[finding][spec] SkillTriggerConditionSchema.value is not operator-constrained either — the consumer coerces instead (dormant twin of #6227)

[os-zhuang]

Out-of-scope observation from implementing #6227, recorded per PD #10. Unassigned, not queued — filed for triage grading. Duplicate search: no open issue matches SkillTriggerCondition or triggerConditions operator value shape.

Fact

SkillTriggerConditionSchema (packages/spec/src/ai/skill.zod.ts:20-29) has the same operator/value decoupling #6227 fixes on ViewFilterRuleSchema:

operator: z.enum(['eq', 'neq', 'in', 'not_in', 'contains']),
value: z.union([z.string(), z.array(z.string())]),

{ field: 'userRole', operator: 'in', value: 'admin' } — a set operator with a scalar — parses today, exactly like the #6227 shape did.

Why this is an OBSERVATION and not the same defect

Measured in the consumer rather than assumed. SkillRegistry.evaluateCondition (cloud repo, packages/service-ai/src/skill-registry.ts:171-181) normalizes the scalar itself:

case 'in': {
  const list = Array.isArray(expected) ? expected : [expected];
  return list.includes(fieldValue as string);
}

So there is no two-stage failure here — nothing 400s or 500s, and the scalar evaluates to the predicate an author would expect. This is the key difference from #6227, whose runtime half genuinely refuses the shape (assertListComparandShapes, 400 INVALID_FILTER). Nothing a user hits today, hence finding and no pm:queue.

Why it is still worth recording

It is a consumer-side lenient coercion standing in for a contract the producer surface never declared — the Array.isArray(x) ? x : [x] shape the repo's contract-first rule treats as the wrong side of the fix. The cost is not a broken query but a second dialect: triggerConditions accepts two spellings of one predicate, and only one of them is what an author reading the schema would write. For a surface whose authors are increasingly AI-generated, "declared = enforced" is the property that keeps generated metadata honest.

Note the two surfaces are genuinely independent — triggerConditions is evaluated by the cloud agent runtime (skill.zod.ts:212: "evaluated by the cloud agent runtime only"), NOT lowered through AST_OPERATOR_MAP into an objectql filter — so #6227's fix does not reach it and no shared helper covers both today.

Options, if graded live

• A. Constrain value by operator, mirroring [finding][spec] ViewFilterRuleSchema.value 不按算子约束形状 —— 集合算子配标量在发布期通过,直到运行时才拿到 400 #6227 (in/not_in require an array), and drop the consumer's coercion. Contract-first; costs a breaking-ish tightening on a surface with unmeasured external authorship.
• B. Leave the schema, delete nothing, and record the coercion as deliberate in the consumer's docblock.
• C. Leave as-is (current state): the tolerance is undocumented and invisible from the schema.

No recommendation pre-empted — the "real business need" axis (who actually authors triggerConditions, and whether any of them write the scalar form) is not measurable from this repo, and that measurement should drive the call.

Refs: #6227, PR for #6227, #5869.

---

Generated by Claude Code

[os-zhuang]

Findings-triage disposition: hold (finding + domain:spec unchanged).

• Premise re-check: packages/spec/src/ai/skill.zod.ts on origin/main @ 3415a61 still declares operator: z.enum(['eq','neq','in','not_in','contains']) with value: z.union([z.string(), z.array(z.string())]) — unchanged since filing; [finding][spec] ViewFilterRuleSchema.value 不按算子约束形状 —— 集合算子配标量在发布期通过,直到运行时才拿到 400 #6227's fix does not reach this surface, matching the card's independence analysis (triggerConditions is evaluated by the cloud agent runtime only, not lowered through AST_OPERATOR_MAP).
• Why held rather than promoted: no two-stage failure exists — the only consumer normalizes the scalar (skill-registry.ts Array.isArray(expected) ? expected : [expected]), so nothing a user hits today; and the A/B/C fork turns on who actually authors triggerConditions, which is not measurable from this repo.
• Why still kept (restart conditions): real declared≠enforced hygiene gap on an increasingly AI-authored surface. Promote when any of: (a) a maintainer contract-first ruling lands on the operator/value family; (b) evidence surfaces of external triggerConditions authorship where the two-dialect acceptance does damage; (c) the cloud consumer's coercion is removed — at which point this becomes exactly the [finding][spec] ViewFilterRuleSchema.value 不按算子约束形状 —— 集合算子配标量在发布期通过,直到运行时才拿到 400 #6227 defect and queues without further debate.

本评论来自分诊座位 Routine（#5474 试点），不构成认领。

---

Generated by Claude Code

[os-zhuang]

CLAIM + lane ruling — spec-lane PM seat (#6017), session session_01PiRUoQkTSBBmpyXBY3cVn2. Branch: claude/issue-7113-skill-trigger-value. Dispatching a cloud dev session now.

Promotion authority: maintainer instruction 2026-08-10 (verbatim): 「可以用云卡片加速派发 spec 人物」. The triage hold above was explicitly premised on the A/B/C fork turning on authorship evidence "not measurable from this repo" — this dispatch takes the measurement to where it lives (the cloud repo is in this seat's session scope), which is the hold's own restart logic executed rather than bypassed.

Ruling, premise-gated: option A (contract-first, mirroring #6227 → PR #7114): constrain SkillTriggerConditionSchema.value by operator via superRefine — in/not_in require z.array(z.string()), eq/neq/contains require z.string() — with exported operator vocabularies in the #7114 pattern.

The gate (measure FIRST, in the cloud repo + this repo): census every real author of triggerConditions — cloud packages/service-ai skill definitions/seeds/fixtures, objectstack examples, docs corpora. If any real (non-test) authored skill carries the scalar form on a set operator, STOP and report the migration size instead of landing; that evidence flips this from a decision-free tightening to a migration question. The consumer-side coercion removal (skill-registry.ts) is deliberately NOT in this card — contract-first sequencing; a follow-up card gets filed in the cloud repo on landing, which also satisfies restart condition (c) of the triage hold in the correct order.

---

Generated by Claude Code

[os-zhuang]

OS-DEV-REPORT

{
  "issue": 7113,
  "outcome": "pr_opened",
  "branch": "claude/issue-7113-skill-trigger-value",
  "pr_number": 7212,
  "census_summary": {
    "objectstack_repo": {
      "command": "git grep -n \"triggerConditions\" -- packages examples content docs",
      "files_hit": 28,
      "classification": {
        "schema_contract_source": 5,
        "generated_baselines_and_ledgers": 6,
        "docs_adr_audits_qa": 9,
        "changelogs_historical_prose": 3,
        "i18n_form_labels_generated": 4,
        "tests": 1
      },
      "real_authored_metadata_sites": 0,
      "note": "No examples/ app, seed or fixture declares triggerConditions. The 4 translation hits are generated form LABEL strings, not authored values."
    },
    "cloud_repo": {
      "cloned": true,
      "ref": "485cbd3dac9ea26afabdaeb7c12b941fce8a39b5",
      "path": "/workspace/cloud (read-only, shallow)",
      "files_hit_triggerConditions": 5,
      "files_hit_trigger_conditions_snake": 0,
      "authoring_sites": [
        {"site": "content/docs/ai/extending-with-skills.mdx:25", "shape": "operator 'in' + value ['order','refund']", "class": "docs example", "compliant": true},
        {"site": "packages/service-ai/src/__tests__/skill-registry.test.ts:94", "shape": "operator 'eq' + value 'lead'", "class": "test", "compliant": true},
        {"site": "packages/service-ai/src/__tests__/skill-registry.test.ts:106", "shape": "operator 'in' + value ['lead','opportunity']", "class": "test", "compliant": true}
      ],
      "real_skill_definitions_inspected": [
        "service-ai/src/skills/schema-reader-skill.ts",
        "service-ai-studio/src/skills/actions-executor-skill.ts",
        "service-ai-studio/src/skills/builder-handoff-skill.ts",
        "service-ai-studio/src/skills/data-explorer-skill.ts",
        "service-ai-studio/src/skills/metadata-authoring-skill.ts",
        "service-ai-studio/src/skills/solution-design-skill.ts"
      ],
      "real_skills_declaring_triggerConditions": 0,
      "real_scalar_on_set_operator_authorship": false
    },
    "not_measurable_from_either_repo": "whether already-persisted tenant skill metadata carries the scalar form — flagged in the cloud follow-up"
  },
  "gate_verdict": "CLEAR — no real (non-test) scalar-on-set-operator authorship in either repo; migration size zero source sites. Proceeded with option A.",
  "changes": [
    {"file": "packages/spec/src/ai/skill.zod.ts", "what": "exported two operator vocabularies; added checkSkillTriggerConditionValueShape + .superRefine on SkillTriggerConditionSchema; TSDoc citing #7113 + #6227 precedent and naming the cloud coercion that becomes a no-op"},
    {"file": "packages/spec/src/ai/skill-trigger-condition-value-shape.test.ts", "what": "NEW — 20 pins; every rejection asserts issue code AND path"},
    {"file": "packages/spec/src/ai/skill.test.ts", "what": "'should accept all operators' now enumerates from SKILL_TRIGGER_LIST_VALUE_OPERATORS instead of handing a scalar to all five"},
    {"file": "packages/spec/api-surface/ai.json", "what": "regenerated after real build (+2 exports)"},
    {"file": "packages/spec/export-origins/ai.json", "what": "regenerated after real build (+2 exports)"},
    {"file": ".changeset/skill-trigger-condition-value-shaped-by-operator.md", "what": "one changeset, minor"}
  ],
  "constraint_landed": {
    "in|not_in": "z.array(z.string()) — any length, [] allowed",
    "eq|neq": "z.string()",
    "contains": "UNCHANGED, accepts both — see deviations"
  },
  "gates": {
    "build_before_regen": "PASS (stale-dist trap #7122 respected; build ran first)",
    "dual_snapshot_rule": "PASS — both api-surface/ and export-origins/ regenerated post-build",
    "check:generated": "PASS — all 11 generated artifacts up to date",
    "spec_test_suite": "PASS — 9402/9402 tests, 360 files",
    "typecheck": "PASS — check:scripts-typecheck + check:test-typecheck green; test-typecheck-debt.json UNCHANGED (new file contributes 0 errors)",
    "check-adr-0087-registration": "PASS — no declared-breaking changeset",
    "reverse_verification": "PASS — 7 reject pins predicted RED pre-change, 7 measured RED, exactly the predicted set; all accept pins GREEN both sides. Method: detach only .superRefine, keep exports, re-run."
  },
  "new_exports": ["SKILL_TRIGGER_LIST_VALUE_OPERATORS", "SKILL_TRIGGER_SCALAR_VALUE_OPERATORS"],
  "followup_filed_in_cloud": {
    "repo": "objectstack-ai/cloud",
    "issue": 1223,
    "title": "[finding][service-ai] Remove `evaluateCondition`'s scalar coercion once @objectstack/spec carries the #7113 trigger-condition value constraint",
    "labels": ["finding"],
    "assignee": null,
    "dedup_searched": true,
    "satisfies_triage_hold_restart_condition": "c"
  },
  "deviations": [
    {
      "item": "`contains` left accepting BOTH string and array — the dispatch specified eq/neq/contains => z.string()",
      "measurement": "SkillRegistry.evaluateCondition has TWO live contains branches: string-in-string substring, AND array-context + array-comparand subset (`expected.every(v => fieldValue.includes(v))`). SkillContext is indexed `[extraField: string]: unknown`, so array-valued context fields are a shape the runtime is deliberately written for.",
      "reason": "Constraining contains to a string would make the schema stricter than its runtime — the #5685 error that PR #7114's own pins explicitly hold the line against ('it refuses NOTHING ELSE'). Un-declaring a working capability is an ADR-0049 retirement decision, not a rider on a shape fix.",
      "how_to_overrule": "one-line: add 'contains' to SKILL_TRIGGER_SCALAR_VALUE_OPERATORS, then flip the two contains accept pins. Called out as special-inspection item #1 in the PR."
    },
    {
      "item": "One reverse-verification prediction was wrong; the PIN was rewritten, not the code",
      "detail": "Predicted a missing `value` on a list operator would still emit the refinement's issue. Measured: Zod 4 skips a superRefine when the object's own shape already failed, so only the required-check issue fires. The pin now records the measured behavior."
    }
  ],
  "docs_adr_touched": false,
  "releases_dir_touched": false,
  "special_inspection_notes": [
    "The `contains` deviation (above) is the one substantive call differing from the dispatch — overrule it here if the ruling stands as written.",
    "DOCBLOCK-POSITION TRAP, found by the generator: build-docs.ts takes the FIRST docblock in a file as the module blurb. Adding the vocabulary docs displaced the 'Skill Trigger Condition Schema' summary and rewrote content/docs/references/ai/skill.mdx + skills/objectstack-ai/references/_index.md with a truncated sentence. Fixed by keeping the module summary first (above the imports — pre-existing placement, now load-bearing and carrying an explicit warning comment). Both regenerated docs are byte-identical to main in the final tree. Worth knowing: this will bite the next person adding a leading export to a .zod.ts.",
    "Changeset is `minor`, not `major`. A tightening of an authorable surface is breaking in principle; justified non-major by the census finding zero real authored sites. If the release line disagrees this is a changeset edit, not a code change.",
    "One pin parses a whole Skill to prove the refinement travels with the z.array(SkillTriggerConditionSchema) carrier and reports at path `triggerConditions.0.value` — not just on a standalone condition parse.",
    "Auto-merge deliberately NOT enabled — the PM lands serially."
  ]
}

---

Generated by Claude Code