[objectstack-fleet]

Requested by Tesla · project thread

Fixes #30759

Clause-②: no
Metadata-protocol: none

Text only; the runtime is unchanged. Triage 6674582930 routed the card to domain:spec as one PR over every surface that carries the wording, so no second dialect opens. The notify-node.ts hunks are domain:server code carried by the spec lane per that triage.

维护者速读

• 改了什么:notify 节点"不写链接就默认链到触发记录"的四处文案(spec 的 describe 与 docblock、服务端 notify 节点的 docblock 与配置说明、Flows 文档、automation skill 示例注释)现在写明:before-create 运行的记录还没有 id,只有调用方自己写了 id 时才会链接;sourceId 求值为 null 是"写了键但为空",半对被丢弃、不回退到默认。参考页 io-node-config.mdx 随之再生成。
• 为什么改:原文把"删除以外的记录变更"都说成会链接,但 before-create 运行在驱动发 id 之前执行,通知其实不带链接;作者或 AI 会据此省掉 pair 而等不到链接。
• 风险与回滚:零运行时改动;skill 文件净增 0 行、token 6443→6446(上限 6449)。回滚即 revert。
• 先看:packages/spec/src/automation/io-node-config.zod.ts(docblock + 两条 describe)、packages/services/service-automation/src/builtin/notify-node.ts(docblock、null 失败文案、两条 description)、skills/objectstack-automation/SKILL.md :125。
• 两分钟验证:pnpm --filter @objectstack/service-automation exec vitest run src/builtin/notify-node.test.ts,预期 25/25 通过,其中两条新 pin:无 id 的 beforeInsert 运行不写 source,调用方写了 id 的则链接。
• 需要你做的:Tier H(skills/**)——席位复审后交 os-zhuang 批准。

What changes

• The shared wording. The four texts now enumerate the runs that link as "a record change other than a delete or a before-create, a time-relative sweep, an action on a record", and each says why a before-create run is out: its record is the write's input, and the driver issues the id at the write, after the hook, so it links only when the caller wrote the id (the card's precision note: the input carries an id only when the caller wrote one).
  • packages/spec/src/automation/io-node-config.zod.ts: the NotifyConfigSchema docblock bullet and the sourceObject / sourceId describes.
  • packages/services/service-automation/src/builtin/notify-node.ts: the defaultClickThroughOf docblock and the sourceObject / sourceId config descriptions (the wording source).
  • content/docs/automation/flows.mdx: the click-through paragraph.
  • skills/objectstack-automation/SKILL.md :125: // trigger record (not on delete/before-create), if any — the comment edited in place, net +0 lines, no re-wrap.
  • content/docs/references/automation/io-node-config.mdx: regenerated from the describes by check:generated --fix, not hand-edited.
• The null clause. notify-node.ts's sourceId failure text and flows.mdx said a null "leaves the key unset". Both now say it is a written sourceId that resolves to nothing, so the pair is dropped rather than defaulted — what defaultClickThroughOf does, since it reads the written config, not the evaluated value.
• Pins in the existing notify click-through defaults to the trigger record describe of notify-node.test.ts: a notify on a beforeInsert run whose record carries no id writes no source (so no source_object / source_id) and no URL; one whose caller wrote the id links to it.
• Changeset .changeset/30759-notify-click-through-before-create.md: @objectstack/spec patch, @objectstack/service-automation patch, @objectstack/skills patch.

Measured on origin/main 6ff27b3 before editing: record-before-create → beforeInsert and record-before-write → beforeInsert + beforeUpdate (packages/triggers/trigger-record-change/README.md mapping table; record-change-trigger.ts :520 treats beforeInsert as the insert event), so the insert leg of record-before-write and record-before-create are the only before-create shapes. On that leg buildContext has no ctx.result and no ctx.previous, so record is ctx.input.data (:477-485); driver-memory create issues id: data.id || this.generateId(object) at the write (:845); triggerRecordOf needs both context.object and context.record.id (:155-159).

Why

A before-create run is "a record change other than a delete", so the shipped text promised a link the runtime cannot make: the notification links nowhere, not to a dead link. An author or an AI leaving the pair out on a before-create notify expected a link that never comes.

Tests

• pnpm --filter @objectstack/service-automation exec vitest run --maxWorkers=2 src/builtin/notify-node.test.ts under the verify lock: Test Files 1 passed (1), Tests 25 passed (25) (VERDICT command-exit 0).
• Reverse verification of the new pin, committed fix first: triggerRecordOf mutated on disk to return object ? { object, id: id ?? 'PIN_ABLATION_ID' } : undefined (1 hit of the marker, 0 of the original line), same run → Tests 2 failed | 23 passed (25): writes no source on a before-create run whose record has no id yet and links nowhere on a run with no trigger record both red (VERDICT command-exit 1). Restored with git checkout HEAD -- <path>: git hash-object equals the HEAD blob, git diff HEAD empty; rerun 25/25.
• pnpm --filter @objectstack/service-automation typecheck: VERDICT command-exit 0.
• Affected packages' own suites under the lock: pnpm --filter @objectstack/spec typecheck VERDICT command-exit 0; pnpm --filter @objectstack/spec test VERDICT command-exit 0, Tests 25838 passed | 1 todo (25839); pnpm --filter @objectstack/service-automation test VERDICT command-exit 0, Tests 3431 passed (3431). test:repo (the vitest repo project) is CI's run (Test Core), not run here.

Verification

• pnpm --filter @objectstack/spec build (dts emitted 38/38), then check:generated --fix: 1 of 14 artifacts stale (check:docs), regenerated content/docs/references/automation/io-node-config.mdx (2 lines), nothing else moved.
• Gate families derived against the actual diff with node scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack at 87404ae (the final commit): 114 lines, every one run with </dev/null, exit codes landed on disk before any pipe, reconciled with --ran: 114 derived, 114 run, 0 NOT-MEASURED, 0 UNRUN, every line exit 0 on the final tree. Three lines first exited 3 (PREREQUISITE NOT MET, nothing measured): check-plugin-teardown-shape.mjs --self-test (its positive control is a commit the shallow clone could not reach; fetched that sha, rerun → exit 0, 48 cases), check:skill-examples and check:dual-build-cjs-loads (no dist/ for the unbuilt packages; pnpm build under the lock, VERDICT command-exit 0, rerun → exit 0: 283 prose examples type-check; 116 entries / 71 packages / 2456 cjs files load). The ran list is /mnt/project-files/spec-seat-1/out/30759-gates-ran-87404ae2.list.
• Local-only readings (not derived, left local): pnpm check:skill-frame-freshness exit 0; pnpm check:dev-prereqs exit 1 — "the workspace is not built" (71 of 106 packages have no dist in this worktree, which built only @objectstack/spec and the service-automation closure); a precondition of this box, not a finding about the diff.
• node scripts/check-skills-token-ratchet.mjs: skills/objectstack-automation/SKILL.md 6446 tokens (ceiling 6449; headroom 3), from 6443 on main.
• Tier: node scripts/pm/check-governed-merges.mjs --test skills/objectstack-automation/SKILL.md … → Tier H (skills/**). Changed lines: 76 additions + 22 deletions, generated file included.
• skills/** readings, lines: skills/objectstack-automation/SKILL.md 452 before / 452 after; the whole published catalog (11 SKILL.md files) 4525 before / 4525 after. Tokens (check-skills-token-ratchet): 6443 before / 6446 after.
• Open PRs touching the same files at dispatch time (gh api search/issues): docs(skills): the automation skill teaches the mail option formatter for a notice's select label and the active-member rule for a user target; the docs page gains organization-authored assignment rules #30866 (skills/objectstack-automation/SKILL.md, a different hunk at :127 and later; no overlap with :125) and docs(skills): the platform skill's operations reference says the official image refuses to start without its database and secrets #30875 (skills/objectstack-platform/**, no overlap). None touches notify-node.ts or io-node-config.zod.ts. The later lander merges main.

Acceptance notes

• Two unconsumed changesets from the PRs that wrote the wording, .changeset/30254-notify-default-link-skill.md and .changeset/30689-notify-click-through-default-describe.md, repeat the old enumeration in their prose. Not edited here (another PR's release input); this PR's changeset states the corrected enumeration, and the release compiles both. Noted for the release-notes compiler, not filed.
• packages/triggers/trigger-record-change/README.md's mapping table is the only in-repo text that states record-before-write fires on beforeInsert + beforeUpdate; the code spells the union through its hook registration, not a literal table. Read-only observation, not a defect.
• The spec describes and the notify-node.ts descriptions were not byte-identical before this PR (the spec adds "a literal," and "not defaulted"); this PR keeps each text's own frame and changes the shared enumeration and the before-create clause in all of them.

Human Minimum

• 1. Schemas start as Zod; packages/spec holds no runtime logic.
• 2. Every name I added uses the casing for its kind.
• 3. No formatter ran; pnpm lint is CI's run.
• 4. One issue, one branch off an up-to-date main, one pull request.
• 5. A change that publishes carries a changeset, and both declaration lines are filled in here and in the changeset.
• 6. packages/spec touched: its generated artifacts are regenerated and committed.
• 7. No hand edits to content/docs/references/, packages/*/CHANGELOG.md or content/docs/releases/.
• 8. Title, body and comments are in English.
• The Before Submitting list in CONTRIBUTING.md is done.

🤖 Generated with Claude Code

https://claude.ai/code/session_01L9qibwUogojquWFFVyz422

[github-actions]

📓 Docs Drift Check

3 anchor(s) derived from 2 changed package(s); no hand-written page names any of them, so this run has nothing to list — not a clean bill of health. This check sees only pages that NAME a derived anchor: one that documents this change in prose, or enumerates it in an authoring dialect, names none and stays invisible to it on every run.

What this run could not see

• the SDK route bridge reached 63 of 222 client-bound route-ledger rows — the other 159 have no registrar path: tail to select them, so pages documenting THEIR client methods cannot appear above, on this or any run. Of those 159: 0 are remediable by widening that discovery convention (an in-repo file declares the path; the convention did not scan it); 55 are structural — on a ledger where NOT ONE row is declared in-repo, so no discovery change reaches them at any price; 104 are undecided (no in-repo declaration, on a ledger that has other in-repo registrars — absence and an unreadable spelling are not distinguishable here). The rows themselves: node scripts/docs-audit/affected-docs.mjs --bridge-coverage
• a page that states a rule by its inputs shares no identifier with the emitter that implements the rule, so an emitter-only diff cannot list it — not on this run and not on any run. Measured on fix(driver-sql): emit varchar(maxLength) for a text field a declared index keys on #11430: content/docs/protocol/objectql/types.mdx documents the text-family column mapping by the ObjectQL type names it maps FROM (text / textarea / html) while the diff changed createColumn; it went unlisted, and it was the page that diff falsified, in four places. No shared token exists to detect this on, so a rule your change carries has to be re-read by hand in the pages that restate it.
• a key NAME is not a key, so the hand re-read the line above prescribes can land on the wrong schema. The same spelling is authorable on one governed type and a [REMOVED] tombstone on another for each of active, aria, joins, objects, template, tools and version (censused on [finding] tools is a key on BOTH AgentSchema (tombstoned, dead) and SkillSchema (live, cloud-attested), so a name-based search attributes skill examples to the agent key — it produced a false stop-the-line alarm on PR #19059 #19093 over the liveness ledger's governed types, top-level keys); nothing in a search result distinguishes the two, so a grep hit on a LIVE example reads as evidence about the DEAD key. Measured on fix(spec): the agent.tools liveness row says dead — it claimed live on a key the schema tombstoned #19059: content/docs/ai/agents.mdx was reported as contradicting the agent.tools tombstone over its tools: example at :161, which is inside the defineSkill({ block opened at :155 — the page was already correct. Settle ownership by PARSING the value against both schemas, never by the name: that literal PASSES SkillSchema, and as an AgentSchema it FAILS at tools with the tombstone prescription. ⛔ These names are not the whole class — a key retired through a .strict() guidance map leaves no tombstone in the walked shape and none of them here (tool.category, live as AIToolDefinition.category).

Coarse fallback — 170 page(s) merely mention a changed package (the pre-#9192 predicate, kept for the deliberately-wide backstop): node scripts/docs-audit/affected-docs.mjs --json 26c965cb8f61ca59804d55e27e737bf7ff91b383 → packageMentionDocs.

Which tree this was computed on

This run read content/docs from 6337c092fbfff6e1f12d1da144adb1ac43b05c7a — the merge of head 7cf467745e681280fc3a6a7130f10ec308c61c8c into base 26c965cb8f61ca59804d55e27e737bf7ff91b383, which is what actions/checkout gives a pull_request run. Not the PR head.

A worktree cut from an older main holds a different content/docs, so re-deriving there can legitimately return a different list — that is a different tree, not a wrong row. To answer on the same tree:

# while this PR is open — GitHub drops the merge commit once it closes
git fetch origin 6337c092fbfff6e1f12d1da144adb1ac43b05c7a && git checkout 6337c092fbfff6e1f12d1da144adb1ac43b05c7a
# afterwards, rebuild it from the two parents, which stay fetchable
git fetch origin 26c965cb8f61ca59804d55e27e737bf7ff91b383 7cf467745e681280fc3a6a7130f10ec308c61c8c && git checkout -B drift-repro 26c965cb8f61ca59804d55e27e737bf7ff91b383 && git merge --no-ff 7cf467745e681280fc3a6a7130f10ec308c61c8c

node scripts/docs-audit/affected-docs.mjs --json 26c965cb8f61ca59804d55e27e737bf7ff91b383

⚠️ That checkout carried uncommitted changes, so the commit above does not fully identify what was read.

[objectstack-fleet]

domain:spec#1 · os-tesla · session_01L9qibzzqHZdAVqgBAvxXKB

Contract review

Served-tier: CONTRACT_REVIEW_TIER
Head-sha: 87404ae27500d18285c27708b4383ab88581a28f
Local-runs: none
Scope: packages/spec/src/automation/io-node-config.zod.ts, skills/objectstack-automation/SKILL.md + packages/services/service-automation/src/builtin/notify-node.ts, packages/services/service-automation/src/builtin/notify-node.test.ts, packages/triggers/trigger-record-change/src/record-change-trigger.ts, packages/triggers/trigger-record-change/README.md, packages/objectql/src/engine.ts, packages/drivers/driver-memory/src/memory-driver.ts, packages/drivers/driver-sql/src/sql-driver-crud.ts, content/docs/references/automation/io-node-config.mdx, content/docs/automation/flows.mdx, skills/objectstack-automation/evals/flows-triggers-approvals.json, .changeset/30759-notify-click-through-before-create.md

Inputs: card #30759 (body + 4 comments: the sequencing note 6674423070, triage 6674582930, the claim 6683585123, the os-dev-report 6684368393), PR #30926 (body, 7-file list, net diff origin/main...87404ae2), the check-runs on 87404ae2. The dispatch order was not read.

① Derived judgments

• Accept-set of NotifyConfigSchema: unchanged — right. The hunks in io-node-config.zod.ts touch one docblock bullet (:645-657) and two .describe() strings (:766, :775). No key, optional-ness, union arm, default or superRefine moves; notifyLiteralStringSchema('sourceObject') and NotifySourceIdSlotSchema are the same schemas as on main. What a notify config parses to is identical before and after.
• Published surface: description text only — right. The two describes feed the JSON-schema description and the generated reference page; the regenerated content/docs/references/automation/io-node-config.mdx :106-107 carries the new describe strings byte-for-byte (read against the describes at the head). No field is added or removed from the reference table.
• Runtime: unchanged — right. defaultClickThroughOf (:183-191) and triggerRecordOf (:155-159) are the same code as on main; the one executable hunk in notify-node.ts is the text of a thrown failure (:712), a message, not a code. The node-registration description strings (:888, :892) are Studio form text.
• The enumeration against the tree — right. On a beforeInsert run buildContext has no ctx.result and no ctx.previous, so record is ctx.input.data (record-change-trigger.ts :477-485); the engine hands that hook input.data = the defaulted row (engine.ts :5540-5543), and applyFieldDefaults fills declared defaultValues only, never an id; the id is minted at the write by the driver (memory-driver.ts :845 data.id || this.generateId(object); sql-driver-crud.ts :259 toInsert.id = nanoid(...)); triggerRecordOf needs record.id. So "links only when the caller wrote the id" is what the tree does on both shipped driver families. record-before-create → beforeInsert and record-before-write → beforeInsert + beforeUpdate (trigger README :40, :43), so "a before-create" covers both shapes the card names.
• The null clause against the tree — right. resolveStringSlot returns value: undefined for a null envelope (:707), and defaultClickThroughOf is handed the parsed config, where cfg.sourceId != null is the written envelope (:188, :1027), so the default is suppressed and no source is written: "a written sourceId that resolves to nothing, so the pair is dropped rather than defaulted" is exact.
• Pins (lane seat's file, read as the first consumer of the claim). The new beforeInsert test feeds a context straight to engine.execute with record: { name: 'Pump' } and no id; it pins triggerRecordOf + defaultClickThroughOf, not the trigger's buildContext. Enough for the card's pin ("writes no source_object / source_id"); the trigger side is covered by the cited code, not by a test in this PR. The second pin (caller-written id links) is the converse. Right.
• Rest of the diff (flows.mdx, the test, the changeset, the regenerated reference): the lane seat's ACCEPT at TIER_DEFAULT.

② Semver level

Clause-②: no and Metadata-protocol: none on the PR body and in .changeset/30759-notify-click-through-before-create.md; the changeset bumps @objectstack/spec, @objectstack/service-automation and @objectstack/skills (packages/skills/package.json) as patch. The diff publishes description text and a failure-message string and nothing that changes an accept-set, so patch is the level and the Clause-②: no line is true. The fixed group in .changeset/config.json already ties spec and skills together. Right.

③ Boundary flags

open_questions: []. Deviations and out-of-scope findings in the os-dev-report, each answered:

1. check:dev-prereqs exit 1 before the workspace build — a box precondition, not a tree finding. Accepted.
2. Full workspace build taken to meet two gates' prerequisites — process, no tree effect. Accepted.
3. Two older unconsumed changesets (.changeset/30254-…, .changeset/30689-…) repeat "a record change other than a delete" — they are the release input of merged PRs and are compiled chronologically beside this one, whose prose states the correction. Not edited here; accepted, no escalation.
4. No GitHub MCP write tools; assignees/labels set at hand-over — process. Accepted.
5. Open PR docs(skills): the automation skill teaches the mail option formatter for a notice's select label and the active-member rule for a user target; the docs page gains organization-authored assignment rules #30866 touches skills/objectstack-automation/SKILL.md at other hunks and is still open at review time; no overlap with :125. The later lander merges main and carries the token ratchet (this PR reports headroom 3 on the skill). Accepted; the ratchet is the later lander's constraint, not this PR's.

• security_findings: none — agreed, prose only.
• README mapping-table observation — read, agrees with record-change-trigger.ts; not a defect.

Tier H (skills/**): approval by os-zhuang is owed after this record, per the claim; not this record's to give.

④ Design judgment

packages/spec/src/automation/io-node-config.zod.ts — judged against the card's ruling (four texts, one shared wording; a before-create run has no id until the driver issues one) and its first consumers (defaultClickThroughOf / triggerRecordOf in notify-node.ts, the generated reference page).

• The docblock bullet and both describes now say "a record change other than a delete or a before-create" and carry the reason and the exception ("…so it links only when the caller wrote the id"). The exception is what the tree does (① above), and it is the one clause the card's own suggested wording lacked; keeping it in the describe is right, because an author who writes id on the insert input does get a link.
• Tried to break it: a record-before-write flow on its insert leg, a batch insert (per-row beforeInsert, same input shape, :5482-5543), a caller-supplied id on the input, an afterInsert run (after-row carries the id, links as before), a beforeUpdate run (previous carries the id). The text holds for each.
• should-fix (a second dialect in the same file). :435, the notifyStringSlotSchema docblock that NotifySourceIdSlotSchema (:457) is built from, still reads "An envelope that evaluates to null or a blank string leaves the key unset, as a token that resolved to nothing did" — the exact wording the card's null clause retires, now sitting 200 lines above a describe that says the pair is "dropped … not defaulted". It is a docblock, not published through .describe(), and not one of the two locations the card named, so it is not a FAIL; but the changeset says "the schema say so", and the schema file now says both. One sentence ("…leaves the key unset; it is still a written key, so the pair is dropped rather than defaulted to the trigger record") closes it. The sibling docblock in notify-node.ts :694 says the same old sentence — cited code, for the lane seat, not reviewed here.
• should-fix (cosmetic). :653-654, the re-wrapped bullet ends a line on "caller wrote the id. An" and starts the next on "explicit pair …", a stranded two-word tail; re-wrap the bullet.

skills/objectstack-automation/SKILL.md — judged against the card's eval surface (evals/flows-triggers-approvals.json :248-259: a record-after-update notify that opens its case, expecting no sourceObject / sourceId / actionUrl) and the spec it teaches on main.

• The one edit, :125 // trigger record (not on delete/before-create), if any, is the shared enumeration compressed to a comment; the eval's case (record-after-update) is unaffected and the expected output still follows from the comment. The comment leaves out the exception (a before-create whose caller wrote the id does link). That reads as over-strict rather than false: an AI following it on a before-create flow writes the pair explicitly, which the runtime honours, or drops when record.id resolves to null — never a dead link, never a missing one it was promised. Under a 3-token headroom the compression is the right trade; accepted, not a finding.
• Net +0 lines, no re-wrap, so the open overlap with PR docs(skills): the automation skill teaches the mail option formatter for a notice's select label and the active-member rule for a user target; the docs page gains organization-authored assignment rules #30866 stays disjoint.

No must-fix.

Check-runs on 87404ae2 at the time of this record: 23 completed/success (among them Check Changeset, Governed Surface Queue Guard, PR-body guards, Spec property liveness, Lint · docs & code guards, Lint · ESLint, Type Check · source gates, Lint & Repo Gates, Build Docs, Dogfood Regression Gate), 8 completed/skipped (opt-in and gated jobs), 0 failure, 31 still in progress (Test Core 1-8, Test UI 1-16, Type Check · workspace / debt ledger / consumer gates, Type Check UI, Build Core, Build & E2E UI, Live E2E UI). Every derived gate family that has concluded is green; this verdict stands on those conclusions and on the tree read above. The in-progress runs are the seat's to read at posting: a failure landing on any of them is a gate verdict this record does not override.

Implemented-by: claude/project-thread-mkn6rg
Reviewed-by: session_01L9qibzzqHZdAVqgBAvxXKB

VERDICT: PASS

[objectstack-fleet]

domain:spec#1 · os-tesla · session_01L9qibzzqHZdAVqgBAvxXKB

Contract review

Served-tier: CONTRACT_REVIEW_TIER
Head-sha: 7cf467745e681280fc3a6a7130f10ec308c61c8c
Local-runs: none
Scope: packages/spec/src/automation/io-node-config.zod.ts, skills/objectstack-automation/SKILL.md + packages/services/service-automation/src/builtin/notify-node.ts, packages/services/service-automation/src/builtin/notify-node.test.ts, packages/triggers/trigger-record-change/src/record-change-trigger.ts, packages/triggers/trigger-record-change/README.md, packages/objectql/src/engine.ts, packages/drivers/driver-memory/src/memory-driver.ts, packages/drivers/driver-sql/src/sql-driver-crud.ts, content/docs/references/automation/io-node-config.mdx, content/docs/automation/flows.mdx, skills/objectstack-automation/evals/flows-triggers-approvals.json, .changeset/30759-notify-click-through-before-create.md

Delta review of head 7cf467745 over the earlier record on 87404ae2 (PR comment 6684513754, PASS with two should-fixes). One commit, 87404ae2..7cf46774, 2 files, +14/-11: packages/spec/src/automation/io-node-config.zod.ts (:432-439 docblock, :652-657 re-wrap) and packages/services/service-automation/src/builtin/notify-node.ts (:691-699 docblock). Nothing else moved between the heads: the describes, the generated reference page, flows.mdx, the test, the changeset and SKILL.md are byte-identical to 87404ae2. Inputs: the card, the PR body and file list, the delta diff, the net diff against origin/main, the check-runs on 7cf467745. The dispatch order was not read. Every judgment of the earlier record is re-read against the new head and carried below.

① Derived judgments

• Accept-set of NotifyConfigSchema: unchanged — right (carried). The delta edits two comment blocks and re-wraps a third; notifyStringSlotSchema (:443) and NotifySourceIdSlotSchema (:458) are the same union, superRefine and .meta() as on main. No describe string changes in the delta.
• Published surface: unchanged by the delta — right. Docblocks are not .describe() text, so no regeneration of content/docs/references/automation/io-node-config.mdx is owed and none happened; Check Changeset and Spec property liveness are green on the new head.
• Runtime: unchanged — right (carried). resolveStringSlot's code (:703-716) is untouched by the delta; only the docblock above it changes. defaultClickThroughOf (:183-191) and triggerRecordOf (:155-159) are as on main.
• The enumeration and the before-create clause — right (carried). Re-read against record-change-trigger.ts :477-485, engine.ts :5540-5543, memory-driver.ts :845, sql-driver-crud.ts :259, triggerRecordOf :155-159 and the trigger README :40/:43 — same tree as at the first record.
• The null clause — right, now in every text of the two files. The two docblocks that still carried "leaves the key unset" now say a null or blank envelope "is a written key that resolves to nothing: the click-through pair is dropped rather than defaulted to the trigger record". That is what the code does: resolveStringSlot returns value: undefined (:710), the executor is handed the parsed config in which cfg.sourceId is the written envelope (:188, :1031), so the default is suppressed and no source is written. The service docblock adds "(defaultClickThroughOf reads the written config)", which is :188's exact behaviour.
• Pins (lane seat's file) — right (carried). Unchanged between heads.
• Rest of the diff (flows.mdx, the test, the changeset, the regenerated reference): the lane seat's ACCEPT at TIER_DEFAULT.

② Semver level

Clause-②: no and Metadata-protocol: none on the PR body and in .changeset/30759-notify-click-through-before-create.md; patch on @objectstack/spec, @objectstack/service-automation and @objectstack/skills. The delta adds no published text (comments only), so the level and both lines stay true. Right (carried).

③ Boundary flags

open_questions: [] on the os-dev-report; no new dev report, deviation or question was filed for the delta commit. The five deviations and two out-of-scope notes answered in the earlier record are unchanged by the delta and stay accepted. PR #30866 is still open and still touches SKILL.md only at hunks disjoint from :125; the delta does not touch SKILL.md. Tier H (skills/**) approval by os-zhuang is owed after the seat's review, per the claim; not this record's to give.

④ Design judgment

packages/spec/src/automation/io-node-config.zod.ts — against the card's ruling and its first consumers (resolveStringSlot / defaultClickThroughOf in notify-node.ts, the generated reference page).

• Should-fix 1 of the earlier record (second dialect at :435): resolved. :435-437 now reads "is a written key that resolves to nothing: the click-through pair is dropped rather than defaulted to the trigger record". Tried to break it: with sourceObject also unwritten there is no pair to drop, but the sentence's effect (no link, and no default) is exactly the runtime's, since cfg.sourceId != null at :188 is true for the written envelope whatever sourceObject says. Holds. The file now carries one dialect for sourceId and null: the slot docblock, the click-through bullet (:652-657) and both describes agree.
• Should-fix 2 (re-wrap at :653-654): resolved. The bullet reads "caller wrote the id. An explicit pair or an explicit actionUrl is the author's word and wins, and a half-written pair is still dropped, never defaulted. The executor / owns the default …" — no stranded tail.
• :398, the recipients slot docblock, left as written — right. "An envelope that evaluates to null or a blank string adds no recipient, as a token that resolved to nothing did, and the executor's guard still fails a node left with none" is a claim about recipients, not sourceId, and the executor does exactly that: a null entry is skipped (if (item == null) continue, notify-node.ts :268) and a node left with no recipient fails on the post-evaluation guard. There is no default for a recipient to fall back to, so the "written key … dropped rather than defaulted" frame does not apply there; changing it would have been wrong. The same reading covers notify-node.ts :234 (recipients) and the general "post-evaluation guards still own 'resolved to nothing'" notes at zod :27 and notify-node.ts :923, which name the guards, not the sourceId default. No finding.
• No new finding on this file.

skills/objectstack-automation/SKILL.md — not touched by the delta; byte-identical to 87404ae2. The earlier judgment is carried: :125 // trigger record (not on delete/before-create), if any is the shared enumeration compressed under a 3-token headroom, over-strict rather than false against the spec it teaches and the eval at evals/flows-triggers-approvals.json :248-259 (record-after-update); an AI following it writes the pair on a before-create, which the runtime honours or drops to nothing, never a promised link that fails to come. Accepted, not a finding.

packages/services/service-automation/src/builtin/notify-node.ts (cited, read for the claim, not reviewed): the :694 sibling of should-fix 1, noted for the lane seat in the earlier record, is closed by the same commit and agrees with the spec's text.

No must-fix. No should-fix outstanding.

Check-runs on 7cf467745 at the time of this record: 8 completed/success (Auto Label, Check PR Size, Governed Surface Queue Guard, Check Changeset, Spec property liveness, Flag docs affected by code changes, PR-body guards, Check Documentation Links), 3 completed/skipped (opt-in and gated jobs), 0 failure, 13 in progress (Type Check · workspace / source gates / consumer gates / debt ledger, Lint · ESLint / docs & code guards / PM dispatch-gates / repo & tooling gates / CI, release & engine gates / scoped x3, filter; the Test Core and Test UI shards had not yet been scheduled). The earlier head's Test Core / Test UI runs were cancelled by the push that superseded it, not failed; on 87404ae2 every gate that concluded was green (35 success, 0 failure). This verdict stands on the conclusions that have landed and on the tree read above; the in-progress runs are the seat's to read at posting, and a failure landing on any of them is a gate verdict this record does not override.

Implemented-by: claude/project-thread-mkn6rg
Reviewed-by: session_01L9qibzzqHZdAVqgBAvxXKB

VERDICT: PASS

[objectstack-fleet]

domain:spec#1 · os-tesla · session_01L9qibzzqHZdAVqgBAvxXKB

维护者速读

• 改了什么：通知节点「三个链接键都不写就默认链到触发记录」的四处文案（spec 的 describe 与 docblock、服务端 notify 节点的说明、Flows 文档、automation 技能示例注释）改为同一说法：删除和 before-create 以外的记录变更才会默认链接；before-create 运行时记录还没有 id，只有调用方自己写了 id 才链接。sourceId 写成 null 是「写了键但为空」，链接对被丢弃，不回退到触发记录。参考页随之再生成，并新增两条测试钉住这两点。
• 为什么改：原文把所有非删除的记录变更都说成会链接，但 before-create 运行在驱动发 id 之前执行，通知其实没有链接。作者或 AI 会据此省掉链接键，然后等不到链接。
• 风险与回滚：只改文字和测试，运行时行为与 schema 接受集不变；技能文件净增 0 行（token 6443→6446，上限 6449）。回滚即 revert。
• 先看：skills/objectstack-automation/SKILL.md :125，packages/spec/src/automation/io-node-config.zod.ts 的 notify 链接段。
• 两分钟验证：pnpm --filter @objectstack/service-automation exec vitest run src/builtin/notify-node.test.ts，预期 25/25，含两条新 pin。
• 需要你做的：Tier H（改到 skills/**）。请以 os-zhuang 批准 PR docs(automation): the notify click-through default says a before-create run links to its record only when the caller wrote the id #30926；批准后席位落地。契约复审 PASS 记录 6684677358，Head-sha 7cf467745e681280fc3a6a7130f10ec308c61c8c。

席位意见：可以落地，待授权批准。

席位核对

ACCEPT · seat domain:spec#1 · head 7cf46774 · 2026-11-06T12:54Z

1. Shape: Fixes #30759, draft, base main, 7 files (changeset, content/docs/automation/flows.mdx, regenerated content/docs/references/automation/io-node-config.mdx, notify-node.ts + notify-node.test.ts, io-node-config.zod.ts, skills/objectstack-automation/SKILL.md), +89/−32. Three patch changesets (@objectstack/spec, @objectstack/service-automation, @objectstack/skills) with Clause-②: no / Metadata-protocol: none, matching claim 6683585123.
2. Ruling: the card's own wanted list (one PR, all four texts share one wording, the pin, the null clause), triage 6674582930. All four texts now say "a record change other than a delete or a before-create"; the null clause says the written key drops the pair.
3. Contract review: at CONTRACT_REVIEW_TIER by an isolated reviewer, scope = hits io-node-config.zod.ts + skills/objectstack-automation/SKILL.md plus the cited trigger, engine and driver code. PASS 6684513754 on 87404ae2 (two should-fixes: the notifyStringSlotSchema / resolveStringSlot docblocks still carried the retired null wording; a stranded line wrap), both fixed in 7cf46774; delta PASS 6684677358 on 7cf46774, no findings. Accept set of NotifyConfigSchema byte-identical.
4. Gates / check-runs: dev: 114 of 114 derived gates exit 0 on 7cf46774, --ran reconciled (ran list in the project files); notify-node 25/25, spec and service-automation suites + typechecks green (report 6684368393). On head 7cf46774: 65 runs, 57 success, 8 skipped, 0 failed, 0 running; mergeable_state clean. check-governed-merges --pr 30926: Tier H (skills/** ×1), 121 changed lines.
5. Accepted deviations: SKILL.md :125 compresses the caller-wrote-the-id exception out of the 3-token headroom (over-strict, never wrong; reviewer accepted). Neighbour PR docs(skills): the automation skill teaches the mail option formatter for a notice's select label and the active-member rule for a user target; the docs page gains organization-authored assignment rules #30866 edits the same SKILL.md at disjoint hunks.
6. Landing: Tier H: pm:awaiting-merge + review requested from the approvers; ⛔ no ready, queue or auto-merge before an APPROVED review from a GOVERNED_APPROVERS account. After it, the seat lands.