docs(skills): an unevaluable hook condition aborts the write (#4775)

[os-zhuang]

Third pass of the rc.2 catch-up, after #4877 (release page) and #4881 (docs sweep run 5): the same window applied to skills/ and .claude/skills/.

Skills matter more than prose here — they are what an AI author reads before emitting metadata, and #4001 closed the whole authorable surface in this window, so a stale skill now produces metadata that hard-fails at parse.

The sweep came back almost clean

Checked all 10 published skills and the 3 internal ones against the window's retirements and behaviour changes. Already correct, updated by the PRs that changed them:

• objectstack-automation documents script 的 config 契约要接入 #4277 的执行期 parse,先得有判别式(actionType)形态 #4343's script-node retirements (actionType: 'email' | 'slack' as logger-backed stubs, inline config.script never executed).
• objectstack-query carries the [P2] data: QueryAST declares 12 members no executor runs — the liveness ledger governs metadata types, not the request surface #4286 prescriptions for cursor, joins and windowFunctions.
• objectstack-formula documents Script validation rules are silently skipped when their predicate fails to evaluate — fail-open is the wrong direction for a validation #4649 fail-closed predicates and has(x) reads as a null guard and is not one — a publish-time lint should reject un-guarded nullable comparisons in CEL predicates #4763's build/publish rejection of has() as a null guard, with the null < null worked example.
• objectstack-data/references/data-hooks.md was updated for bug(objectql): hook 条件对着「只含本次更新字段」的残缺 record 求值，取不到的字段被吞成 false —— 审计 hook 静默不触发 #4770 (the condition reads the record, not the payload) and bug(objectql/docs): hook condition 的 CEL 作用域只绑定 record —— 文档教的 previous.x / ctx.record 根本不存在,过渡型条件写不出来 #4784 (previous is bound), including the transition idiom.

The has(vars.picked) in objectstack-automation is over a flow variable, not a declared record field — that is the "was this written at all" use #4763 explicitly leaves untouched. Not a finding.

One finding

skills/objectstack-data/references/data-hooks.md had been updated for #4770 and #4784 but never for #4775, and closed the condition section with:

An undeclared key (a typo) stays unevaluable: the condition is logged at WARN and treated as false.

That is exactly the behaviour #4775 replaced. An unevaluable condition now aborts the operation.

Wrong in the dangerous direction for an AI-authoring reference: it tells an author a typo is a soft failure. It also silently downgraded the two bullets immediately above it — previous is unbound on inserts and on multi: true writes, and has() is not a null guard — from "this breaks your write" to "this quietly disables your hook". Both are now write-failing mistakes, which is what makes those bullets load-bearing rather than stylistic.

Replaced with a callout carrying the #4775 rule, the reason the two outcomes had to split (a before* guard swallowed into false lets writes through; an audit hook swallowed into false drops records — opposite failures from one collapsed outcome), and the practical authoring consequence.

Verification

• node scripts/check-doc-authoring.mjs — 215 files clean

Docs-only; releases nothing.

---

Generated by Claude Code

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

1 Skipped Deployment

Project	Deployment	Actions	Updated (UTC)
objectstack	Ignored		Aug 3, 2026 1:08pm