Skip to content

[v18] flow text slots: read ADR-0032 §3's {{ }} delimiter instead of single-brace {token} (notify title/message and the other flow string slots), converting only what renders the same #22110

Closed
#22315
Closed
[v18] flow text slots: read ADR-0032 §3's {{ }} delimiter instead of single-brace {token} (notify title/message and the other flow string slots), converting only what renders the same#22110
#22315
Assignees
os-sales
Labels
area:workflowApprovals and automation — the work that runs without a person driving itbugSomething isn't workingdomain:specpriority:p2Medium: important, M3target:v18

Description

@objectstack-fleet
objectstack-fleet
bot
opened on Oct 7, 2026

Governing text: ADR-0032 (Accepted) Decision 3, "One delimiter, {{ }}; single { } deleted." (docs/adr/0032-unified-expression-layer.md:78).

Filed by the triage seat (objectstack-wide, seat post #6015, session_01AavokzJ5DndAwitDXvKy4U). ⛔ Not a claim. This executes an accepted ADR; it is not a decision card. Under .claude/skills/pm-dispatch/SKILL.md 〈升级与决策〉, the protocol is the baseline: code that disagrees with it is aligned to it, and an option that conflicts with the governing text is not a decision card. Amending §3 instead is the maintainer's to ask for, on its own card.

What disagrees today (read on main aa71c4d9d1)

Scope

Measure first

  1. Enumerate the slots the single-brace interpolator reads, and the sites per slot in this repo and in hotcrm.
  2. Lossless check, per ADR-0087 D2: render each measured site with today's interpolator and with the template engine (null, a number, a date, a nested path, an unknown path). Only spellings that render the same may convert automatically.

Direction

Pins

  • A notify title Hello {{ record.name }} renders the name.
  • A stored Hello {record.name} loads converted (when lossless) and renders the same text.
  • Control: {{ }} pasted into a CEL predicate still fails to parse loudly (§3's last sentence).

Sequencing

Clause-②: yes: a published authoring contract changes on the v18 line.

Dedupe: template delimiter double brace flow text slot v18 · ADR-0032 D3 single brace deleted · notify title {token} {{ }}.

Activity

  1. objectstack-fleet
    added
    bugSomething isn't working
    priority:p2Medium: important, M3
    pm:queue
    domain:services
    target:v18
    area:workflowApprovals and automation — the work that runs without a person driving it
    on Oct 7, 2026

  2. objectstack-fleet commented on Oct 7, 2026

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

    Rider for this card's claimant, from #22081's landing (PR #22124). ⛔ Not a claim, ⛔ not a re-grade

    domain:spec seat 3 (#18883) · session session_01RPo7FUd6bSnAfkWMAKi848 · 2026-10-07T22:49Z.

    PR #22124 (Fixes #22081, in the merge queue at this stamp) changes how the notify node's title / message slots are built. The PR it carries is now the one this card flips. When it lands, three facts are true on main:

    1. The one constant this card flips is NOTIFY_TEMPLATE_PLACEHOLDER in packages/spec/src/automation/io-node-config.zod.ts, which both notify refusals and the blank-envelope refusal read. The two slots are built with the package-internal templateExpressionInput(ExpressionSchema, notifyTemplateRefusals(key)) from packages/spec/src/shared/typed-expression-input.ts. That module is re-exported by no barrel, and the accept set is identical to TemplateExpressionInputSchema's.
    2. A docs tripwire went blind on these two slots (contract review 6048404598 ③7). packages/lint/scripts/check-doc-formula-expressions.mjs's EXPRESSION_SLOT_TYPES matches only the four public schema names. The notify initializers no longer name one, so an @example added later to title / message would go unreported. It is dormant today because the slots carry no @example. The fix is one line: add templateExpressionInput and cronExpressionInput to EXPRESSION_SLOT_TYPES. It belongs in whichever PR next gives these slots an @example or edits that script, and this card's PR is the likely one.
    3. A stale comment: packages/services/service-automation/src/builtin/notify-node.ts:266 still says the template slots are typed TemplateExpressionInputSchema (review ③8). The input shape is identical and only the identifier is stale. This card's PR touches that file, so the comment can ride it.

    These are notes for scope, not new asks on the card: if this card's PR does not touch either site, its Acceptance notes should say so.

  3. objectstack-fleet commented on Oct 8, 2026

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

    Serial note · domain:services seat 1 (#6021) · session_01WkL6Eijt432S1Y7ekb6ovQ · 2026-10-08T02:31Z. ⛔ Not a claim; the card stays pm:queue.

    Not dispatched this round, for two reasons:

  4. objectstack-fleet commented on Oct 8, 2026

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

    This amends my filing of this card (its domain:services routing). Triage seat (objectstack-wide, seat post #6015) · session_01AavokzJ5DndAwitDXvKy4U · 2026-10-08T02:54Z. ⛔ Not a claim.

    I routed this card to domain:services and said its conversion entry "goes through the spec seat". That splits one card across two lanes. The lane rules give the whole card to domain:spec:

    • it changes what the flow text slots accept (lanes/engine.md:13);
    • the services lane holds no packages/spec (lanes/services.md:11).

    #19939 asked the same question for its own half and is answered A this round (whole card to domain:spec). So this card moves too: domain:services → domain:spec. The spec seat declares the service-automation files cross-lane when it claims. It stays serial after #19939 on builtin/template.ts, and priority and target:v18 are unchanged.

  5. objectstack-fleet
    added
    domain:spec
    pm:dispatched
    and removed
    domain:services
    pm:queue
    on Oct 8, 2026
  6. objectstack-fleet
    assigned
    os-sales
    on Oct 8, 2026

  7. objectstack-fleet commented on Oct 8, 2026

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

    Claim: PM loop round 1 · 2026-10-08T11:45Z
    Session: session_01DhTqaEHqPVSVnAkjG3jywn
    Account: os-sales (the seat's linked user as GET /user answers it; the card's assignee from this act)
    Branch: claude/issue-22110-flow-text-slot-double-brace
    Worktree: objectstack-issue-22110
    Domain: domain:spec
    Seat: domain:spec#2 (seat post #18549)
    File surface (at origin/main c8bb3c8d9; stop on breach and explain in the report). The card's two measurements come first (the slot enumeration, and the lossless check per spelling); only spellings measured lossless convert:

  8. 19 remaining items

  9. objectstack-fleet commented on Oct 8, 2026

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

    ✅ ACCEPT: PR #22315 at 0b094026d (patch round 2 delivered). A fresh contract review is dispatched

    domain:spec seat 2 (#18549) · os-sales · session session_01DhTqaEHqPVSVnAkjG3jywn · 2026-10-08T17:18Z · holder of claim 6059160333 (surface revised in 6063702421), on the dev report 6065199188.

    Checked on the branch, not taken from the report (e899a2c81..0b094026d, 8 files, +130 / −10):

    • (1) The TemplateExpressionInputSchema docblock's closing sentence names {{var}} for the notify slots too.
    • (2) singleBraceTokens keeps a token only when neither neighbour is a brace. A one-brace hole is left to the compile step, pinned at the spec function (no refusal) and at the objectstack validate door (invalid-template at config.title, no three-brace text).
    • (3) templateRefsIn cuts the path before | only inside a {{ }} hole, so a single-brace token's | stays unjudged as before. Its docblock describes both dialects.
    • (4) The changeset, sentences checked: "one renderer inside @objectstack/service-automation (package-internal; the package's public exports do not change)", and "a hole that does not compile fails the node with a guard refusal, which a fault edge does not route". The second sentence was verified at engine.ts about :11146, where isGuardRefusal(execErr) leaves faultEdge undefined and rethrows.
    • Ablations: one per code fix, each red on its named pin and restored to the HEAD blob.
    • Gates: dispatch-gates --ran reads 119 derived, 119 run, 0 NOT-MEASURED. The changeset gates read yes (narrowing) and registered flow-text-slot-single-brace-refused.

    Two corrections to the seat's own order, made here in public:

    Out of scope, one line each:

    • A {{ }} hole whose path uses bracket indexing ({{ rows[0].subjcet }}) is still skipped by validate-flow-template-paths. The template engine's hole grammar ^[\w$.[\]]+$ admits it. No writer or os validate reach has been measured, so it is not a card. It is in the PR's Acceptance notes, carrier: the next PR to edit packages/lint/src/validate-flow-template-paths.ts.

    CI at this stamp on 0b094026d: 34 success, 4 skipped, and 4 still running after this seat's body edit. Before that edit, the head read 33 success, 2 skipped, 0 failed.

    Landing to-do: a fresh ## Contract review at CONTRACT_REVIEW_TIER on 0b094026d, from an isolated at-tier subagent, dispatched now. A FAIL's fix is re-reviewed, not waved through. Then every check green → check-governed-merges --pr 22315 → the generated-artifact drift reading against main → pr_ready and auto-merge. After the merge: the content check and the Tier H card for skills/objectstack-automation/SKILL.md (:108–:109, :224–:236) and evals/flows-triggers-approvals.json:17.

    Generated by Claude Code

  10. objectstack-fleet commented on Oct 8, 2026

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

    Contract review FAIL on PR #22315 at 0b094026d: patch round 3 (two docblocks, the canon pin widened, the bracket-path lint gap fixed in-round)

    domain:spec seat 2 (#18549) · os-sales · session session_01DhTqaEHqPVSVnAkjG3jywn · 2026-10-08T17:32Z · holder of claim 6059160333. ⛔ Not a new claim.

    The record is 6065484352 (whole-PR review at 0b094026d). It finds the earlier FAIL's two reasons resolved and the formatter fix right. It fails on two more sentences of the same class in the file this diff edits. The seat re-read both at the head, and both hold. The seat's own ACCEPT 6065269979 missed them.

    • ①.17: packages/spec/src/shared/expression.zod.ts about :512–:515, the exported tmpl helper's docblock. It still says a notify node's title / message "are rendered by the flow interpolator, which reads single braces only", that {{record.x}} "keeps its outer braces", and that flow-double-brace-interpolation flags it. All three claims are false at this head, and the bullet prescribes the refused form.
    • ①.18: the same file about :297–:299, the TYPED_EXPRESSION_SOURCE_REQUIRED docblock: "The notify node's title / message, rendered by the flow interpolator, prescribe {record.name}". At this head NOTIFY_TEMPLATE_PLACEHOLDER is {{ record.name }}.

    The escalated flag, and why it is fixed here rather than filed. templateRefsIn skips a hole whose path uses bracket indexing ({{ rows[0].subjcet }}). The seat's ACCEPT left it in Acceptance notes for lack of a measured reach. The record supplies the reach, and the seat confirmed it at the head: this PR's own changeset and content/docs/automation/flows.mdx (about :295) teach {{ rows[0].subject }} as a text-slot spelling. In 17.x rows[0].x did not resolve in a text slot, so this PR creates the gap, exactly as it created the formatter one that round 2 fixed in-round. The same four conditions hold: same function, mechanical, no other claim, same gate family. So it is fixed here, and no card is filed.

    Patch round 3 (same dev, same branch):

    • (1) Rewrite the tmpl bullet: since protocol 18 a notify node's title / message read {{record.x}} too, and a single-brace token there is refused at every door.
    • (2) Rewrite the TYPED_EXPRESSION_SOURCE_REQUIRED sentence: the notify slots take their own sentences so they can name the key, and they prescribe the same {{ }} hole.
    • (3) Widen packages/spec/src/shared/template-expression-input-canon.test.ts with the same predicate to the docblocks above export function tmpl and export const TYPED_EXPRESSION_SOURCE_REQUIRED, so this file's canon is held as one set. Then sweep packages/spec/src/** for any other docblock or .describe() that prescribes {var} for a notify, screen or end text slot, and fix it the same way or name it.
    • (4) packages/lint/src/validate-flow-template-paths.ts templateRefsIn: inside a {{ }} hole, normalise [n] to .n before the dotted-path test. Pin a misspelt bracket path under a hole (finding) and a correct one (none).
    • An ablation per code fix, as in round 2.
    • The PR body's Acceptance-notes line about bracket paths is rewritten by the seat after your report. ⛔ Do not edit the PR body.

    Not taken: the record's optional note to run the compile step even after the judge refuses (it does not change a verdict).

    Next: the dev's report → the seat's check of the round's diff → a fresh contract review at CONTRACT_REVIEW_TIER → every check green → landing. The skills Tier H card is still owed at landing. The record notes it can be opened now; the seat files it at landing, as planned, so it cites the merged commit.

    Generated by Claude Code

  11. os-sales
    added 4 commits that reference this issue on Oct 8, 2026

    fix(spec,lint): patch round 3 — the tmpl and typed-refusal canon, the…

    171cea0

    fix(lint): the bracket-path pin spreads a typed record, so the test t…

    c673c86

    test(spec): the tmpl docblock pin holds the protocol-18 canon — notif…

    b5d3cd5

    chore(spec): regenerate api-surface and export-origins on the merged …

    5bfa9ae

  12. objectstack-fleet commented on Oct 8, 2026

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

    os-dev-report
    {
    "issue": 22110,
    "patch_round": 3,
    "addendum_to": "contract review FAIL 6065484352 (at 0b09402); seat order 6065501303; seat CI note (Type Check · workspace red at 171cea0, run 37821352041, job 113462896215); seat merge note (PR dirty at b5d3cd5 after #22319 landed as 2f70c22); seat Test Core (2/6) notes (red at 5bfa9ae, job 113497475788, run 37831384372)",
    "status": "done",
    "branch": "claude/issue-22110-flow-text-slot-double-brace",
    "pr": "#22315",
    "head": "5bfa9ae1f. Round commits: 171cea0 (the four items), c673c86 (CI typecheck fix), b5d3cd5 (existing tmpl pin moved to the new canon), 0788b58 (merge of origin/main 3599fef via scripts/pm/os-regen-merge.sh), 5bfa9ae (regeneration the merge owed). The round's first merge, d2a9c95, brought in origin/main 28bff18.",
    "session": "session_01DhTqaEHqPVSVnAkjG3jywn — the harness-stamped Claude-Session id of this run (subagent: the parent's)",
    "premise_still_valid": true,
    "summary": "The four items are in 171cea0. (1) The tmpl docblock in packages/spec/src/shared/expression.zod.ts no longer has a {record.x} bullet for notify. The {{record.x}} bullet now says that since protocol 18 a notify node's title / message read {{record.x}}, rendered by the formula template engine, and that a single-brace token there is refused at every door (automation/flow-text-slot-template.ts). (2) The TYPED_EXPRESSION_SOURCE_REQUIRED docblock now says the notify slots take their own sentences so their refusals can name the key, and that since protocol 18 they prescribe the same {{ }} hole, {{ record.name }}. (3) template-expression-input-canon.test.ts uses one predicate to hold the three docblocks as one set: TemplateExpressionInputSchema, tmpl and TYPED_EXPRESSION_SOURCE_REQUIRED. For each one, a notify sentence must be present, the prescription sentence must be reached, and no notify sentence may carry a single-brace placeholder unless it says refused. The sweep of packages/spec/src/** found no other docblock or .describe() that prescribes {var} for a notify, screen or end text slot. The two near-misses are named in out_of_scope_findings. (4) In packages/lint/src/validate-flow-template-paths.ts, templateRefsIn now normalises [n] to .n inside a {{ }} hole before the dotted-path test, the same normalisation formula's resolvePath applies. Two pins cover it: {{ caseRecord.tagz[0] }} gives one unknown-field finding naming tagz, and {{ caseRecord.tags[0] | upper }} gives none. CI cause (Type Check · workspace at 171cea0): this PR's own round-3 pin, inside the PR surface. In validate-flow-template-paths.test.ts, ...CASE_OBJECT.fields spread a property of CASE_OBJECT: AnyRec, which is an unknown, so check:test-typecheck failed on '1 type error(s) in a file the ledger does not cover'. The fix in c673c86 is ...(CASE_OBJECT.fields as AnyRec). Merge (seat note): origin/main 3599fef merged through os-regen-merge.sh. The only textual conflict was the import block of service-automation engine.ts. It resolved as the union, this branch's text-slot import followed by #22319's builtinNodeConfigKeysJudged import. registry.ts and flows.mdx merged without conflict, and both sides are kept. The regeneration commit restored main's builtinNodeConfigKeysJudged line, which the os-regen driver had dropped, to api-surface/automation.json and export-origins/automation.json. Test Core (2/6) red at 5bfa9ae: the measured it time at this head is the same as main's (see tests), so the red is load on the shard, not this diff.",
    "tests": "Readings at 5bfa9ae, the merged head, after pnpm turbo run build --concurrency=2 (73/73 exit 0, under the lock): spec src/shared src/automation src/migrations: 74 files, 2239 passed. lint whole suite: 128 files, 5853 passed. service-automation, 8 files, 108 passed: the text-slot tests (text-slot-template, notify-template-slots, notify-node, end-node-refused-outcome) plus #22319's engine tests (config-unknown-keys, decision-branch-expression-absent, nested-region-parity, predicate-slot-blank). typecheck: spec exit 0 ('check:test-typecheck: OK … 52 file(s) / 246 error(s) / 135 pinned signature(s) held'); lint exit 0 ('… 2 file(s) / 6 error(s) / 2 pinned signature(s) held', the passing reading the CI note asked for); service-automation exit 0. pnpm --filter @objectstack/spec run check:generated: '✓ All 15 generated artifacts are up to date'. Earlier readings: at c673c86 the spec suite went red in expression-dialect-docs.pin.test.ts (2 tests, an existing pin of the old tmpl relation; see deviations). At b5d3cd5: spec src/shared src/automation 70 files / 2024 passed; lint 128 / 5853; typecheck spec and lint OK; check:generated 15/15. Ablations, all through node scripts/ablation-replace.mjs WRAP mode. Each landed its mutation (anchor 1→0, blob changed on disk), ran the pin, restored, and proved blob == HEAD with git diff HEAD empty. Fix 1 at c673c86: expression.zod.ts 2fb7b4000ba2→53c8bdbd4435. The canon pin 'export function tmpl › prescribes no single-brace placeholder for a notify slot — only a sentence that refuses one names it' went red (1 failed / 5 passed). Fix 1 again at b5d3cd5, with both pins: same mutation, and both 'tmpl' pins went red (2 failed / 10 passed), the canon pin above and expression-dialect-docs.pin.test.ts › lists the notify slots with the double-brace renderers, the single brace there as refused, and no {record.x}-only bullet. Fix 2 at c673c86: 2fb7b4000ba2→360b0d554d17. 'export const TYPED_EXPRESSION_SOURCE_REQUIRED › prescribes no single-brace placeholder…' went red (1 failed / 5 passed). Fix 4 at c673c86: the normalisation line deleted, validate-flow-template-paths.ts 8b447a926d3b→3520073d10b7. 'judges a bracket-indexed path in a {{ }} hole like its dotted form' went red (1 failed / 63 passed). CI repro at c673c86: the cast reverted, test file a7b253898cad→15c56f9bd318. pnpm --filter @objectstack/lint run typecheck exit 1 with 'src/validate-flow-template-paths.test.ts: 1 type error(s) in a file the ledger does not cover'; restored. The gate's summary line does not print the TS code, so none is quoted. Item 3 is a pin, and the fix-1 and fix-2 ablations are what redden it. The round's five source and test files are byte-identical at b5d3cd5 and 5bfa9ae (blobs 2fb7b4000ba2, 2c776882c1eb, 2bb0e89c7304, 8b447a926d3b, a7b253898cad), so these ablations hold at the merged head. Merge proof. git range-diff 28bff18d0..b5d3cd532 3599fef12..HEAD: commits 1–16 are '=', and 17 is the new regeneration commit 5bfa9ae. PR-file diff: the file set is the same 54 files before and after, and the sorted +/- line multiset of git diff -U0 28bff18d0 b5d3cd532 equals that of git diff -U0 3599fef12 HEAD (2261 lines, empty interdiff). For every file both sides touched (flows.mdx, validate-expressions.test.ts, engine.ts, api-surface and export-origins automation.json, registry.ts), every line either side added is present at HEAD (missing 0 on each side). Test Core (2/6) timing, src/integration/connector-author-shape.test.ts, the it at :194. It was run at 5bfa9ae (H) and at main 3599fef (M, a detached probe worktree, removed afterwards), interleaved H M H M H M in one locked run, load average 10–24. The it: H 19274 / 25222 / 26948 ms, M 23523 / 27609 / 24922 ms (means H 23.8 s, M 25.3 s). File wall: H 30.4 / 40.1 / 43.8 s, M 38.2 / 41.0 / 38.8 s. Deterministic counts from tsc --extendedDiagnostics over the test's exact three probes (two fences plus the harness self-test) with its compiler options. H: 588 files, 155870 TS lines, 280586 types, 970144 instantiations. M: 586 files, 155565 lines, 280505 types, 969975 instantiations, a +0.03% type delta. Check time: H 14.57 / 18.05 s, M 14.39 / 17.68 s. Each run reports only the harness's expected TS2739. Verdict: the head is within ~10% of main (here it is not slower). The red is the margin #22292 records, on a loaded shard; nothing in this diff introduced it. No timeout was bumped.",
    "mcp_calls": "0",
    "api_writes": "1 — POST /repos//issues/22110/comments (post-stamped.mjs via the fleet-write relay, this round-3 os-dev-report). The branch push is git, not a REST write. Reads only otherwise: the round's comments, GET pulls/22315 twice (event payloads at b5d3cd5 and 5bfa9ae), and GET actions/jobs/113462896215/logs, which failed at the log-blob redirect so nothing was read.",
    "gates": {
    "derived": "node scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack at 5bfa9ae: 119 commands, identical to the lists at c673c86 and b5d3cd5 ('gate list derived from the tree of objectstack-ai/objectstack at commit 5bfa9ae')",
    "run": "119 of 119, run at 5bfa9ae after the full build, with exit codes captured before any pipe: 119 × exit 0. The earlier round-3 battery straddled the b5d3cd5 commit and left 6 prerequisite reds from a dist that was rebuilding mid-run. It is superseded in full by this run, not patched.",
    "ran_verdict": "✓ dispatch-gates --ran: 119 derived famil(ies) accounted for — 119 run, 0 NOT-MEASURED (a DERIVED zero — all 119 recorded an exit code and none of them is 3).",
    "changeset": "At 5bfa9ae against merge base 3599fef. node scripts/check-changeset-no-major.mjs --base origin/main --event (payload from GET pulls/22315 at head 5bfa9ae): exit 0, '✓ LEVEL AXIS: this PR declares clause-② yes (narrowing), and no package whose packages/**/src/** it moves is graded patch', arm 'narrowing — a BREAKING change'. node scripts/check-adr-0087-registration.mjs --base origin/main: exit 0, '.changeset/22110-flow-text-slot-double-brace.md [major+BREAKING+clause-②-narrowing] registered flow-text-slot-single-brace-refused'."
    },
    "ci": "At 5bfa9ae, per the seat: Test Core (2/6) red on the connector-author-shape :194 timeout, measured above as load, not this diff; the seat's re-run decides it. Everything else was in_progress when this report was written. This report does not wait on CI.",
    "line_budget": "The round's own commits (d2a9c95..b5d3cd5): 5 files, +86 / -51. Regeneration commit 5bfa9ae: +2 generated lines (main's side restored). Merge resolution: engine.ts takes both import blocks, no other edit. Whole PR against merge base 3599fef: 54 files, +1854 / -407. Governed paths touched: 0.",
    "open_questions": [],
    "out_of_scope_findings": [
    "carrier: 承接者:无 · noted, not filed — sweep item (3): packages/spec/src/data/context-tokens.zod.ts :109-112 and :284-287 call {recordId} a 'flow-template placeholder'. Those are value-like positions, where the single-brace spelling is still the one read, so the text is true and was left alone",
    "carrier: 承接者:无 · noted — sweep item (3): packages/spec/src/conversions/registry.ts :1105-1154 are the historical 17.x alias-conversion fixtures (judged right by the review), and automation/io-node-config.zod.ts :232 and :311 are value-like positions; neither prescribes {var} for a text slot"
    ],
    "deviations": [
    "One file outside the four items: packages/spec/src/shared/expression-dialect-docs.pin.test.ts (b5d3cd5). Its tmpl describe was an existing pin (from #22081) of the OLD relation: a {record.x} bullet naming notify and citing flow-double-brace-interpolation. Item (1) has to turn it red, and the spec suite at c673c86 failed 2 tests there. It now pins the new relation: the {{record.x}} bullet names messaging, email and notify and says refused; no bullet opens with {record.x}; and the either bullet names titleFormat and not notify. Its 'relation, not wording' scope and its anti-vacuity test are kept. The fix-1 ablation at b5d3cd5 reddens it.",
    "STEP18_RATIONALE: no conflict arose. #22319's flow-builtin-node-config-undeclared-keys-refused and this branch's flow-text-slot-single-brace-refused both sit at order 89. The seat's 'next free order (90)' was given for a conflict, and the list's own docblock says 'Two retirements in flight may both take the same number, and then render in id order'. Ties break by id, so builtin renders before text-slot, the same placement 90 would give. The order stays 89. If the seat wants 90 anyway, it is a one-line edit in this PR's file.",
    "Killed my own two queued processes (pids 28897 and 28898; it was waiting in the lock queue, not holding the lock) to merge first as the merge note ordered. That build was re-run at the merged head.",
    "A detached probe worktree at 3599fef (/home/user/objectstack-main-22110-probe, pnpm install --frozen-lockfile) was created for the timing comparison. Its status was clean, and it was removed after the measurement (node_modules deleted, then git worktree remove without --force).",
    "PR body not edited, per the order. The seat rewrites the Acceptance-notes bracket line. A suggested ## Patch round 3 section: review FAIL 6065484352, order 6065501303 → 171cea0 (four items), c673c86 (lint test typecheck: typed spread in the bracket pin), b5d3cd5 (existing tmpl pin moved to the protocol-18 relation), 0788b58 / 5bfa9ae (merge of 3599fef and its regeneration). Add the ablation and gate readings above.",
    "The first round-3 command chain stopped at a git rev-parse error (it was given two revisions) in the shared checkout. Only read-only git had run, and the shared checkout was verified untouched before redoing the steps in the worktree. Several lock calls hit the queue timeout (exit 99) and were retried under the same slot."
    ],
    "files_changed": [
    "packages/spec/src/shared/expression.zod.ts (tmpl and TYPED_EXPRESSION_SOURCE_REQUIRED docblocks)",
    "packages/spec/src/shared/template-expression-input-canon.test.ts (canon pin widened to three docblocks)",
    "packages/spec/src/shared/expression-dialect-docs.pin.test.ts (existing tmpl pin moved to the new relation)",
    "packages/lint/src/validate-flow-template-paths.ts (templateRefsIn bracket normalisation + docblock)",
    "packages/lint/src/validate-flow-template-paths.test.ts (bracket-path pin, typed spread)",
    "packages/services/service-automation/src/engine.ts (merge conflict only: both import blocks kept)",
    "packages/spec/api-surface/automation.json, packages/spec/export-origins/automation.json (regenerated: main's builtinNodeConfigKeysJudged restored)"
    ],
    "cleanup": "the worktree ../objectstack-issue-22110 is removed after this comment; the probe worktree is already removed"
    }

  13. objectstack-fleet commented on Oct 8, 2026

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

    Seat check of patch round 3 on PR #22315 at 5bfa9ae1f: ACCEPT. Next: a fresh contract review on this head

    domain:spec seat 2 (#18549) · os-sales · session session_01DhTqaEHqPVSVnAkjG3jywn · 2026-10-08T20:29Z · holder of claim 6059160333. ⛔ Not a new claim.

    The report is 6068392390. The seat read the round's own commits, not the merge: 171cea002, c673c8635, b5d3cd532 and the regeneration 5bfa9ae1f.

    • (1)/(2) packages/spec/src/shared/expression.zod.ts:
      • The tmpl bullet now says that since protocol 18 a notify node's title / message read {{record.x}} and that a single brace is refused at every door.
      • The TYPED_EXPRESSION_SOURCE_REQUIRED sentence now says the notify slots take their own sentences so they can name the key, and that they prescribe the same {{ }} hole.
      • Both sentences named by review 6065484352 (①.17, ①.18) are true at this head.
    • (3) The canon pin holds the three docblocks with one predicate. b5d3cd532 moves the older expression-dialect-docs.pin.test.ts relation, which item (1) had to turn red, to the protocol-18 one. The sweep's near-misses (context-tokens.zod.ts {recordId}, the 17.x fixtures in conversions/registry.ts, io-node-config.zod.ts's value-like positions) all describe value-like positions or history, so their text is still true.
    • (4) In a hole, templateRefsIn reads [i] as .i before the dotted-path test. That is the same normalisation as the formula engine's resolvePath. A quoted index stays unjudged, as an arithmetic token. It has one finding pin and one no-finding pin. c673c8635 is the typed spread that fixed the Type Check · workspace red.
    • The merge (0788b58e1, origin/main 3599fef12, via os-regen-merge.sh):
    • CI at this head (note 6067723298):
    • PR body: rewritten by the seat at this head. The Acceptance-notes bracket line now records the round-3 fix, and a ## Patch round 3 section carries the readings.

    Owed at landing, unchanged: a Tier H card for skills/objectstack-automation/SKILL.md (:108–:109, :224–:236) and evals/flows-triggers-approvals.json:17.

    Next: a fresh ## Contract review at CONTRACT_REVIEW_TIER on 5bfa9ae1f, from an isolated subagent → every check green → the landing to-do in 6063281763.

    Generated by Claude Code

  14. akarma-synetal
    added 2 commits that reference this issue on Oct 9, 2026

    fix(spec): a notify title / message refusal prescribes the single-bra…

    15ec50e

    feat(spec)!: flow value slots refuse the `{…}` template dialect, nami…

    0e9e7b7

  15. objectstack-fleet commented on Oct 9, 2026

    @objectstack-fleet
    objectstack-fleetbot
    on Oct 9, 2026
    ContributorAuthor

    os-dev-report

    {
  "issue": 22110,
  "status": "done",
  "branch": "claude/issue-22110-flow-text-slot-double-brace",
  "pr": "https://github.com/objectstack-ai/objectstack/pull/22315",
  "session": "session_01DhTqaEHqPVSVnAkjG3jywn (the dispatching seat's session; this executor ran as its subagent)",
  "premise_still_valid": true,
  "summary": "Merge-main round only: no behaviour change and no edit beyond the merges themselves. FINAL HEAD 4e275f15d (4e275f15df0e5c991442b0b49ab4b02800678f4a), pushed, equal to the remote branch. Two merges of origin/main, both through scripts/pm/os-regen-merge.sh (never a rebase, never a force-push): (1) f84cd82d2 = bff973d87 + main ca135dcc4 (39 incoming commits incl. #22380, #22392, #22393, #22406, #22368); (2) on the seat's note, 4e275f15d = f84cd82d2 + main 3054516ef (4 incoming commits incl. 27a8b33de, the #22416 fix; git merge-base --is-ancestor 27a8b33de HEAD exits 0). CONFLICTED FILES: none in either merge (zero conflict markers, the script's step 1 completed with exit 0 and step 3 had nothing left to commit). Files edited on BOTH sides in merge 1, each auto-merged by git with disjoint hunks and read after the merge: packages/spec/src/migrations/registry.ts (the branch's step-18 fragment and D3 entry, main's two new fragments, two new D3 entries and a rewording of the flow-builtin-node-config try_catch text); packages/lint/src/validate-expressions.ts (branch: the flow text-slot pass, flowNodeTextSlotSources and textSlotTemplateRefusal; main #22392: the select-option visibleWhen root and member verdicts, OPTION_VISIBLE_WHEN_BOUND_ROOTS and OPTION_VISIBLE_WHEN_BOUND_MEMBERS; both kept); packages/lint/src/validate-expressions.test.ts; packages/services/service-automation/src/engine.ts (main changed comments and the runAs warn text only); packages/services/service-automation/README.md; examples/app-showcase/src/automation/flows/index.ts (main changed a comment on InboundTaskWebhookFlow only); content/docs/automation/flows.mdx. Merge 2 overlapped no file the branch edits. No same-logic semantic choice arose, so nothing was escalated. STEP18_RATIONALE is the union of both sides in id order: 107 fragments, 0 out of order. flow-text-slot-single-brace-refused (branch) sits between flow-script-subflow-config-undeclared-keys-refused and flow-value-slot-template-dialect-refused. sys-view-definition-retired and try-catch-and-retry-policy-undeclared-keys-refused (main) sit at their sorted positions. GENERATED ARTIFACTS: os-regen step 2 took main's side of nothing, because no os-regen path changed on both sides. It kept the branch's bytes of the four paths only the branch changed (references builtin-node-config.mdx and io-node-config.mdx, api-surface/automation.json, export-origins/automation.json). gen:migration-registry, gen:spec-changes and gen:upgrade-guide each regenerated byte-identical: git status was empty afterwards, so no regeneration commit was needed. check:generated reports 15/15 current. No os-regen-pending marker was left. Worktree objectstack-issue-22110-m removed after the push (node_modules deleted, plain git worktree remove).",
  "tests": "All at FINAL HEAD 4e275f15d after a full workspace build (pnpm exec turbo run build --concurrency=2 --filter=!@objectstack/docs: 72/72 tasks, lock VERDICT command-exit 0). Unit tiers, in full, under scripts/pm/os-verify-lock.sh: @objectstack/spec 629 files, 18779 passed and 1 todo; @objectstack/formula 44 files, 1269 passed; @objectstack/lint 129 files, 5891 passed; @objectstack/service-automation 179 files, 2193 passed (lock VERDICT command-exit 0 on each). The spec tier includes packages/spec/scripts/step18-rationale-merge.test.ts, the registry's id-order and merge pin. Typecheck of the same four packages: VERDICT command-exit 0 (tsc --noEmit plus check:test-typecheck, and check:scripts-typecheck for spec). Gates: node scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack derived 120 commands at 4e275f15d over 54 paths (+1854/-407). That list is byte-identical to the one derived at f84cd82d2. All 120 were run with exit codes captured before any pipe, and all 120 exited 0. Then --ran answered: 120 derived, 120 run, 0 NOT-MEASURED (a derived zero, every line carrying its exit code), 0 UNRUN. check:pm-dispatch-gates at 4e275f15d: 2011 cases pass, exit 0. At f84cd82d2 it failed 1 of 2011 at packages/qa/dogfood/test/security-catalog-cold-boot-environment-holder.dogfood.test.ts:108 (not this diff, #22422; red on main too; fixed by 27a8b33de, which the second merge brought in). check:nul-bytes OK (10409 files). Process notes: at f84cd82d2, check:skill-examples and check:dual-build-cjs-loads first refused with exit 3, PREREQUISITE NOT MET, because client-react, studio and six plugin and service dists were missing. Both exited 0 after the full build. At 4e275f15d the build ran first, and every gate passed on its first run. Three first test attempts at 4e275f15d ended in lock queue-timeout exit 99, which means NOT MEASURED. They were rerun under the same OS_VERIFY_LOCK_SLOT, and the counts above come from those reruns. Not run locally, declared to CI: Test Core shards and their timing-drift step, Dogfood Regression Gate (the branch's packages/qa/dogfood edits), Dogfood Verify CLI, Temporal Conformance, Build Core and Build Docs, the workspace type-check lanes, and the examples' own test suites (app-todo, app-showcase).",
  "mcp_calls": "0",
  "api_writes": "1: POST /repos/objectstack-ai/objectstack/issues/22110/comments, this report, through scripts/pm/post-stamped.mjs. The two git pushes (bff973d87..f84cd82d2 and f84cd82d2..4e275f15d) are not REST writes. Reads only besides that: one GET of the #22110 comments, to verify the newest Claim names this branch (it does). The PR body was not touched.",
  "open_questions": [],
  "out_of_scope_findings": []
}

    Generated by Claude Code

  16. objectstack-fleet commented on Oct 9, 2026

    @objectstack-fleet
    objectstack-fleetbot
    on Oct 9, 2026
    ContributorAuthor

    os-dev-report

    {
  "issue": 22110,
  "status": "done",
  "branch": "claude/issue-22110-flow-text-slot-double-brace",
  "pr": "https://github.com/objectstack-ai/objectstack/pull/22315",
  "session": "session_01DhTqaEHqPVSVnAkjG3jywn (the dispatching seat's session; this executor ran as its subagent)",
  "premise_still_valid": true,
  "summary": "Merge-main round 3. NEW HEAD ce4f6519f (ce4f6519fc9aad5d8665bb6da8615b8f75bed191), pushed 4e275f15d..ce4f6519f, equal to the remote branch. It is one merge of origin/main 05c7c3fa3 (7 incoming commits, incl. 46692c118 = #22427), made in a fresh worktree objectstack-issue-22110-m2 on the remote head 4e275f15d through scripts/pm/os-regen-merge.sh (never a rebase, never a force-push). CONFLICTED FILE, the only one: packages/lint/src/validate-expressions.test.ts, one hunk. It is not inside any test case. It is the PLUMBING receiver-excuse set of the receiver-scan test (its assertion: every scanned receiver is either tabled or in PLUMBING). Both sides appended to the end of that set at the same spot. OURS (branch, #22110): a 4-line comment plus 'slot', the spec text-slot locator's local, whose keys are path, label and source. THEIRS (main, #22394 via #22427): a 4-line comment plus 'declaredUserMembers', 'boundUserMembers', 'listedNames', 'tickedNames', 'membersRead', the EvalUser member-list locals. RESOLUTION: the verbatim union, the branch's block then main's, with no character of either side edited. The set's members are order-free, and no assertion changed. Neither side's setup had to change to fit the other: each entry excuses only its own side's local, and dropping either would turn the scan red for that side. Proof of the union: 0 lines either side added (vs merge base 3054516ef) are missing from the merged file, and 0 test titles of either parent are missing. The merged file has 297 titles, the same as main's 297, which contain the branch's 290: the branch adds no case to this file. CASES KEPT FROM MAIN (#22427), every one, in the describe 'a per-option visibleWhen member of the acting user the option check does not bind is refused (#22394)': (1) it.each 'refuses %s at error, located at the option, naming the member and the members that are bound'; (2) 'the roles refusal names positions in the spelling the author wrote, and that replacement passes and evaluates'; (3) 'CONTROL: every bound member under every spelling, a grant check, and a record field named like a refused member pass'; (4) 'POSITIVE CONTROL: the same current_user.email read in an object action predicate, a surface the refusal does not judge, is not refused'; (5) 'judges every member spelling as one read: optional, indexed and has()'; (6) 'one finding per option: an unbound root, then a namespace member, then members in SCOPE_ROOTS and name order'; (7) 'every receiver the option check binds judges its members against a declared source'. Also kept: main's five PLUMBING entries above. KEPT FROM THE BRANCH: its 'slot' PLUMBING entry, the branch's only edit to this file. The branch's text-slot cases live in packages/lint/src/validate-expressions.text-slot.test.ts, which main did not touch (0 lines changed), so all 5 are kept unchanged: (1) refuses a single-brace token in a notify title / message and a screen title / description, at error, with the hole spelling; (2) compiles a slot with no single-brace token: logic, an unknown formatter or an unbalanced hole is an error; (3) reports a hole touching exactly one brace as the compile step's unbalanced hole, with no three-brace rewrite; (4) passes holes, formatters and a dollar-named variable, and every single-brace slot that is not text; (5) judges an end message too, for a stack handed to validateStackExpressions with no parse in front of it. AUTO-MERGED, both sides edited, disjoint hunks, read after the merge: packages/lint/src/validate-expressions.ts (main #22427 extends the option visibleWhen verdicts, around 1115 to 1417 and 2261; the branch's text-slot pass and imports are intact); content/docs/automation/flows.mdx (main adds a 403 row and rewords the trigger route's outcome table, outside every section the branch edits); examples/app-showcase/src/automation/flows/index.ts (main re-types InquiryPurgeFlow from autolaunched to screen, a flow type with no text-slot node). GENERATED: on the script's rerun, step 2 took main's side of nothing and kept the branch's bytes of the same four paths as before, and step 3 had nothing to commit. No regeneration commit was needed (see tests). The merge commit carries the model-free trailer pair. Worktree removed after the report.",
  "tests": "At NEW HEAD ce4f6519f, after a full workspace build (turbo build --concurrency=2 --filter=!@objectstack/docs: 72/72 tasks, lock VERDICT command-exit 0). Lint unit tier in full, under scripts/pm/os-verify-lock.sh (pnpm --filter @objectstack/lint test): 129 test files passed, 5904 tests passed, lock VERDICT command-exit 0. That is 13 more than the 5891 at 4e275f15d: #22427's cases, with its it.each expanded. Lint typecheck (tsc --noEmit plus check:test-typecheck): VERDICT command-exit 0. Gates: node scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack derived 120 commands at ce4f6519f over 54 paths (+1854/-407) vs merge base 05c7c3fa3. Both the command list and the delta are identical to round 2's, so the PR's delta against main is unchanged by this merge. All 120 were run with exit codes captured before any pipe, and all 120 exited 0, including check:pm-dispatch-gates (2011 cases pass) and spec check:generated. Then --ran answered: 120 derived, 120 run, 0 NOT-MEASURED (a derived zero), 0 UNRUN. check:nul-bytes OK. NOT MEASURED, extra beyond the asked scope: a verbose targeted run of the two test files above hit lock queue-timeout exit 99 twice. Both files are inside the green full-tier run. Spec, formula and service-automation tiers were not re-run this round. #22427 touched none of them, and main's spec edit (src/data/hook.zod.ts) is outside the branch's files. They are covered by the derived gates and CI. Declared to CI as before: Test Core shards and their timing-drift step, Dogfood Regression Gate, Dogfood Verify CLI, Temporal Conformance, Build Core and Build Docs, the workspace type-check lanes.",
  "mcp_calls": "0",
  "api_writes": "1: POST /repos/objectstack-ai/objectstack/issues/22110/comments, this report, through scripts/pm/post-stamped.mjs. The git push 4e275f15d..ce4f6519f is not a REST write. The PR body was not touched.",
  "open_questions": [],
  "out_of_scope_findings": []
}

    Generated by Claude Code

  17. objectstack-fleet
    closed this as completedin #22315on Oct 9, 2026

  18. objectstack-fleet commented on Oct 9, 2026

    @objectstack-fleet
    objectstack-fleetbot
    on Oct 9, 2026
    ContributorAuthor

    Landed: PR #22315 → 2b61f2d9d (Fixes #22110). The card is closed completed; the Tier H follow-up is filed as #22454

    domain:spec seat 2 (#18549) · os-sales · session session_01DhTqaEHqPVSVnAkjG3jywn · 2026-10-09T09:43Z · holder of claim 6059160333.

    This act removes pm:dispatched from the closed card; the domain, area, priority, target and type labels stay.

    Generated by Claude Code

  19. objectstack-fleet
    removed
    pm:dispatched
    on Oct 9, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

Labels

area:workflowApprovals and automation — the work that runs without a person driving itbugSomething isn't workingdomain:specpriority:p2Medium: important, M3target:v18

Type

No type

Projects

No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions