D3: retire compound-name addressing — remove the mounted compound arities, unify SDK URL spelling, close out the five broken methods (#12176 stage 3)

[os-trump]

Blocked-by discharged 2026-08-25 — #12194 is CLOSED: D1 landed as PR #12244 (merge commit 311433f6 on main). No legal item name can contain / any more, so this stage is now the dead-code removal the ruling planned. Released to dispatch and claimed in the same stroke by PM session session_01H9StxQgG2DPA26XzZZqnJB.

Stage 3 of #12176's maintainer-ruled retirement (provenance on #12176 and #12194). ⛔ Not dispatchable until D1 (#12194) lands — after D1, no legal item name can contain /, which is what makes this stage dead-code removal rather than a behavior break.

Ruled scope (execute on the post-D1 ref; re-verify counts against #12176's census report)

1. Retire the mounted compound arities in packages/rest (census: 3 mounted — the getItem/saveItem twins and the /published read; ledger rows and ordering pins go with them; packages/runtime/src/domains/meta.ts folds likewise). ⚠️ Census correction to carry: at the compound arity the :type segment holds an OBJECT name, not a metadata type — read Ban / in metadata item names — retire compound-name addressing (maintainer-ruled direction): census, migration surface, staged plan #12176's report before assuming a route's semantics.
2. Unify SDK URL spelling in packages/client: with slashes illegal, the 6-unencoded/4-encoded split is moot — pick one spelling (encoding a slash-free name is a no-op; keeping encodeURIComponent everywhere is the defensive default), update the doc comments that promise "compound names pass through unencoded" (index.ts:595, :986 region), and delete/rework the tests pinning unencoded compound URLs (client.test.ts:325-334, meta-automation-descriptors.test.ts:30, the protocol tests Ban / in metadata item names — retire compound-name addressing (maintainer-ruled direction): census, migration surface, staged plan #12176's census lists — 26 test files measured; re-derive on the current ref).
3. Close out the parked family: Four SDK metadata methods build compound-name URLs against routes mounted in one arity only — measured 404, and getHistory/deleteItem encode while getAudit/getReferences/diffItem/rollbackItem do not #12106 and A compound-name draft has no REST promotion door — POST /meta/:type/:name/publish is mounted in one arity while its read twin /published is mounted in both #11932 are pm:blocked on Ban / in metadata item names — retire compound-name addressing (maintainer-ruled direction): census, migration surface, staged plan #12176 — this stage's landing is what resolves them; the landing PR references both (⛔ Part of, closure is the PM's re-verification call).

Clause-②: yes — public routes are removed (contract surface change; post-D1 they are unreachable by any legal name, but the removal is still a published-surface move); contract-review tier at dispatch.

Refs: #12176 (census: 5 source files, 3 routes, 26 test files, 2 docs pages) · #12194 (D1) · #12106 · #11932 · closed PR #12105 (its shadowing measurements and capability-enumeration table are the reference material for un-mounting safely).

[os-trump]

Claim: PM direct-dispatch (the #12176 program — maintainer 2026-08-25, verbatim: 「斜杠不该存在，现在就立卡，你负责派发，然后给我结果」)
Session: session_01H9StxQgG2DPA26XzZZqnJB
Branch: claude/issue-12195-retire-compound-arities
Worktree: objectstack-issue-12195
Domain: domain:cli
File surface: packages/rest/src/rest-server.ts (the 3 mounted compound arities + their ledger rows + ordering pins) + packages/runtime/src/domains/meta.ts (compound folds) + packages/client/src/index.ts (URL-spelling unification + the two docblock regions) + the census's test files re-derived on the post-D1 ref + changeset(s). ⛔ packages/spec is read-only for this card — D1 landed the grammar there; a spec touch STOPS the card.
Container & model: mode:subagent under the PM session. Clause-②: yes (public routes removed). Per the contract-review chain rules, a mode:subagent clause-② dispatch keeps needs:contract-review regardless of the parent's tier — hung on this card in the same stroke as the claim; ⛔ hang it on the PR the moment the PR exists (dual carrier — the #12198 half-state is the lesson). The chain reviews and lands; the dev never flips ready or arms auto-merge.

Serial constraints: rest-server.ts is free — PR #12280 (#11718) merged; no open PR holds the file (verified by queue/PR state, not by assumption). This claim now holds it; the lane's frozen order (#11473 → #11719 → #11735 → #11926) stays queued behind this card, one at a time, per the standing fence discipline.

Blocked-by discharged in the body in this same stroke: #12194 closed via PR #12244 (311433f6), so every compound name is now refused at the publish door before it could ever be stored — the three mounted arities are reachable only by names that can no longer exist, which is what makes this removal dead-code cleanup by construction rather than a capability break. PR #12105 (closed unmerged, by ruling) carries the shadowing measurements and capability-enumeration table to un-mount against.

---

Generated by Claude Code

[os-trump]

os-dev-report

{
 "issue": 12195,
 "status": "done",
 "branch": "claude/issue-12195-retire-compound-arities",
 "pr": "https://github.com/objectstack-ai/objectstack/pull/12378",
 "premise_still_valid": true,
 "summary": "D3 landed as draft PR #12378 (draft, needs:contract-review hung on it and confirmed by read-back alongside the size-labeler's own labels). All three ruled limbs done. (1) The 3 mounted compound arities are un-mounted in packages/rest — GET and PUT /meta/:type/:section/:name and GET .../:section/:name/published — with their 3 route-ledger rows and the ordering pins that existed only because the three-segment compound route was a CATCH-ALL shadowing /history, /audit, /diff, /published and /layers; the matching folds in packages/runtime/src/domains/meta.ts go with them (item branch now requires exactly 2 segments, /published exactly 3, instead of re-joining every trailing segment). (2) packages/client unifies on encodeURIComponent for every /meta item address, and the two docblock regions promising unencoded compound pass-through are corrected. (3) The PR references #12106 and #11932 as 'Part of', never 'Fixes'; first body line is 'Fixes #12195'. packages/spec untouched — no spec fork arose. FOUR CORRECTIONS TO THE BRIEF, all measured, none reconciled to the census. (a) The SDK split re-derived on this ref is 10 unencoded / 4 encoded, NOT the 6/4 the card carried: 8 unscoped `${route}/${type}/${name}` sites plus 2 scoped `/meta/${type}/${name}` sites were unencoded, with deleteItem and getHistory already encoded on both clients; all 14 now encode. (b) The card's census listed 26 test files; re-derived by anchored grep the raw population is 37, of which 8 are the page:section component-type false positives the census itself flagged — but the reliable instrument turned out to be the suites, not grep: the removal turned 14 rest files / 80 tests and 6 runtime files / 10 tests and 2 client files / 4 tests red, and that 22-file set is what was actually reworked. (c) A CONSTRAINT THE BRIEF DID NOT CARRY and that shaped the design: #12194's landed migration entry states as acceptance criteria that 'reads and deleteMetaItem still answer for pre-grammar residue rows, so any stored junk name remains listable and clearable'. Un-mounting alone would have broken that on the dispatcher transport, because packages/runtime splits the RAW path and — unlike the Hono routes — nothing decodes its parameters (measured: c.req.path returns /meta/lead/views%2Fall_leads verbatim while c.req.param('name') on the same request yields views/all_leads). So the un-fold ships with decodeMetaNameSegment on the :name segment; the sibling domains/packages.ts already decodes its own id segments the same way, so the meta domain was the outlier, and this is the one addition beyond pure removal. With it, %2F reaches the single-segment door on BOTH transports and decodes back to the stored spelling — the capability is re-expressed, not removed, and meta-published-overlay.test.ts pins it end-to-end on a stored views/all_leads row. (d) The dispatcher's `>= 2` fold was also swallowing paths that were never compound names at all — /metadata/object/foo/references folded to the item name 'foo/references' and was answered as a metadata READ of a name nothing stores. Requiring exactly two segments ends that too; visible in meta-state-plural-tolerance.test.ts, where /meta/objectss/task/state/status moves from RESOURCE_NOT_FOUND (a claim about an item) to ROUTE_NOT_FOUND (a claim about the route), the accurate one. One consequential judgment call, flagged for review rather than buried: the meta domain's tail `return { handled: false }` was UNREACHABLE dead code before this card (0, 1 and — through the fold — every 2+ segment path were all covered), and un-folding makes it reachable for the first time. Left as-is it would have handed the adapter an anonymous 404 'Not Found' on exactly the shape this retirement newly produces, so it became a LOCATED deps.routeNotFound — the same form domains/ai.ts and domains/share-links.ts already use. Minor observation, no issue filed because it is a no-op with no compound dimension: meta.getItems(type) and a few sibling /meta sites still interpolate a bare `${type}` single segment; type is always canonical snake_case there, so encoding it would change no bytes.",
 "tests": "All runs at c6c3d98d (git rev-parse --short HEAD; working tree clean). Heavy runs through scripts/pm/os-verify-lock.sh; every exit code captured BEFORE any pipe (redirect-then-capture), and every gate quoted by its OWN verdict line, never a bare $?. SUITES, 'os-verify-lock: VERDICT command-exit 0 · held the lock 347s (5m47s)': packages/rest 'Test Files 147 passed (147) / Tests 2328 passed (2328)'; packages/runtime 'Test Files 191 passed (191) / Tests 2815 passed (2815)'; packages/client 'Test Files 25 passed (25) / Tests 346 passed (346)'. TYPECHECK all three: 'VERDICT command-exit 0 · held the lock 32s', 'packages/rest typecheck: Done', 'packages/runtime typecheck: Done', 'packages/client typecheck: Done' plus 'check:test-typecheck: OK — 0 file(s) / 0 error(s) held in test-typecheck-debt.json'. RED-BEFORE: the removal was made first and the existing pins run against it — 80 rest tests, 10 runtime tests and 4 client tests went red, every one a pin naming a retired arity, and the client pair failed with precisely the predicted diff (views/all_leads -> views%2Fall_leads). ABLATION, direction predicted IN WRITING before running and recorded in the run log: restoring the pre-removal rest-server.ts from the FIXED base sha 24910641 (never origin/main — a shared moving ref) re-mounts the arities, so the new ABSENCE pins must go RED with a NON-EMPTY actual containing the compound route keys. Mutation CONFIRMED ON DISK before reading anything — ':section' occurrences 7 -> 10 and 2 compound `path:` registrations present (an editing tool's exit code is not evidence). Observed: 'Test Files 3 failed (3) / Tests 6 failed | 33 passed (39)', actual values exactly as predicted, e.g. 'expected [ Array(1) ] to deeply equal []' with '+ [\"GET /api/v1/meta/:type/:section/:name/published\"]'. No rebuild leg was owed: the ablated subject is imported relatively ('./rest-server.js') inside its own package, so vitest reads src — not a dist-resolved dependency. Script carried trap restore EXIT INT TERM; restoration proven BYTE-IDENTICAL rather than by --stat: git status --porcelain empty, git diff HEAD empty, git diff --cached empty, and git hash-object == git rev-parse HEAD:<path> for both ablated files (rest-server.ts 8ac62c80..., domains/meta.ts c3e7a63c...). GATE FAMILY derived with 'node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack' (no paths — self-derived change set of 27 files; header confirmed 'derived from the tree of objectstack-ai/objectstack at commit c96c5e9c' and the --repo assertion held). Every runnable family run and green: check-adr-0087-registration '1 declared-breaking changeset(s), each carrying an ADR-0087 disposition'; check-changeset-no-major 'This diff introduces no major bump'; check-empty-changeset '1 declaring changeset(s) added'; check:changeset-gate-self-tests (118+212+116 assertions); check:objectui-changeset (15 cases); check:authz-resolver 'single shared authorization resolver intact'; check:cross-package-test-inputs 'OK: 18 package(s) read outside themselves, all declared'; check:test-source-alias 'OK — 72 packages with tests scanned'; check:type-source-resolution 'OK — 93 tsc program(s) across 77 packages'; check:published-files '69 publishable package(s)'; check:slot-lookup 'ratchet holds: 107 unswept site(s) in 25 file(s), none new'; check:page-declaration-shape 'OK — 34 page entries across 2190 sources'; check:nul-bytes 'OK (scanned 6844 text file(s) ... no raw ASCII control bytes)'; release-rehearsal-clone --self-test all cases. TWO MEASUREMENTS DECLARED RATHER THAN COUNTED. (i) check:skill-examples first REFUSED — 'packages/client-react/dist holds no .d.ts declarations — the package is not built', i.e. it would have reached a verdict without reading the artifact under test. The cause was mine and is worth recording: the earlier dependency build used the filter '@objectstack/client^...', which builds a package's DEPENDENCIES and not the package, so neither client nor client-react had a dist. Both were built and it was re-run; only the second run is reported: '256 prose examples type-check across 3 surface(s) — every marked block parsed, so tsc ran the SEMANTIC pass on all of them'. (ii) The ADR-0087 gate's FIRST run reported '0 non-breaking changeset(s) seen' because the changeset was still untracked — it measured nothing, so it is not counted; it was re-run after the commit and only that run is quoted above. One repo-scope scan NOT run and declared as such rather than reported green: repo-wide 'pnpm lint' is CI's run, not owed here. NOT claimed as a proven narrowing either — the three evidence items that would make a narrowing a measurement were not collected, so it is simply 'not run'.",
 "open_questions": [],
 "out_of_scope_findings": []
}

---

Generated by Claude Code