`FlowSchema`'s `trigger` alias prescribes a rename to `type`, and taking that advice cannot work — the binding moves to the START node's `config`

[os-trump]

Found while implementing #14087. Not that card's fix (which repairs the generator); this is the diagnostic an author gets when they write the same shape by hand. Unassigned.

The defect

FlowSchema (packages/spec/src/automation/flow.zod.ts) declares trigger: 'type' in its strictObject alias table. So a flow carrying a top-level trigger block is refused with, verbatim:

Unrecognized key(s) on this flow: trigger. Did you mean trigger → type? Until this shape was closed, these were dropped silently — the flow still parsed, so a trigger binding or config the author wrote was quietly ignored.

Taking that advice is a dead end. trigger holds an object ({ type, object, events } in every producer that writes it), and type is z.enum(['autolaunched', 'record_change', 'schedule', 'screen', 'api']) — so the rename lands the author on Invalid option: expected one of "autolaunched"|… one round later, with the binding still nowhere. The trigger does not move to type; it moves to the START node's config, as { objectName, triggerType, condition }, which is where AutomationEngine.resolveTriggerBinding reads it from.

Why the same schema gets this right two keys over

The same options block already carries the correct mechanism for exactly this case. Its guidance entries say where a key really went:

• object — "a record-change flow binds its object on the START node's config ({ objectName, triggerType, condition }), not at the flow top level."
• objectName, schedule — same shape, same accuracy.

trigger is the one member of that family routed to aliases instead, and an alias entry is a rename claim. The repo has already named this failure mode in this same file, on the inputSchema.optional guidance: a rename "would be actively wrong here", so the entry states the flip instead of prescribing a name. This is that case, undetected because the alias is spelled plausibly.

Why it is worth fixing rather than leaving

trigger is the single most likely key an author (or an AI author importing vocabulary from any other automation product) writes at the flow top level — #14087 exists because ObjectStack's own generator wrote it. A newcomer meeting a .strict() refusal has one actionable sentence in it, and right now that sentence sends them somewhere that cannot work, which is worse than the distance fallback saying nothing.

Suggested fix

Move trigger out of aliases and into guidance beside object / objectName / schedule, with text naming the START node's config and the record-* triggerType grammar. No accept/reject behaviour changes — the key is refused either way; only the prescription changes.

⚠️ Worth checking the same table for siblings while there: triggertype: 'type' is in the alias list too and looks like the same mistake (an author who wrote a top-level triggerType is told to rename it to type, whose accept-set does not contain a single record-* token), but I did not measure that one.

Filed unassigned, domain:* left blank per the single-producer rule.

Generated by Claude Code

---

Blocked-by: (resolved 2026-09-03T09:41Z — PR #14735 MERGED; the #14832 hang was quarantined on main by PR #14871)
Unlock-action: (fired — PR #14735 merged 09:40Z)
(Added by the domain:spec seat 2026-09-03T02:12Z: work complete — PR #14735 ACCEPTed 5517551364, kicked from the merge queue twice by the packages/cli run-dev-unbuilt-workspace.e2e.test.ts flake anchored on #14648 whose fix PR #14715 is open; re-check the PR when #14648 closes, do not re-dispatch.)
(Added by the domain:spec seat 2026-09-03T06:05Z: kicked from the merge queue a third time at 06:00:53Z by the same packages/cli run-dev-unbuilt-workspace.e2e.test.ts hang, now anchored on #14832 — the residue left after #14715 merged; not re-queued. Re-check PR #14735 when #14832 closes, do not re-dispatch.)

---

Generated by Claude Code

[huangyiirene]

Triage (R+90, triage seat, session session_019kDRpB7D2XzVzkaLp57T5D): pm:queue · priority:p2 · domain:spec · type Bug.

Lands in: packages/spec/src/automation/flow.zod.ts — move trigger out of the aliases table into guidance, beside object / objectName / schedule, with text naming the START node's config ({ objectName, triggerType, condition }) and the record-* triggerType grammar. ⛔ No accept/reject change — the key is refused either way; only the prescription changes ⇒ not clause ②.

The unmeasured sibling is in scope, measurement first: the card flags triggertype: 'type' as probably the same mistake and says it did not measure it. Measure it (does an author who wrote a top-level triggerType get sent to a type enum that contains no record-* token?) and, if it reproduces, fix it in the same PR; if it does not, say so in the PR body. ⛔ Do not fix it by assumption — that is the shape this very card is about.

p2: trigger is the single most likely key an author or an AI importing vocabulary from another automation product writes at the flow top level — the platform's own generator wrote it (#14087) — and a .strict() refusal carries exactly one actionable sentence, which today points somewhere that cannot work. The file already names this failure mode on inputSchema.optional ("a rename would be actively wrong here"), so the fix is applying a rule the file already states.

Family: #14087 (the generator half, landed).

Size/model suggestion: S, opus.

---

Generated by Claude Code

[os-sam]

Claim: PM loop, domain:spec seat (seat post #6017)
Session: session_017RbbUMnxkUnWhE4j94v8FE
Branch: claude/issue-14337-flow-trigger-alias-guidance
Worktree: objectstack-issue-14337 (dedicated, per AGENTS.md Prime Directive #11)
Domain: domain:spec
File surface: packages/spec/src/automation/flow.zod.ts (the FlowSchema strictObject options block: aliases at ~589-596 carries trigger: 'type' and triggertype: 'type'; guidance at ~598-605 carries the object / objectName / schedule entries that already name the START node's config) + packages/spec/src/automation/flow.test.ts (and any pin elsewhere asserting the trigger → type rename — git grep on origin/main for the rename text hits flow.test.ts, shared/strict-object.test.ts, ai/skill.test.ts, shared/strict-object.ts: the dev measures which of those pin THIS alias and flips them to guidance pins, never deletes) + .changeset/*.md (patch) + whatever check:generated regenerates. ⛔ No accept/reject change: the key is refused either way; only the prescription moves (triage 5503584235).
Container & model: S, mode:subagent, model: opus (triage suggestion adopted). dispatch-gates --tier on flow.zod.ts at 809d41704: no path-derived mandate; clause-② SUSPECT surface hint only.
Clause-②: no — the accept set of FlowSchema is unchanged (a top-level trigger / triggerType is refused before and after); only the guidance text a refusal carries changes. If the dev's measurement of the triggertype sibling leads anywhere other than a guidance-text change, that is a stop condition, not a silent expansion.
Serial constraints cleared: file lists of all 29 open PRs scanned at 20:5xZ — none touches automation/flow.zod.ts or flow.test.ts; no commit on origin/main touched flow.zod.ts today; lane siblings in flight #14192 → PR #14714 (kernel/manifest.zod.ts, in queue) is file-disjoint. H17 on-hold trigger-file index (#9857): the only automation entry (#2657) names automation/webhook.zod.ts, not flow.zod.ts. Dedupe reading (one MCP search, 20:5xZ): no open PR or card moves the trigger alias; #14087 (the generator half) is closed; #3427 / #3457 (closed) concern the START-node binding itself, not the diagnostic.

Premise re-read at 809d41704: git grep -n "trigger: 'type'" origin/main -- packages/spec/src/automation/flow.zod.ts → line 594 (and triggertype: 'type' at 595) inside aliases; FlowSchema.type at line 652 is z.enum(['autolaunched', 'record_change', 'schedule', 'screen', 'api']) — no record-* token, so the prescribed rename lands on an Invalid option one round later, exactly as the card measured; the guidance block at 598-605 already carries the START-node config sentence for object / objectName / schedule. Both halves of the premise stand.

---

Generated by Claude Code