finding	落点	before	after
UI-B-02	SKILL.md:554-558	"See the Production Pattern section below for the full Dashboard shape with refreshInterval, header actions, date range, global filters, widget options, and the period-over-period (compareTo) modifier; date bucketing comes from the bound dataset dimension's dateGranularity (ADR-0021)."	deleted — a forward reference to a section that is now the next thing on the page (-75)
UI-B-02	SKILL.md:1316-1317	"Dashboards (Dashboard) are first-class metadata. Beyond the basic widget layout shown above, the production-grade pattern uses:"	deleted — a second introduction to a section that no longer restarts (-33)
UI-B-02	SKILL.md:1381-1384	the "Tokens in filters" blockquote, pointing at Date Macros for the second time in the same file	deleted — the merged file carries that pointer once (-69)
UI-B-05 residue	SKILL.md:1029-1032	"Real component props/callbacks flow through — e.g. ObjectForm honors objectName / mode / recordId / formType / onSuccess / onCancel; ListView honors objectName / fields / onRowClick / navigation."	deleted — the very next blockquote says "Do not guess props — read the contract … author against it, not from memory", and this sentence is a from-memory copy of that generated contract (-57 with its trailing clause)
UI-B-02 (split-exposed)	SKILL.md## Report Types	"a report visualizes via its embedded chart: config (see the example below)"	the parenthetical deleted — the report example is now in rules/dashboards.md (-6)
UI-B-02 (split-exposed)	rules/navigation.md Record Presentation	"for bespoke layout, assign a record Page (below)"	the parenthetical deleted — Pages is now rules/pages.md (-2)
UI-B-02 (split-exposed)	rules/pages.md page type list	"configured via interfaceConfig (see the iron rule above)"	the parenthetical deleted — the iron rule is now in rules/navigation.md (-6)

Gross -248. No decision-changing construct was deleted, and nothing was deleted to reach a number — condition (a) is satisfied by the split being byte-conserving, not by these.

Every "above" / "below" in the new tree was swept for a target that left its file; those three were the only ones, and each is deleted rather than rewritten — a dead pointer is removed text, not an occasion for new prose.

The one-topic-two-files cost, declared

check:skill-identifier-liveness Leg 2 binds three sections to skills/objectstack-ui/SKILL.mdby path (report-types, action-types, navigation-item-types); a bound heading that leaves its file yields [stale-binding] … A renamed heading unbinds the assertion, so it fails here rather than quietly checking nothing, pinned by that script's own self-test. Nothing in the script was changed. So three topics are split across two files:

• ## Report Types keeps its exhaustive table in the entry; ### Report Configuration moved to rules/dashboards.md (promoted to ##).
• ### Action Types keeps its table in the entry; the rest of Actions moved to rules/actions.md (headings shifted up one level).
• ### Navigation Item Types keeps its table in the entry; the App.create example moved to rules/navigation.md.

That is a real legibility cost of the split as constrained, and it is the reason those three entry sections look thin.

Link rewrites

Three links crossed a directory boundary and were rewritten, following the skills/objectstack-platform/rules/ precedent: #date-macros--filter-placeholders became ../SKILL.md#date-macros--filter-placeholders, and ./references/react-blocks.md / ./contracts/react-blocks.contract.json became ../references/… / ../contracts/….

Not in this PR

• UI-B-04 (the duplicated generated contracts/react-blocks.contract.json, 5,352 tokens) belongs to [Decision] Skills optimization program — batch 1 (4 items): new-file ceilings for splits · planned-eval stubs · react-blocks double rendering · published pm-dispatch scope #14296 item 3 and is the spec lane's seam card Seam (spec lane, ruled): gen:react-blocks stops emitting skills/objectstack-ui/contracts/react-blocks.contract.json — the markdown rendering becomes the single AI-facing output (#14296 item 3 = A) #14559. Generated output is untouched here.
• UI-H-01 (add an eval fixture) is retired: [Decision] Skills optimization program — batch 1 (4 items): new-file ceilings for splits · planned-eval stubs · react-blocks double rendering · published pm-dispatch scope #14296 item 2 = A rules that eval stubs are deleted rather than built out, so the proposal to add a fixture no longer stands. evals/ is untouched; the real fixture stays.

premise_false: none — no audit finding was skipped this round for a false premise.

Gates

Head d72e3dcc. Family re-derived on the final commit with node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack --commands (25 commands, from the script's own merge-base changeset — not a hand-written path list), and the whole union re-run on that head under scripts/pm/os-verify-lock.sh with OS_VERIFY_LOCK_SLOT=issue-14298-r2. Every exit code captured before any pipe. 31 ran, 30 exit 0.

The six beyond the derivation, each owed for a named reason: check-skills-token-ratchet.mjs --self-test (this PR edits that gate, so its own self-test is owed on top of the derived family — 64 cases pass), check:skill-examples and check:skill-refs (the dispatch named them; skills/** changed), check:doc-anchors and check:published-readme-links (links crossed a directory), check:nul-bytes (any edit).

The verdicts that answer this PR's specific risks, quoted from the gates themselves:

• check:skill-examples — "256 prose examples type-check across 3 surface(s) — every marked block parsed, so tsc ran the SEMANTIC pass on all of them". All 15 os:check markers travelled with their fences (0 in the entry, 15 across the five rules files; 15 before, 15 after) and the marked blocks still compile from their new files.
• check:skill-identifier-liveness — "Leg 1: 465 citation(s) over 46 published file(s) … Leg 2: 8 registered exhaustive section(s), 0 ledgered gap(s)". The three path-keyed ui bindings still resolve inside SKILL.md.
• check:published-readme-links — "176 outbound link(s) across 60 published markdown file(s) … 1 anchor(s) verified, 103/103 relative target(s) found". The new ../SKILL.md#date-macros--filter-placeholders and ../references/… links resolve.
• check:doc-anchors — "295 internal #fragment link(s) across 409 source file(s) all resolve to a real heading".
• check:skill-docs / check:skill-refs — both in sync; no generated file needed regenerating, and none was touched.
• check:role-word — green after the baseline re-point (its own commit; see below).
• node scripts/check-skills-token-ratchet.mjs — "36 authored bundle file(s) within their ceilings", every objectstack-ui row at exactly zero headroom.

check:test-completeness exited 3, which is NOT MEASURED, not a failure — the gate says so itself: "Nothing was measured: this gate exited before parsing a single summary line … ⛔ It is NOT a finding". Recorded as such rather than counted green.

check:role-word needed the baseline re-pointed, and that is a pure relocation. The baseline is keyed by PATH; the two occurrences it already carried for SKILL.md ("the tab-bar role", "semantic roles") moved into rules/list-views.md and rules/navigation.md as byte-identical lines. --update therefore produced a three-line diff — one row removed, two added, total unchanged at 2 — and re-baselined no other file. It is in its own commit. The count did not move, which is what the brief asked to be shown.

Repo-wide pnpm lint was not run; the narrowing is measured, not assumed. (1) The population comes from eslint's own config, not a guess: every files glob in eslint.config.mjs is {ts,tsx,mts,cts,js,jsx,mjs,cjs}, no markdown plugin is configured, and .md appears in no glob — so the six markdown files in this diff are outside eslint's population entirely. (2) Counted from --format json: of the eight changed files eslint claims exactly one, scripts/check-skills-token-ratchet.mjs, at 0 errors / 0 warnings; role-word-baseline.json reports "File ignored because no matching configuration was supplied". (3) Invariance for untouched files: the config configures no projectService and no project, so type-aware linting is off and nothing in this diff can move the verdict of a file it does not contain. CI's Lint job runs the full sweep regardless.

🤖 Generated with Claude Code

https://claude.ai/code/session_01LraLgQVGq8egUwfYZpbYt1

---

Generated by Claude Code

[os-litant]

Part of #14298

Ruling: "#14296 item 1 = A, comment 5507419465". Condition (a) was read at programme level by "#14568 = A, comment 5511939006" (director seat, verbatim reply 「同意」): the split PR itself must land token-neutral or negative, and the package arithmetic is carried by the programme.

The three conditions, with numbers

(a) Token-neutral or negative — measured -58.node scripts/check-skills-token-ratchet.mjs before and after, at head d72e3dcc:

	before	after
SKILL.md	22,856	3,815
rules/*.md (5 new files)	—	18,983
generated (contracts/react-blocks.contract.json, references/react-blocks.md, references/_index.md)	9,710	9,710
evals/	1,391	1,391
package	33,957	33,899

Net -58 (-0.17%). Gross deletions -248, additions +190 (five file titles, one ## Detailed Rules heading, five one-line pointers). Nothing in this PR is positive.

Programme account for the package: round 1 #14338-2,585 + this split -58 + seam card #14559-5,352 = -7,995 against the pre-programme 36,542 = -21.9%.

(b) Ceilings. Five rows added, each pinned AT its landed count so every new file starts with exactly zero headroom; the entry row re-locked to its landed count. No other row changed.

row	ceiling
skills/objectstack-ui/SKILL.md	25,445 → 3,815 (re-lock)
skills/objectstack-ui/rules/actions.md	2,108 (new)
skills/objectstack-ui/rules/dashboards.md	6,090 (new)
skills/objectstack-ui/rules/list-views.md	3,011 (new)
skills/objectstack-ui/rules/navigation.md	2,273 (new)
skills/objectstack-ui/rules/pages.md	5,501 (new)

(c) Not a channel for added prose — proved line by line, not asserted. Comparing the multiset of non-blank lines before (1,591) and after (1,586): 28 lines in the new tree are not lines of the old SKILL.md, and not one of them is a new sentence — 5 file titles, 1## Detailed Rules heading, 5 one-line pointers, 10 headings whose only change is a level shift or a deletion inside the heading text, and 7 lines whose only change is a rewritten relative link or a deleted dead clause. That is the entire set of text this PR authored. The check is reproducible from the diff.

File map

old section (spans at the merge base)	new home	tokens
## Configuring a List View · ## Configuring Kanban Views · ## Configuring Gantt Views	rules/list-views.md	3,011
## App Navigation (the App.create example) · ## Three Run Modes · ## Record Presentation	rules/navigation.md	2,273
## Dashboards · ## Dashboards (cont.) · ### Report Configuration · ## Analytics Cubes	rules/dashboards.md	6,090
## Pages — Lightning-Style Page Layouts · ## Docs — Package Documentation	rules/pages.md	5,501
## Actions (body below ### Action Types)	rules/actions.md	2,108
frontmatter · ## View Types · ## Detailed Rules · ## App Navigation head · ## Report Types · ## CRM UI Blueprint · ## Date Macros · ## Actions head · ## Common Pitfalls · ## Verify your work · ## References	stays in SKILL.md	3,815

Five files, not the audit's suggested nine — the brief says fewer is better, and two of the nine were folded deliberately: rules/filter-tokens.md was not created because ## Date Macros is cross-cutting (list views, dashboards, reports, pages) and is the target of the package's only in-file anchors, so it stays in the entry; rules/master-detail.md was not created because ### Master-Detail Forms is a routing ladder ("you almost never need a custom page or form config — prefer, in order"), which is exactly the kind of construct the entry keeps.

Dashboards now have one home. The two sections that sat 780 lines apart are adjacent in rules/dashboards.md, the second demoted to ### KPI Widgets, Filters, Drilldown under the first — which is what makes the three UI-B-02 deletions below dead text rather than a judgement call.

Deletions, per finding

finding	落点	before	after
UI-B-02	SKILL.md:554-558	"See the Production Pattern section below for the full Dashboard shape with refreshInterval, header actions, date range, global filters, widget options, and the period-over-period (compareTo) modifier; date bucketing comes from the bound dataset dimension's dateGranularity (ADR-0021)."	deleted — a forward reference to a section that is now the next thing on the page (-75)
UI-B-02	SKILL.md:1316-1317	"Dashboards (Dashboard) are first-class metadata. Beyond the basic widget layout shown above, the production-grade pattern uses:"	deleted — a second introduction to a section that no longer restarts (-33)
UI-B-02	SKILL.md:1381-1384	the "Tokens in filters" blockquote, pointing at Date Macros for the second time in the same file	deleted — the merged file carries that pointer once (-69)
UI-B-05 residue	SKILL.md:1029-1032	"Real component props/callbacks flow through — e.g. ObjectForm honors objectName / mode / recordId / formType / onSuccess / onCancel; ListView honors objectName / fields / onRowClick / navigation."	deleted — the very next blockquote says "Do not guess props — read the contract … author against it, not from memory", and this sentence is a from-memory copy of that generated contract (-57 with its trailing clause)
UI-B-02 (split-exposed)	SKILL.md## Report Types	"a report visualizes via its embedded chart: config (see the example below)"	the parenthetical deleted — the report example is now in rules/dashboards.md (-6)
UI-B-02 (split-exposed)	rules/navigation.md Record Presentation	"for bespoke layout, assign a record Page (below)"	the parenthetical deleted — Pages is now rules/pages.md (-2)
UI-B-02 (split-exposed)	rules/pages.md page type list	"configured via interfaceConfig (see the iron rule above)"	the parenthetical deleted — the iron rule is now in rules/navigation.md (-6)

Gross -248. No decision-changing construct was deleted, and nothing was deleted to reach a number — condition (a) is satisfied by the split being byte-conserving, not by these.

Every "above" / "below" in the new tree was swept for a target that left its file; those three were the only ones, and each is deleted rather than rewritten — a dead pointer is removed text, not an occasion for new prose.

The one-topic-two-files cost, declared

check:skill-identifier-liveness Leg 2 binds three sections to skills/objectstack-ui/SKILL.mdby path (report-types, action-types, navigation-item-types); a bound heading that leaves its file yields [stale-binding] … A renamed heading unbinds the assertion, so it fails here rather than quietly checking nothing, pinned by that script's own self-test. Nothing in the script was changed. So three topics are split across two files:

• ## Report Types keeps its exhaustive table in the entry; ### Report Configuration moved to rules/dashboards.md (promoted to ##).
• ### Action Types keeps its table in the entry; the rest of Actions moved to rules/actions.md (headings shifted up one level).
• ### Navigation Item Types keeps its table in the entry; the App.create example moved to rules/navigation.md.

That is a real legibility cost of the split as constrained, and it is the reason those three entry sections look thin.

Link rewrites

Three links crossed a directory boundary and were rewritten, following the skills/objectstack-platform/rules/ precedent: #date-macros--filter-placeholders became ../SKILL.md#date-macros--filter-placeholders, and ./references/react-blocks.md / ./contracts/react-blocks.contract.json became ../references/… / ../contracts/….

Not in this PR

• UI-B-04 (the duplicated generated contracts/react-blocks.contract.json, 5,352 tokens) belongs to [Decision] Skills optimization program — batch 1 (4 items): new-file ceilings for splits · planned-eval stubs · react-blocks double rendering · published pm-dispatch scope #14296 item 3 and is the spec lane's seam card Seam (spec lane, ruled): gen:react-blocks stops emitting skills/objectstack-ui/contracts/react-blocks.contract.json — the markdown rendering becomes the single AI-facing output (#14296 item 3 = A) #14559. Generated output is untouched here.
• UI-H-01 (add an eval fixture) is retired: [Decision] Skills optimization program — batch 1 (4 items): new-file ceilings for splits · planned-eval stubs · react-blocks double rendering · published pm-dispatch scope #14296 item 2 = A rules that eval stubs are deleted rather than built out, so the proposal to add a fixture no longer stands. evals/ is untouched; the real fixture stays.

premise_false: none — no audit finding was skipped this round for a false premise.

Gates

Head d72e3dcc. Family re-derived on the final commit with node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack --commands (25 commands, from the script's own merge-base changeset — not a hand-written path list), and the whole union re-run on that head under scripts/pm/os-verify-lock.sh with OS_VERIFY_LOCK_SLOT=issue-14298-r2. Every exit code captured before any pipe. 31 ran, 30 exit 0.

The six beyond the derivation, each owed for a named reason: check-skills-token-ratchet.mjs --self-test (this PR edits that gate, so its own self-test is owed on top of the derived family — 64 cases pass), check:skill-examples and check:skill-refs (the dispatch named them; skills/** changed), check:doc-anchors and check:published-readme-links (links crossed a directory), check:nul-bytes (any edit).

The verdicts that answer this PR's specific risks, quoted from the gates themselves:

• check:skill-examples — "256 prose examples type-check across 3 surface(s) — every marked block parsed, so tsc ran the SEMANTIC pass on all of them". All 15 os:check markers travelled with their fences (0 in the entry, 15 across the five rules files; 15 before, 15 after) and the marked blocks still compile from their new files.
• check:skill-identifier-liveness — "Leg 1: 465 citation(s) over 46 published file(s) … Leg 2: 8 registered exhaustive section(s), 0 ledgered gap(s)". The three path-keyed ui bindings still resolve inside SKILL.md.
• check:published-readme-links — "176 outbound link(s) across 60 published markdown file(s) … 1 anchor(s) verified, 103/103 relative target(s) found". The new ../SKILL.md#date-macros--filter-placeholders and ../references/… links resolve.
• check:doc-anchors — "295 internal #fragment link(s) across 409 source file(s) all resolve to a real heading".
• check:skill-docs / check:skill-refs — both in sync; no generated file needed regenerating, and none was touched.
• check:role-word — green after the baseline re-point (its own commit; see below).
• node scripts/check-skills-token-ratchet.mjs — "36 authored bundle file(s) within their ceilings", every objectstack-ui row at exactly zero headroom.

check:test-completeness exited 3, which is NOT MEASURED, not a failure — the gate says so itself: "Nothing was measured: this gate exited before parsing a single summary line … ⛔ It is NOT a finding". Recorded as such rather than counted green.

check:role-word needed the baseline re-pointed, and that is a pure relocation. The baseline is keyed by PATH; the two occurrences it already carried for SKILL.md ("the tab-bar role", "semantic roles") moved into rules/list-views.md and rules/navigation.md as byte-identical lines. --update therefore produced a three-line diff — one row removed, two added, total unchanged at 2 — and re-baselined no other file. It is in its own commit. The count did not move, which is what the brief asked to be shown.

Repo-wide pnpm lint was not run; the narrowing is measured, not assumed. (1) The population comes from eslint's own config, not a guess: every files glob in eslint.config.mjs is {ts,tsx,mts,cts,js,jsx,mjs,cjs}, no markdown plugin is configured, and .md appears in no glob — so the six markdown files in this diff are outside eslint's population entirely. (2) Counted from --format json: of the eight changed files eslint claims exactly one, scripts/check-skills-token-ratchet.mjs, at 0 errors / 0 warnings; role-word-baseline.json reports "File ignored because no matching configuration was supplied". (3) Invariance for untouched files: the config configures no projectService and no project, so type-aware linting is off and nothing in this diff can move the verdict of a file it does not contain. CI's Lint job runs the full sweep regardless.

🤖 Generated with Claude Code

https://claude.ai/code/session_01LraLgQVGq8egUwfYZpbYt1

---

Generated by Claude Code