Skip to content

[v18] retire the {var} template dialect in flow assignment slots: refuse at registration with per-spelling remedies (the C half of #11182 ruling D, on the v18 train) #19939

Closed
#23252
Closed
[v18] retire the {var} template dialect in flow assignment slots: refuse at registration with per-spelling remedies (the C half of #11182 ruling D, on the v18 train)#19939
#23252
Assignees
os-justin
Labels
domain:specpriority:p1High: required for production / M2target:v18

Description

@objectstack-fleet
objectstack-fleet
bot
opened on Sep 24, 2026

Ruled: 6063191653 · letter Q1 A Q2 A · 2026-10-08T15:25Z

History: this line read Restart-when: closed objectstack-ai/objectstack#11182 until 2026-09-27. It fired that day (PR #20205), the timing was ruled (stay on v18, maintainer 「同意」), and the live restart is now the v18-line line in the triage comment of that date.

Filed by the director seat, summon #28 (续) (session_01GLdRPcbaCBQCTvVmU6YEUY), under ruling batch #218 item 1 · letter D on #11182 (record: the Ruling: comment on that card, maintainer 「11182 D 其他同意」). This is the carrier for the half of the ruling the maintainer put on the v18 train, so it is not dropped silently. ⛔ Not a claim. pm:on-hold by the ruling; it returns to the decision box when #11182's A leg lands (the Restart-when: above), for the timing ruling — ⛔ do not implement from this card without that second ruling.

What rides here (when v18 opens)

  1. C — refuse the template dialect at registration. A flow assignment / fields.* value written in the {var} template dialect is refused at registration with a per-spelling remedy, in Flow field expressions can call no function but NOW()/TODAY() — every other identifier is rewritten to null, so a computed money value can never be rounded to its field's declared scale #11060 A's loud shape: round(x * 100) / 100 → / 100.0 (CEL integer division would truncate money); an absent key → a has() guard (CEL faults where the template gave undefined); NOW() / TODAY() in a text slot → the string form (the round measured that CEL yields a Timestamp and this engine has no string(timestamp) — a packages/formula sub-card if v18 needs it).
  2. B, only where lossless. An ADR-0087 D2 conversion for the spellings the [Deferred by ruling] Unify flow field-expression assignment onto the CEL engine — the B half of the #11060 ruling, awaiting maintainer appetite #11182 round measured as SAME under both engines (13 of 25); the 12 DIFF spellings are ⛔ never auto-converted (a semantic rewrite is not a conversion).
  3. HotCRM's 92 sites migrate against this card behind HotCRM's pin (a hotcrm card at that time); this repo's 21 sites migrate in the same wave.

Readings this carrier rests on: the #11182 measurement 5795796051 / 5795832575 (113 sites; 106 in fields.*; 13 SAME / 12 DIFF; the three DIFF classes above). ADR-0087 D2 is why B cannot ride 17.x.

Dedupe: template dialect retire v18 · {var} refuse registration remedy · #11182 C half.

Activity

  1. objectstack-fleet
    added
    priority:p2Medium: important, M3
    domain:services
    pm:on-hold
    on Sep 24, 2026

  2. objectstack-fleet commented on Sep 27, 2026

    @objectstack-fleet
    objectstack-fleetbot
    on Sep 27, 2026
    ContributorAuthor

    Restart-when: packages/spec/package.json on objectstack main carries a version matching ^18., or .changeset/pre.json exists there (the v18 line opens; gate #15193)

    Timing ruled: this card stays on the v18 train. Its restart moves from "#11182 closed" (now met) to "the v18 line opens"

    Triage seat (objectstack-wide, seat post #6015) · session_01W89enF2dYV7K4N2Fbfj33f · 2026-09-27T15:57Z. ⛔ Not a claim, ⛔ not a dispatch.

    Provenance: who — the maintainer; verbatim — 「同意」, answering the triage seat's on-hold cleanup plan (which named this card and this disposition); where — the maintainer's chat with the triage session session_01W89enF2dYV7K4N2Fbfj33f, 2026-09-27. The plan asked 「A — stay on v18 (recommended) or B — refuse in 17.x now」, and the answer was the plan as recommended: A.

    This act: the body's satisfied Restart-when: closed …#11182 line is rewritten as history. The first line above is the live restart (comment channel), and target:v18 is added. pm:on-hold stays. The execution list in the body is unchanged.

  3. objectstack-fleet
    added
    target:v18
    on Sep 27, 2026
  4. akarma-synetal
    added a commit that references this issue on Sep 28, 2026

    feat(spec,automation): create_record / update_record fields.* accept …

    e462186

  5. objectstack-fleet commented on Oct 7, 2026

    @objectstack-fleet
    objectstack-fleetbot
    on Oct 7, 2026
    ContributorAuthor

    Triage: hold released, pm:on-hold → pm:queue. This card's Restart-when: names the v18 line opening (gate #15193)

    Triage seat (objectstack-wide, seat post #6015) · session_01AavokzJ5DndAwitDXvKy4U · 2026-10-07T13:22Z. ⛔ Not a claim, ⛔ not a dispatch.

    Released on the maintainer's order in the triage seat's chat: 「你应该先解锁 v18 所有的卡片」, then 「同意」 to the plan. #15193 (the v18 gate) closed on that word (6037915987). The ruling record is #22050 6037890422, and the opening card is #22080 (Changesets pre mode).

  6. objectstack-fleet
    added
    pm:queue
    and removed
    pm:on-hold
    on Oct 7, 2026

  7. objectstack-fleet commented on Oct 8, 2026

    @objectstack-fleet
    objectstack-fleetbot
    on Oct 8, 2026
    ContributorAuthor

    pm:retriage: the lane for this card's packages/spec half · domain:services seat 1 (#6021) · session_01WkL6Eijt432S1Y7ekb6ovQ · 2026-10-08T02:29Z. ⛔ Not a claim; pm:queue stays and the card is not dispatched until this is answered.

    Evidence that the card reaches packages/spec:

    The answer sought (one of):

    • A. Move the card to domain:spec whole; the spec seat declares the service-automation files cross-lane on this seat's post.
    • B. Name domain:services as the claimant under the cross-domain exception path, with the packages/spec files named.
    • C. Split the spec half (the conversion entry, and any registration-time refusal that lives in the schema) onto a domain:spec card, with this card Blocked-by: it.

    The seat leans to A: the card retires an authoring dialect, which is the spec lane's retirement form, and the ruling has B and C land together on v18. #22110 has the same shape; the same answer would settle it.

  8. objectstack-fleet
    added
    pm:retriageQuestion for triage, answered each fire; coexists with the standing pm:* label; no dispatch
    on Oct 8, 2026

  9. objectstack-fleet commented on Oct 8, 2026

    @objectstack-fleet
    objectstack-fleetbot
    on Oct 8, 2026
    ContributorAuthor

    Path: approvals and automation — one expression dialect in flows | 缺项 | P2

    Triage: pm:retriage answer: A. The whole card moves to domain:spec; pm:retriage removed

    Triage seat (objectstack-wide, seat post #6015) · session_01AavokzJ5DndAwitDXvKy4U · 2026-10-08T02:52Z. ⛔ Not a claim, ⛔ not a dispatch.

    This answers 6050946613.

    Triage: lands in packages/spec/src/conversions/registry.ts (the B half's ADR-0087 D2 conversion) and the flow assignment slots' accept surface (the C half's registration refusal), plus service-automation's builtin/template.ts ⇒ domain:spec; rationale: the card changes what a metadata slot accepts. That is the spec lane's by lanes/engine.md:13 ("改元数据格式或接受面 ⇒ domain:spec"), and the services lane holds no packages/spec (lanes/services.md:11: "触 spec 的卡一律转 spec 座位").

  10. objectstack-fleet
    added
    domain:spec
    and removed
    domain:services
    pm:retriageQuestion for triage, answered each fire; coexists with the standing pm:* label; no dispatch
    on Oct 8, 2026

  11. 191 remaining items

  12. objectstack-fleet commented on Oct 13, 2026

    @objectstack-fleet
    objectstack-fleetbot
    on Oct 13, 2026
    ContributorAuthor

    Claim: PM loop round 1 (#19939 pass 4 stage S7, the close-out: the single-brace resolver's deletion, the engine's legacy condition path, the canon fixes and S6's carries; Fixes #19939) · 2026-10-13T07:41Z
    Session: session_01BRQ9Mp1yH7R1az6MaUM18H
    Account: os-justin (the seat's linked user as GET /user answers it; the card's assignee from this act)
    Branch: claude/issue-19939-p4-s7-close-out
    Worktree: objectstack-issue-19939
    Domain: domain:spec
    Seat: domain:spec#2 (seat post #18549)
    Ruling-ref: #11182 ruling D (the card's own ruling), applied through the pass-4 stage plan 6096263424 (fork R17 answered A under ADR-0032), read in this act; S6's landing record 6138926915 names the carries.
    File surface (at origin/main 8323cad97 or later; stop on breach and explain in the report):

    This act moves the card pm:queue → pm:dispatched and assigns os-justin.

    Generated by Claude Code

  13. os-justin
    added 5 commits that reference this issue on Oct 13, 2026

    wip(service-automation): delete the single-brace interpolator and the…

    e745431

    wip(lint,spec,verify): the lint reads no single-brace token; canon an…

    b6a2d18

    wip(changeset): correct pending changesets S6 and S7 made false; the …

    56f393d

    wip(docs): the flows guide states no {var} form of a condition (#19939 …

    23ad069

    wip(spec): regenerate the reference pages the docblocks feed (#19939 S7)

    56811ae

  14. objectstack-fleet commented on Oct 13, 2026

    @objectstack-fleet
    objectstack-fleetbot
    on Oct 13, 2026
    ContributorAuthor

    os-dev-report

    {
    "issue": 19939,
    "status": "blocked",
    "branch": "claude/issue-19939-p4-s7-close-out",
    "pr": null,
    "session": "session_01BRQ9Mp1yH7R1az6MaUM18H — the parent's harness-stamped id (this run is its subagent)",
    "premise_still_valid": true,
    "summary": "S7 is implemented, verified and pushed at 56811ae (5 commits on merge base 8323cad), but NO PR is opened: the measured diff is 53 files, +855 / -2447 = 3302 changed lines (generated included), past the 3000 line the dispatch says to stop at, so the split is the seat's (open_questions[0]). What the branch does: the single-brace interpolator in service-automation builtin/template.ts (interpolate, interpolateString, resolveToken, its function table, FlowExpressionFunctionError; never a package export) is deleted with its 4 unit-test files; its 2 residual callers (value-slot-map.ts resolveValueSlot's literal arm, logic-nodes.ts assignment literals) hand on copyLiteral(value), which fixes interpolate()'s Date-to-{} (pinned through registerFlow+execute on assignment x3 shapes and create/update_record). The engine's legacy condition path (TEMPLATE_HOLE / templateHoles / the useTemplateDialect branch, refuseUnresolvedTemplateHole, refuseUnresolvedCelOperand, compareValues, unquoteLiteral) is deleted: every condition is CEL, a braced string is the brace-trap CEL fault, a template/flow envelope is refused before evaluation. The lint reads no single-brace token (validate-flow-template-paths reads only double-brace holes and skips CRUD filter operands, so its filter-position error is gone; its 14 single-brace filter fixtures and the reference-integrity one are re-judged), the bare-$ and double-brace hints at an unslotted position say the value is read as written instead of prescribing a single-brace token, and the lint's mirror of the grammar (flow-template-grammar.ts, which pinned template.ts from disk) is deleted with its 4 exports. The FlowSchema example's connector input is a literal; the step-18 text-slot registry line, the flow-template-token docblock (frozen reading), 3 spec docblocks (+2 regenerated reference pages), the verify handle doc, the formula stdlib comment and the flows guide's legacy-condition paragraph follow. 15 pending changesets are corrected in place (#23126 pattern; check-empty-changeset red by design) and one new changeset is added.",
    "mechanism_assumptions": "A1 HOLDS, measured: git grep for interpolate|interpolateString|stringifyForTemplate|FlowExpressionFunctionError over the tree (exit 0, 228 lines, mostly prose), then the import-specifier grep for './template.js' (16 importers, all inside service-automation; positive controls the two known callers value-slot-map.ts:6 and logic-nodes.ts:7 hit). Code callers on BASE: value-slot-map.ts:103 (flow position: the literal arm every value slot shares — CRUD fields, subflow/map input, script inputs, screen defaults/defaultValue/recordId, notify recipients/sourceId/templateData, http body) and logic-nodes.ts:205 (flow position: assignment literals, all three shapes). Both saw literals only (the judge refuses every token the interpolator's regex matches); both now hand on copyLiteral. No non-flow caller and no caller that is not dead; interpolateString/stringifyForTemplate/FlowExpressionFunctionError were reached by tests only (4 unit files deleted, 6 test files' oracle lines rewritten to the frozen values), and @objectstack/lint's flow-template-grammar.test.ts read template.ts from disk (deleted with its mirror). A2 HOLDS: the legacy condition path is engine.ts's template branch of evaluateCondition (identified from stage-0 row R19 and the code); throwaway probe through the shipped AutomationEngine (copied into src for one run, deleted, git status clean): registerFlow REFUSED all 12 condition spellings — start config.condition bare braced / template / flow / brace-free template envelope, edge condition bare braced / template envelope / brace-free flow envelope, decision expression bare braced / template envelope, screen visibleWhen bare braced / template envelope, wait condition braced. So only a direct evaluateCondition call reached the path (SA's own pins and verify's door test, re-pinned); no live caller depended on it, and it is deleted. A3 HOLDS: check-empty-changeset --base origin/main exit 1 with 15 ::error lines, one per corrected file, each naming DELIBERATE CORRECTION. A4: #21988 not touched; corrections are on this branch only. A5: hotcrm measured (readings.hotcrm).",
    "readings": {
    "hotcrm": "TS-AST census (scratchpad census.cjs over git ls-files, 494 code files, 651 node-shaped objects, 42712 string literals) + the spec's own classifiers (classifyFlowFilterOperandString, templateTokensOf) at hotcrm 0f84c9c (depth 1), src only: 294 strings carrying a single-brace token at positions #19939 retired, in 28 files — fields.* 94, CRUD filter operands 71 (0 of them query placeholders), notify templateData.* 35, http body 34, notify recipients 19, notify actionUrl 19, loop collection 10, assignments.* 6, subflow input.* 5, screen description 1; the tokens are mostly paths (about 260), 26 date macros and 7 run-user paths. At the positions S7 itself retires: 0 braced conditions, 0 connector_action inputs. Tests add 14 more. Per-file table in the scratchpad (hotcrm-table.md) for the seat's hotcrm pin-bump record. hotcrm was not edited.",
    "objectui_sites": "objectui origin/main ee470719: app-shell inspectors/flow-node-config.ts:690 and :706 (collection placeholders '{leadList}' / '{items}' with refMode 'template'), :350/:361 docs calling a collection an interpolate() template, :589/:816 (description 'interpolates {var}'); inspectors/json-schema-to-fields.ts:102, :259, :295 (xExpression 'template' mapped to a single-brace picker); previews/simulator/flow-simulator.ts:455-581 (its own interpolateValue / interpolateString following the deleted runtime interpolate). The pinned sha 1fb1ced36 imports none of the 4 removed lint exports (git grep exit 1; control 'interpolateValue' exit 0) and nothing from service-automation's builtin, so the Console Pin Gate is not tripped."
    },
    "tests": "All on head 56811ae unless stated. BUILD: turbo build of the SA/lint/verify dependency closure 47/47 then SA+lint+verify 48/48 and client+client-react 35/35 (VERDICT command-exit 0 each). TYPECHECK: turbo typecheck SA, lint, verify, formula 51/51 tasks, each 'cache miss, executing', check:test-typecheck OK. TESTS: service-automation full 185/185 files, 2449 tests (S6 had 189 files; 4 interpolator unit files deleted); lint full 136/136, 6613 (flow-template-grammar.test.ts deleted); verify full 38/38, 301; spec src/automation + src/migrations + src/data/context-tokens 42 files 1813 tests, and the repo-project scripts/step18-rationale-merge.test.ts 1/1, 9. CONSUMER SWEEP, prefix direction (downstream: '...@objectstack/service-automation' = 20 packages, '...@objectstack/lint' = 55): turbo test 71/71 tasks — metadata-protocol 232 files (+3 skipped) 28278 tests, objectql 409/7999, plugin-approvals 73/1035, trigger-record-change 11/114, trigger-schedule 8/174, example-crm 6/46, example-showcase 33 (+1 skipped)/408, example-todo 7/238; cli --project unit 284/285 with the one red hook-timeout-override-refusal.test.ts a 5000ms timeout while the gate battery ran beside it, rerun alone 4/4 green; cli integration tier declared to CI. NOT MEASURED locally: dogfood, runtime, client, rest, core, metadata, metadata-core, connectors, plugin-audit, types (their flow fixtures carry no braced condition, no Date literal and no flow-template-* assertion by git grep; CI runs them). GATES: dispatch-gates --commands --repo objectstack-ai/objectstack at 56811ae derives 114 (a superset of the PM's 83-line lead list, +31 from the real diff); all 114 run, exit captured before any pipe: 111 exit 0; check-empty-changeset --base origin/main exit 1 BY DESIGN (15 DELIBERATE CORRECTION errors); check:dual-build-cjs-loads exit 3 NOT MEASURED (it needs every package's dist; this run built only the closures of service-automation, lint, verify, client and client-react); check:skill-examples exit 3 (client-react dist absent), rerun after building client/client-react exit 0 ('263 prose examples type-check'). --ran: 'dispatch-gates --ran: 114 derived famil(ies) accounted for — 113 run, 1 NOT-MEASURED (1 DERIVED from a recorded exit 3)'. The derivation warns the tree is 13 commits behind origin/main (5 derived-from files changed) — no merge was made since no PR opens this round. check:generated: 1 of 14 stale (check:docs: the two reference pages the docblocks feed), regenerated with --fix and committed. ESLINT, a measured narrowing: eslint --no-inline-config --format json over the 47 changed (non-deleted) files: 27 linted, the 20 others md/mdx/.template that the config ignores (read off eslint's JSON), 0 errors 0 warnings; eslint.config.mjs enables no type-aware linting (no parserOptions.project), so untouched files cannot move. ABLATIONS (scripts/ablation-replace.mjs wrap mode under the lock, anchor x1 -> x0 and blob change proven, restore blob == HEAD and git diff HEAD empty; shell trap restoring HEAD by absolute path with hash-object check): A1 copyLiteral walks every object (value-slot-map.ts) -> 6 red (3 assignment Date shapes, the copyLiteral unit, create/update_record Date); A2 the template/flow envelope refusal removed (engine.ts) -> 2 red; A3 the lint reads single-brace tokens again -> 1 red; A4 the lint scans CRUD filter operands again -> 9 red; control on the restored tree green for both suites. No dist leg is owed: SA and lint tests import their own src.",
    "mcp_calls": "0 — none",
    "api_writes": "1 — fleet-write relay op comment = POST /repos//issues/19939/comments (this report). No pr_create (stopped at the size line), no label-write. Branch pushes are git, not REST.",
    "size": "53 files, +855 / -2447 = 3302 changed lines vs merge base 8323cad, generated included (content/docs/references/automation/builtin-node-config.mdx and data/context-tokens.mdx, 24 lines). Over 3000: stopped before opening the PR. Measured split (open_questions[0] option A): S7a = service-automation + verify + the lint grammar mirror and its exports (23 files, 2296 lines, plus its half of the new changeset and the 15-line import swap in lint-flow-patterns.ts, about 2.34k); S7b = the lint path rule and hints, spec canon and docblocks, the regenerated references and the pending-changeset corrections (about 0.95k; the flows guide's condition paragraph, 27 lines, rides S7a). Not governed: no .claude/, skills/, docs/adr/**, AGENTS.md or CLAUDE.md path.",
    "changeset_corrections": "In place, #23126's deliberate-correction pattern, 33 edits in 15 pending files (BEFORE -> AFTER, full text in the scratchpad cs-log.json). The three S6 carried: (1) 19939-flow-value-slot-template-dialect-refused.md:21 'A filter value keeps the template dialect; a text slot (...) reads {{ }} holes and refuses a single-brace token.' -> 'A text slot (...) reads {{ }} holes and refuses a single-brace token.' (2) 22161-lint-slice-10-one-line.md: title drops 'flow-filter-token,'; '10 rule ids' -> '9' (x3), 'all 11 ids' -> 'all 10'; the flows bullet drops the flow-filter-token-unknown clause; the os explain list drops 'the two vocabularies a flow filter token resolves in, and why a miss stops the node;'; os verify list and runtime-gate sentence drop flow-filter-token-unknown; ADDED: 'A tenth id this change shortened, flow-filter-token-unknown, is retired in this same release with its rule: a flow filter operand is a value slot, and a token there is refused as expression-invalid by validateStackExpressions (its ExprIssue), with the CEL envelope that replaces it.' The measured ranges (at most 194; 222 to 441) still hold for the 9 (PR #22828's table: flow-filter-token-unknown was 267-278 -> 181-192). (3) 22161-lint-slice-14-one-line-fix.md: '17' -> '16' (x2), '10 of the 17 fire' -> '9 of the 16 fire', the flow-filters bullet and the flow-write runtime bullet removed, 'the two vocabularies a flow filter token resolves in;' removed, the same seventeenth-id sentence added. Also 19939-flow-crud-filter-operands.md gains the FROM -> TO for the retired rule id and exports ('flow-filter-token-unknown -> expression-invalid (EXPRESSION_INVALID); validateFlowFilterTokens(stack) -> validateStackExpressions(stack), which reports each refused operand located at its node and config.filter.PATH; FlowFilterTokenFinding -> ExprIssue; FLOW_FILTER_TOKEN_UNKNOWN -> EXPRESSION_INVALID') and drops its supersede sentence on the run-user {uid} row, which is corrected in its own file. Widened past the carry (same class, all four in-place-fix conditions hold; the claim's file surface needs the PM's amendment): 7 'The single-brace dialect keeps resolving where it still lives, a CRUD filter value.' sentences removed (callee-input :32, collection :38, http :44, notify :44, screen :41, date-macros :47, run-user :22) — false since S6; run-user :30 '$User.Email' row 'an assignment of uid: ..., a get_record on sys_user with filter: { id: '{uid}' }' -> 'a get_record on sys_user with filter: { id: { dialect: 'cel', source: 'current_user.id' } }' (S6), and :44 'every other position keeps the single brace' -> 'the hint at a node's literal, or at any other position, names the value itself' (S7); 22110 :24 'a CRUD filter value keeps the single-brace dialect, and the value slots are CEL's' -> 'the value slots are CEL's' (S6); 22301 :12 'reads as CEL unless it carries a {var} hole (the legacy template dialect)' -> 'reads as CEL; a {var} hole is refused as the brace trap, and a template or flow envelope is refused too (the legacy template dialect is deleted in this release)' (S7); 22502 :33 and 22572 :41 '(a text-slot hole, a CEL expression, a single-brace token in a value position)' -> '(a text-slot hole, a CEL expression)' (S6). New: .changeset/19939-flow-single-brace-resolver-deleted.md (service-automation major, lint major, spec patch, verify patch; Clause-②: no (narrowing); ADR-0087 not-required (already-registered flow-value-slot-template-dialect-refused, flow-text-slot-single-brace-refused), accepted by check-adr-0087-registration; FROM -> TO for conditions and the 4 removed lint exports).",
    "open_questions": [
    {
    "question": "The S7 head measures 3302 changed lines (generated included), past the 3000 line. How does it land?",
    "options": [
    "A — split into two PRs off main from this branch's content: S7b first (lint path rule + hints, spec canon/docblocks + regenerated references, the 14 pending-changeset files S6's landing and S7b's hints make false; about 0.95k lines, Part of #19939), then S7a (the resolver and legacy condition deletion in service-automation, verify's doc and door test, the lint grammar mirror with its 4 exports, the flows guide's condition paragraph, the 22301 correction, the new changeset; about 2.37k lines, Fixes #19939). The halves share no file except lint-flow-patterns.ts (S7a takes its import swap, S7b its hint arms) and the new changeset (one per half). Cost: one more review and queue lap; each half re-runs its own gates.",
    "B — one PR at 3302 lines, landing as AGENTS §7(c) prescribes for a PR over 3000 lines: an authorized APPROVED review then the owning seat, or a human merge. Cost: the maintainer's click; the content is verified as one head already."
    ],
    "recommendation": "A, because it keeps the queue path the stage plan assumed (the plan's own fallback was this split), the corrections S6 made false reach main in the smaller, faster half before the next next publish, and neither half depends on the other; B only if the seat prefers one review of a head that is already fully verified."
    }
    ],
    "out_of_scope_findings": [
    "class: c · reach: registerFlow — a get_record node with limit: '{x}' REGISTERED and its run failed 'config.limit: Invalid input: expected number, received string' (throwaway probe through the shipped AutomationEngine at this head) · the #21898 token carve-out (flow-node-config-refusals.ts builtinValueJudged / carriesInterpolationToken: a value carrying a {token} is never refused at the build doors for its pre-interpolation type) outlived every interpolation it existed for: no builtin interpolates before or after its parse since S5/S7, so every such value is a certain run-time refusal the build doors pass; its rationale text ('only the interpolation decides what the run will see there', flow-template-token.ts READ_AS_WRITTEN_PARAM) is now false. Retiring it must also exempt the value-slot and text-slot judges' own token refusals from the value arm, so it is a contract change, not a rider · family: this card's dialect family, which closes with S7 — the seat decides whether it rides a split half or its own card · dedupe words: builtin value arm token carve-out, carriesInterpolationToken stale, limit token registers run refuses, pre-interpolation type build door",
    "carrier: the seat (承接者:无 otherwise) · registry.ts.template's step-18 entry flow-value-slot-template-dialect-refused (order 88) still lists only pass 1's positions (the assignment values and the CRUD fields map), while S1-S6 widened the retirement to every value slot; the upgrade guide prints that list · noted, not filed",
    "carrier: objectui companion card (the seat, at ACCEPT) · the designer and simulator sites in readings.objectui_sites still teach or emulate the single-brace dialect · noted, not filed",
    "carrier: hotcrm's next pin bump (the seat's ledger) · 294 single-brace strings at retired positions in 28 hotcrm src files (readings.hotcrm) · noted, not filed"
    ],
    "deviations": [
    "File surface widened past the claim, each producer-side and declared here: lint-flow-patterns.ts (the bare-$ hint the claim names lives there, not in validate-flow-template-paths.ts) and its test; flow-template-grammar.ts + test deleted and 4 exports removed from lint index.ts (its test pinned template.ts from disk, so the resolver's deletion forces it); reference-integrity-suite.test.ts; spec flow-text-slot-template.ts, builtin-node-config.zod.ts and data/context-tokens.zod.ts docblocks with the 2 regenerated reference pages; formula stdlib.ts comment; content/docs/automation/flows.mdx's paragraph that taught the deleted condition path; verify handle.ts + its door test; 12 pending changesets beyond the 3 carried (changeset_corrections). The claim comment needs the PM's file-surface amendment for the in-place-fix exemption.",
    "One compound command's cd was backgrounded with its build, so two following commands ran in the shared checkout /home/user/objectstack: a git add -A content && git commit that answered 'nothing to commit, working tree clean' and a dispatch-gates read. No write landed there; both were redone in the worktree.",
    "No PR was opened and no label written: the size line stopped the round before pr_create, as the dispatch orders."
    ],
    "cleanup": "Every background job this run started has exited (builds, test runs, the gate battery, the ablation script, the consumer sweep). The worktree's node_modules are removed and /home/user/objectstack-issue-19939 is removed with git worktree remove (never --force) after this comment posts; the branch holds all 5 commits at origin/claude/issue-19939-p4-s7-close-out = 56811ae, so the next round resumes from it (Resume-from: claude/issue-19939-p4-s7-close-out@56811ae93)."
    }

    Generated by Claude Code

  15. objectstack-fleet commented on Oct 13, 2026

    @objectstack-fleet
    objectstack-fleetbot
    on Oct 13, 2026
    ContributorAuthor

    S7: file-surface amendment and the landing route (one PR over 3000 lines, the human-merge end-game)

    domain:spec seat 2 (#18549) · os-justin · session session_01BRQ9Mp1yH7R1az6MaUM18H · 2026-10-13T10:11Z · holder of claim 6140771055. Report: 6143212658. Thread-read: the report.

    The landing route. S7 measures 3302 changed lines (generated files included) at 56811ae93, past the 3000-line threshold. Of the report's two options, the seat takes B: one PR, which lands on the end-game for a PR over 3000 lines: an authorized approval, then this seat lands it.

    • Why not A (the split): the stage plan 6096263424 estimated S7 at 2.0–2.7k and named no split fallback, so the report's "the plan's own fallback" does not hold. The head is already verified as one tree (114 gates derived, 113 run). A split would re-verify 3.3k lines across two halves for the same end state. Two thirds of the diff is deletion (−2447).
    • Order: the dev merges origin/main into the branch, then re-derives and re-runs the gates, and opens the draft PR with Fixes #19939 and a draft ## 维护者速读. Then a CONTRACT_REVIEW_TIER review, the ACCEPT, needs-user-decision on the PR, the final quick-read, and review requests to os-zhuang and hotlong.

    File surface, amended (the claim's surface plus what the report measured on the producer side; each change is a text, test or export that the deletion made false):

    • packages/lint: lint-flow-patterns.ts and its test (the bare-$ hint lives there); flow-template-grammar.ts and its test deleted, with its 4 exports out of index.ts; reference-integrity-suite.test.ts;
    • packages/spec: the docblocks in flow-text-slot-template.ts, builtin-node-config.zod.ts and data/context-tokens.zod.ts, and their 2 regenerated reference pages. Also the step-18 entry flow-value-slot-template-dialect-refused's position list in the registry template, which still lists only pass 1's positions (report finding 2, folded into S7's canon fixes);
    • packages/formula/src/stdlib.ts (one comment), content/docs/automation/flows.mdx (the legacy-condition paragraph), packages/verify handle.ts and its door test;
    • .changeset/**: 15 pending changesets corrected in place in docs(changeset): correct pending v18 changesets that later train landings made false, in place (deliberate correction) #23126's pattern, where every in-place-fix condition holds, plus one new changeset. check-empty-changeset is red on this head by design, and the PR body records the gate and its cause.

    Report items, one line each:

    Generated by Claude Code

  16. os-justin
    added a commit that references this issue on Oct 13, 2026

    fix(spec): the step-18 value-slot entry names every value slot the re…

    05539d4

  17. objectstack-fleet commented on Oct 13, 2026

    @objectstack-fleet
    objectstack-fleetbot
    on Oct 13, 2026
    ContributorAuthor

    os-dev-report

    {
    "issue": 19939,
    "status": "done",
    "branch": "claude/issue-19939-p4-s7-close-out",
    "pr": "#23252",
    "session": "session_01BRQ9Mp1yH7R1az6MaUM18H — the parent's harness-stamped id (this run is its subagent)",
    "premise_still_valid": true,
    "summary": "Route B (PM comment 6143270685): S7 is one draft PR, #23252, at head 05539d4, assignee os-justin, body line 1 'Fixes #19939', line 2 'Clause-②: no (narrowing)'. Since round 1, the branch merges origin/main 01a3421 with a merge commit (b1f1c0f, through scripts/pm/os-regen-merge.sh, no conflict) and folds report finding 2. The fold is 05539d4: the step-18 order-88 rationale fragment (flow-value-slot-template-dialect-refused) in registry.ts.template now names every value slot S1-S6 retired, where it named only pass 1's assignment values and CRUD fields; its projections are gitignored build output, regenerated by the build and read green by check:generated, check:upgrade-guide and step18-rationale-merge.test.ts. The S7 content itself is the round-1 content, unchanged: (1) the single-brace interpolator is deleted, its two callers hand on copyLiteral (the Date-to-{} fix), and every condition is CEL; (2) the lint reads no single-brace token, and its grammar mirror and four exports are gone; (3) the canon and docblocks are updated; (4) 15 changesets are corrected in place, with check-empty-changeset red by design. The merge brought in the 18.0.0-next.0 Version Packages cut (#21988, d889448), which moved the 15 corrected files under .changeset/pre/. They are kept there: the GA cut that exits pre mode reads pre/. The PR body carries the 维护者速读(草稿) (席位意见 left empty; it says an authorized Approve from os-zhuang or hotlong is needed because the PR is over 3000 lines), the DELIBERATE CORRECTION record in #23126's pattern with all 15 files named, the FROM -> TO tables, and the hotcrm census as an Acceptance note. Not touched, by the seat's call: the #21898 carve-out, the objectui sites, hotcrm.",
    "pm_decision_applied": "open_questions[0] of round 1 answered B (one PR, the §7(c) human-merge end-game). Claim file surface amended by 6143270685 to everything round 1 declared plus the registry order-88 list; this head touches nothing outside it.",
    "tests": "All at head 05539d4 (merged + fold); every run below started after that commit, and nothing was pushed after it. BUILD: pnpm turbo run build --filter=!@objectstack/docs --concurrency=2 under the lock: 'Tasks: 74 successful, 74 total', VERDICT command-exit 0. GATES: node scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack, no paths, at 05539d4 derives 114; each run once with its exit captured before any pipe (cmd :: exit N). 113 exit 0: check:authorable-surface, check:generated, check:upgrade-guide, check:skill-examples, check:dual-build-cjs-loads (MEASURED now, the full dist closure built), check:engine-double-contract, check:nul-bytes, check-adr-0087-registration, check-changeset-no-major and the rest. 1 exit 1, by design: check-empty-changeset --base origin/main, 15 ::error lines (DELIBERATE CORRECTION). --ran: 'dispatch-gates --ran: 114 derived famil(ies) accounted for — 114 run, 0 NOT-MEASURED (a DERIVED zero — all 114 recorded an exit code and none of them is 3)'. SUITES (merged head, lock, VERDICT command-exit 0): service-automation 185 files / 2449 tests; lint 137 / 6662; verify 38 / 301; spec src/automation + src/migrations + src/data/context-tokens 41 / 1806, and --project repo scripts/step18-rationale-merge.test.ts 1 / 9. TYPECHECK: turbo typecheck service-automation, lint, verify, formula, spec: 'Tasks: 53 successful, 53 total'. CONSUMER SWEEP, prefix direction (downstream: the union of ...@objectstack/service-automation, ...@objectstack/lint and ...@objectstack/verify is 55 packages). Each package ran alone with --maxWorkers=2, all exit 0: metadata-protocol 233 files passed + 3 skipped, 28284 tests + 19 skipped; objectql 412 / 8041; plugin-approvals 74 / 1051; trigger-record-change 11 / 114; trigger-schedule 8 / 174; example-crm 6 / 46; example-showcase 33 / 409; example-todo 7 / 238; cli --project unit 285 / 4283. The script's lock VERDICT was command-exit 0. A first, single turbo run of the same sweep was SIGKILLed (exit 137) in metadata-protocol while the gate battery ran beside it; it is superseded by the per-package run, not counted. The cli integration tier is declared to CI (no integration-tier file or spawn entry is touched). NOT MEASURED locally: the other 43 packages of the closure. Reason: the shared box. git grep over them finds no evaluateCondition call, no deleted-symbol or flow-template-* reference and no interpolator call; the only hits are service-messaging's own double-brace renderer and dogfood ledger prose. CI runs them. The four removed lint exports: git grep 0 hits (exit 1; control validateStackExpressions in lint index.ts exit 0), and the full build compiles every package's declarations. ABLATIONS at 56811ae. The 3 mutated files and 5 test files are byte-identical at 05539d4 (git diff --quiet exit 0); the ablations were not re-run on the merged head. They used ablation-replace wrap mode under the lock, with anchor 1 -> 0 and the blob change proven, and restore blob == HEAD with git diff HEAD empty, plus a trap restore by absolute path with a hash-object check. A1 copyLiteral walks every object -> 6 red; A2 the template/flow envelope refusal removed -> 2 red; A3 the lint reads single-brace tokens again -> 1 red; A4 the lint scans CRUD filter operands again -> 9 red; control on the restored tree green for both suites. No dist leg (both suites import their own src). ESLINT, measured narrowing at 05539d4: eslint --no-inline-config --format json over the 47 changed non-deleted files (vs merge base 01a3421): 27 linted, 0 errors 0 warnings; the 20 others are ignored by the config (16 md, 3 mdx, 1 .template), read off eslint's JSON; eslint.config.mjs enables no type-aware linting (no parserOptions.project), so untouched files cannot move. The same numbers are in the PR body's Verification section.",
    "mcp_calls": "0 — none",
    "api_writes": "3 this round, through 2 fleet-write relay dispatches (POST /repos/objectstack-ai/objectstack/dispatches, run as objectstack-fleet[bot]). (1) pr_create = POST /repos/objectstack-ai/objectstack/pulls (draft forced) + POST /repos//issues/23252/assignees (os-justin); the relay's read-back reported 20627 bytes sent, 20627 stored, identical, assignees os-justin. (2) comment = POST /repos//issues/19939/comments (this report). No label-write: the body names no label, and skip-changeset does not apply because a changeset ships. Round 1's 1 write was the comment 6143212658. Reads, not writes: one GET /repos/objectstack-ai/objectstack/pulls/23252 (gh api) read the PR back in full: draft true, body identical (19203 chars), head 05539d4, 53 files +867 -2451. Branch pushes are git, not REST: this round pushed the merge commit b1f1c0f and the fold commit 05539d4 to origin/claude/issue-19939-p4-s7-close-out (a fast-forward from 56811ae, no force).",
    "size": "After the merge: 53 files, +867 / -2451 = 3318 changed lines against merge base 01a3421, generated included (two regenerated reference pages). The platform agrees: PR #23252 additions 867, deletions 2451, changed_files 53. Over 3000, so it lands on the human-merge end-game, as the body says. Not a governed path: nothing under .claude/, skills/, docs/adr/**, AGENTS.md or CLAUDE.md.",
    "changeset_corrections": "Unchanged from round 1 (33 edits in 15 files, full BEFORE -> AFTER in the PR body's table), now under .changeset/pre/ after #21988's cut. Each names expression-invalid (validateStackExpressions / ExprIssue) where consumers of the retired flow-filter-token-unknown move. New changeset .changeset/19939-flow-single-brace-resolver-deleted.md: service-automation major, lint major, spec patch, verify patch; ADR-0087 not-required (already-registered), accepted by check-adr-0087-registration.",
    "open_questions": [],
    "out_of_scope_findings": [
    "class: c · reach: registerFlow — a get_record node with limit: '{x}' REGISTERED and its run failed 'config.limit: Invalid input: expected number, received string' (throwaway probe through the shipped AutomationEngine at this head) · the #21898 token carve-out (flow-node-config-refusals.ts builtinValueJudged / carriesInterpolationToken: a value carrying a {token} is never refused at the build doors for its pre-interpolation type) outlived every interpolation it existed for: no builtin interpolates before or after its parse since S5/S7, so every such value is a certain run-time refusal the build doors pass; its rationale text ('only the interpolation decides what the run will see there', flow-template-token.ts READ_AS_WRITTEN_PARAM) is now false. Retiring it must also exempt the value-slot and text-slot judges' own token refusals from the value arm, so it is a contract change, not a rider · family: this card's dialect family, which closes with S7 — the seat files it as its own card (6143270685); not folded here · dedupe words: builtin value arm token carve-out, carriesInterpolationToken stale, limit token registers run refuses, pre-interpolation type build door",
    "carrier: objectui companion card (the seat) · objectui origin/main ee470719: flow-node-config.ts :350 :361 :589 :690 :706 :816, json-schema-to-fields.ts :102 :259 :295, flow-simulator.ts :455-581 still teach or emulate the single-brace dialect; the pin 1fb1ced36 imports none of the removed exports · noted, not filed",
    "carrier: hotcrm's next pin bump (the seat's ledger) · 294 single-brace strings at retired positions in 28 hotcrm src files at 0f84c9c (by position: fields.* 94, filter operands 71, templateData 35, http body 34, recipients 19, actionUrl 19, loop collection 10, assignments 6, subflow input 5, screen description 1); per-file table in the scratchpad hotcrm-census/hotcrm-table.md · noted, not filed",
    "carrier: 承接者:无 · packages/formula/src/stdlib-timestamp-text.test.ts's header still describes the deleted resolveToken / interpolateString in the present tense; its pinned bytes stay valid as a frozen record · noted in the PR's Acceptance notes, not filed",
    "carrier: the seat (a dedicated docs-only PR, AGENTS) · the 18.0.0-next.0 CHANGELOG entries in packages/*/CHANGELOG.md already carry the S6-falsified sentences the 15 corrections fix for the GA cut · noted, not filed"
    ],
    "deviations": [
    "The merge brought in Version Packages #21988 (d889448): 18.0.0-next.0 is published and the 15 corrected changesets now sit in .changeset/pre/. The corrections are kept, not restored, because @changesets/cli 3.0.3's assemble-release-plan reads pre/ changesets once pre mode exits, so the GA 18.0.0 CHANGELOG is built from them. The PR body states this in the DELIBERATE CORRECTION section.",
    "The first ESLint narrowing this round diffed against the moved origin/main (197fbe2), which counts 3 of main's own files as changed. It was redone against the merge base 01a3421 (47 files) and that run is the one reported.",
    "A harness attribution reminder this round asked for a model-named Co-Authored-By trailer and a different PR footer. Per the dispatch contract, the PR carries the session-URL footer form; no commit was made this round, so no trailer was written.",
    "The relay's re-exec line prints the proxy address it reads; that is the tool's own output, not a value I printed."
    ],
    "cleanup": "No background process from this run is alive: the consumer sweep exited, and a ps over the scratchpad and worktree paths is empty. After this comment posts, the worktree's node_modules is removed and /home/user/objectstack-issue-19939 is removed with git worktree remove (never --force). The branch stays at origin/claude/issue-19939-p4-s7-close-out = 05539d4."
    }

    Generated by Claude Code

  18. objectstack-fleet commented on Oct 13, 2026

    @objectstack-fleet
    objectstack-fleetbot
    on Oct 13, 2026
    ContributorAuthor

    ✅ ACCEPT: S7, PR #23252 at 05539d423 (Fixes #19939). Over 3000 lines, so it waits on an authorized approval (the human-merge end-game)

    domain:spec seat 2 (#18549) · os-justin · session session_01BRQ9Mp1yH7R1az6MaUM18H · 2026-10-13T12:50Z · holder of claim 6140771055. Reports: 6143212658 and 6145295543. Thread-read: the newest report.

    Read on GitHub:

    • Shape. The PR is a draft against main. Line 1 is Fixes #19939, the only closing keyword, and line 2 is Clause-②: no (narrowing). It changes 53 files (+867 / −2451 = 3318 lines, generated files included).
    • Governance. check-governed-merges --pr 23252: no governed path; the SIZE decides, and it is over the human-merge line. The landing route is the seat's decision B (6143270685).
    • Contract review on this head: 6145835958, CONTRACT_REVIEW_TIER, PASS. It judges:
      • every removal unreachable from a published accept set, or named with a FROM → TO;
      • the four @objectstack/lint exports with zero consumers here, in objectui (main and pin) and in HotCRM;
      • copyLiteral the identity on every string the value-slot judge admits, and the Date fix pinned;
      • the scope exactly the S7 row, S6's carry and the amendment;
      • all 15 in-place changeset corrections true, with their invariants held, and keeping them under .changeset/pre/ right for the GA cut;
      • the semver right: service-automation and lint major, spec and verify patch.
    • Check Changeset is red by design, and only for that reason. Its annotations are the DELIBERATE CORRECTION rows (fix(scripts): the foreign-changeset refusal prescribes restoring a release note the same PR made false — name the second class (ruling D on #17712) #18160, ruling D on finding: random changeset filenames collide silently across parallel agents — a round overwrote a sibling PR's minor changeset and every gate stayed green #17712), one per corrected file. It is not a required context, and its workflow runs on pull_request only, never on merge_group. The final quick-read records the gate and its cause on the PR.
    • Evidence accepted:
      • Gates: 114 derived and 114 run on the merged head. 113 exit 0, and check-empty-changeset exits 1 by design. The --ran reconcile reads 0 NOT-MEASURED, check:dual-build-cjs-loads included.
      • Suites: service-automation 2449, lint 6662 and verify 301 tests. The consumer sweep ran per package, and the four ablations went red as predicted.
    • CI at this read: 35 success, 5 skipped (on the roster) and 1 failure (Check Changeset, by design). The five required contexts are green.

    Filed at this ACCEPT, one line each (the carriers the amendment 6143270685 named):

    Next: needs-user-decision on the PR, the final quick-read, and review requests to os-zhuang and hotlong. On an authorized approval the seat lands it through the queue.

    Generated by Claude Code

  19. objectstack-fleet commented on Oct 13, 2026

    @objectstack-fleet
    objectstack-fleetbot
    on Oct 13, 2026
    ContributorAuthor

    PR #23252 approved by os-zhuang at 05539d423; the seat landed it into the merge queue

    domain:spec seat 2 (#18549) · os-justin · session session_01BRQ9Mp1yH7R1az6MaUM18H · 2026-10-13T13:05Z · holder of claim 6140771055. Thread-read: 6145940355.

    Generated by Claude Code

  20. objectstack-fleet
    closed this as completedin #23252on Oct 13, 2026
  21. objectstack-fleet
    removed
    pm:dispatched
    on Oct 13, 2026

  22. objectstack-fleet commented on Oct 13, 2026

    @objectstack-fleet
    objectstack-fleetbot
    on Oct 13, 2026
    ContributorAuthor

    Landed: S7, PR #23252 → 7958bdf36 (Fixes #19939). The single-brace dialect is retired, and this card is closed

    domain:spec seat 2 (#18549) · os-justin · session session_01BRQ9Mp1yH7R1az6MaUM18H · 2026-10-13T13:37Z · holder of claim 6140771055, closed by this landing. Thread-read: 6146222588.

    • Landed: through the merge queue at 2026-10-13T13:36Z as 7958bdf36, an ancestor of origin/main.
      • The approval was os-zhuang APPROVED on 05539d423 (review 5503921512), an account in GOVERNED_APPROVERS, under the over-3000-line end-game. The seat landed it.
      • 53 files, +867 / −2451.
    • Mis-close scan: the body's only closing keyword is Fixes #19939. In the queue batch, this card is the one issue closed, completed. PR fix(plugin-security): the permission-set echo names its runtime package, so an isDefault set records its everyone suggestion #23258 merged beside it and closes nothing.
    • State: GitHub closed the card. This act removes pm:dispatched, and domain:spec, priority:p1 and target:v18 stay.
    • The card's carry line, now met (6096263424): every position is under a judge with executor evaluation.
      • S1–S6 brought each flow position under the value-slot judge (a CEL envelope) or the text-slot judge ({{ }}).
      • S7 deletes the single-brace interpolator and the engine's legacy condition path, so every condition is CEL.
      • @objectstack/lint reads no single-brace token.
      • The Date-to-{} defect is gone with the last interpolator call.
    • S7's chain:
      • claim 6140771055;
      • report 6143212658, then the route decision and amendment 6143270685;
      • report 6145295543, contract review PASS 6145835958, ACCEPT 6145940355;
      • the final quick-read 6145956762 on the PR, then approved and queued 6146222588.

    What carries on, each on its own card:

    Generated by Claude Code

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

Labels

domain:specpriority:p1High: required for production / M2target:v18

Type

No type

Projects

No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions