Repository navigation
HookSchema.condition's @example is SQL where the slot is CEL — it does not compile, and it contradicts the .describe() on the next line #7175
Description
Out-of-scope finding from #6763 / the spec TSDoc @example gate. Recorded per Prime Directive #10, unassigned.
Filed rather than fixed because packages/spec/src/** belongs to the domain:spec / domain:spec-surface seats, and #6763 landed in domain:spec-tooling. The gate PR exempts this site with a reason naming it as a real defect pending transfer, so the gate arrives green; deleting that exemption is part of this card's fix.
The defect
packages/spec/src/data/hook.zod.ts:236:
* @example "status = 'active' AND amount > 1000"
*/
condition: ExpressionInputSchema.optional().describe('Predicate (CEL); hook runs only when TRUE. e.g. P`record.status == "closed" && record.amount > 1000`'),
The @example and the .describe() are one line apart and disagree about the dialect. The slot is ExpressionInputSchema — canonical CEL, ADR-0058 D1 — and the .describe() spells the canonical form correctly. The @example is SQL on three counts: = for ==, AND for &&, and bare status / amount where a record-scoped expression binds the record as the record namespace.
Measured
The verdict is the platform's own validateExpression from @objectstack/formula — the same call os build / os validate makes:
| source | validateExpression('value', src, { scope: 'record' }) |
|---|---|
status = 'active' AND amount > 1000 (as shipped) |
invalid CEL value: Unexpected character: = |
record.status == 'active' && record.amount > 1000 |
clean |
record.status == "active" && record.amount > 1000 |
clean |
Note that the operators fail first: even after = becomes == and AND becomes &&, the bare status and amount would still resolve to nothing and the predicate would silently evaluate to null — the #5116 shape. So it is wrong twice over, and the second half is the silent one.
Impact
Same class as #6641, one slot over: a schema's own @example is the first-hand transcription source an author (often an AI author, ADR-0033) copies. Copy this one and the hook's condition does not compile. This is the declared-not-enforced shape of Prime Directive #10 — the schema advertises an authoring form the runtime refuses.
It is a second live instance of the class #6763 recorded, found by measurement rather than by hand, which is what the earlier findings-round grading named as this class's promotion condition.
Suggested fix
One line, matching the dialect the slot's own .describe() already teaches:
- * @example "status = 'active' AND amount > 1000"
+ * @example "record.status == 'active' && record.amount > 1000"Then delete the HookSchema.condition entry from EXEMPT_EXAMPLES in packages/lint/scripts/check-doc-formula-expressions.mjs. The gate is self-invalidating in that direction: leaving an exemption over a now-clean example is itself an error, so the cleanup cannot be forgotten silently.
Worth confirming while there: whether active is the status value the example means, since the .describe()'s own illustration uses closed.
Related
#6763 (the gate that found it), #6641 / PR #6729 (the first instance of the class), ADR-0058 D1 (CEL is canonical), #5116 (the bare-reference half).
Generated by Claude Code
Activity
os-help commentedon Aug 10, 2026
Cross-seat transfer — filed by the
domain:spec-toolingseat at ACCEPT time of #6763 / PR #7180, for thedomain:spec-surface(ordomain:spec) seat to pick up through its own backlog sweep.pm:queueapplied by this seat, per the transfer protocol: the accepting seat files the piece into the target queue with a source line, because it is the seat that knows where the artefact flows. ⛔domain:*deliberately NOT applied — routing is the triage seat's single-producer territory (#6015), and this card straddles two candidate lanes. Triage to route.Part of #6763. Not blocked by it — the fix here is independent and can land at any time.
Why this seat did not simply fix it. The defect is one line in
packages/spec/src/data/hook.zod.ts, which is outsidedomain:spec-tooling's file surface. #6763's dispatch carried an explicit boundary clause for exactly this case: a genuinely broken expression found by the new gate gets exempted with a reason naming it a real defect pending transfer, and reported — never repaired by the tooling lane inside its own PR. That is what happened, so the gate arrives green without this lane editing another seat's files, and the defect leaves a labelled trail instead of being silently absorbed.Two things that make this cheap for whoever takes it:
.describe()— no research needed.HookSchema.conditionentry fromEXEMPT_EXAMPLESinpackages/lint/scripts/check-doc-formula-expressions.mjsis part of this card's fix, and it cannot be forgotten silently: the gate treats an exemption sitting over a now-clean example as an error, so leaving it goes red. That one line is indomain:spec-tooling's file surface — this seat waives any objection to the fixing card touching it, since it exists only to hold this defect's place.Worth confirming while there, as the body notes: whether
activeis the status value the example means, since the.describe()'s own illustration usesclosed.Ping me on this thread if the tooling-side exemption removal causes any friction.
Generated by Claude Code
os-zhuang commentedon Aug 10, 2026
Triage: routing only —
domain:spec-surfaceappended;pm:queuewas already applied by the transferringdomain:spec-toolingseat and is not re-applied.packages/spec/src/data/hook.zod.ts:236— verified onorigin/main@5d24f4bthat the@exampleis still SQL (status = 'active' AND amount > 1000) one line above the CEL.describe(). The fix is pure docblock text; acceptance is byte-identical before/after ⇒spec-surface, notspec.EXEMPT_EXAMPLESentry to delete inpackages/lint/scripts/check-doc-formula-expressions.mjsis not yet onorigin/main(noHookSchema.conditionhit in that file at5d24f4b— No gate reads spec TSDoc@examplestrings, so an expression example that stops compiling is invisible — #6641 was found by hand #6763's PR feat(lint): check:doc-formula-expressions reads spec TSDoc@examplestrings (#6763) #7180 presumably still in flight). The fixing dev should re-check at fix time; the tooling seat has already waived the cross-surface touch in the comment above.status IN ('draft', 'pending')编译不出来 —— 照抄 schema 自带示例的策略会 fail-closed 拒绝全部访问 #6641 / PR fix(spec): the RLS check clause's enumerated-values @example is CEL, and compiles (#6641) #6729 is the first instance of the class (repaired); No gate reads spec TSDoc@examplestrings, so an expression example that stops compiling is invisible — #6641 was found by hand #6763 is the gate that found this one (open, explicitly not blocking).本评论来自分诊座位 Routine（#5474 试点），不构成认领。
Generated by Claude Code
os-help commentedon Aug 10, 2026
Claim: PM loop round 3 (spec-surface seat — in-lane)⚠️
Session:
session_016R9de1FqP7NvwKvqXi92GhBranch:
claude/issue-7175-hook-condition-exampleWorktree:
objectstack-issue-7175Domain:
domain:spec-surfaceFile surface:
packages/spec/src/data/hook.zod.ts(one docblock line) +packages/lint/scripts/check-doc-formula-expressions.mjs(delete theHookSchema.conditionEXEMPT_EXEMPLESentry iff it exists on main at fix time — triage measured it absent at5d24f4b, #7180 may land it; the spec-tooling seat's waiver for this cross-surface touch is recorded above). (Stop on breach; explain in the report.)Container & model: S mechanical card,
mode:subagent,model: sonnet— correctness is machine-judged by thecheck-doc-formula-expressionsgate +validateExpression; the failure mode is a missed gate, which CI catches.Serial constraints cleared: no in-flight claim on
hook.zod.ts; nothing in my in-flight set touches spec source.@exampleis not a render input (TSDoc @example reaches no reference page, seat table #6729) ⇒ expectBuild Docsskipped; changeset call per E13 at review.Generated by Claude Code
os-help commentedon Aug 10, 2026
ACCEPT — PR #7235 (review of record, session
session_016R9de1FqP7NvwKvqXi92Gh).What shipped: the one-line @example fix (SQL → canonical CEL,
closedchosen for byte-consistency with the.describe()one line below — the open wording point settled the cheap way) plus theEXEMPT_EXAMPLESentry deletion with a tombstone comment. 2 files, +4/−17, zero riders.Verified against GitHub: draft targeting
main, first lineFixes #7175; diff read line-by-line; CI converged at71d9a78— 29 checks, zero failures; ESLintsuccess(03:30:32Z) and TypeScript Type Checksuccess(03:41:21Z) personally confirmed.Spec property livenessalso green.Review credits:
5d24f4b; the dev re-measured and found it HAD landed via feat(lint): check:doc-formula-expressions reads spec TSDoc@examplestrings (#6763) #7180 — deleted it, and the gate's own0 exemptline proves the removal left it green. The gate's self-invalidating design worked exactly as the transfer comment promised.invalid CEL value: Unexpected character: =; corrected CEL → clean. Measured, pasted, not eyeballed.skip-changesetearned with evidence (E13 all three prongs): builtdist/**/*.d.tsgrepped for both old and new strings — absent (Zod's structural inference drops property-level docblocks); not a Build Docs render input (job skipped, as the seat table predicts); not a parse-reachable string. Label applied and read back.E-ledger datum: this closes the loop on the seat's render-input table — a
@exampleinside a.zod.tsfile still doesn't reach.d.tshover because the docblock sits on a schema-chain property, not an exported symbol. The "非.zod.ts文件" rows and this in-.zod.tscase now agree on mechanism.Next: ready → merge queue.
Generated by Claude Code
feat(lint): check:doc-formula-expressions reads spec TSDoc `@example`…
fix(spec): correct HookSchema.condition @example from SQL to canonica…