skills-sweep ③: objectstack-query (1,550 lines, 6 files) — behavioral-claim verification, content-class execution-first

[huangyiirene]

Member card of program #13658 (published-skills factual sweep). Flight ③ target: skills/objectstack-query/** — 6 files, 1,550 lines, enumerated at origin/main this morning:

• SKILL.md (673) · rules/filters.md (292) · rules/aggregation.md (278) · rules/pagination.md (233) · evals/README.md (39) · references/_index.md (35)

Scope: verify every behavioral claim against the implementation per the program method as BOUND by #13658 comments 5474435934 / 5474464789 / 5475032239 — content-class targeting first (surface/liveness claims: does the key exist, is it retired, is it automatic, what does the engine install), both-error-direction parse probes (deader-than-real is now the majority direction), cross-file contradiction scan inside the package, executed probes for every behavior-bearing claim. The two $exists rows already corrected by merged PR #13577 are prior art, not targets — verify their current state on origin/main rather than re-deriving them.

Constraints: published governed surface — draft PR, human merge, .md content goes to maintainer review (2026-08-26 ruling); token-ratchet ceilings unchanged (raising one is the maintainer's call); net line budget at most 0. NOT MEASURABLE claims are recorded in the flight report, not annotated in the files.

Part of #13658; this card closes with the flight's PR (Fixes).

Generated by Claude Code

[huangyiirene]

Claim: PM loop round 1
Session: session_01EnE7G31tqbxN1rqpQmzurT
Branch: claude/issue-13717-skills-sweep-query
Worktree: objectstack-issue-13717
Domain: domain:skills
File surface: skills/objectstack-query/** (stop on breach; explain in the report)
Container & model: M (claim verification with executed probes — judgment tier), mode:subagent, model: opus — per this hour's node scripts/pm/dispatch-gates.mjs --tier run over this face: "no path-derived mandate: the surface hits none of the 3 declared glob(s) … The tier stays the PM's per-card judgment call"; clause ② not reachable from these paths.
Clause-②: no — published-docs text corrected toward the implementation; no contract accept/reject behavior change, no public-surface widening. (Whether published skills/** text falls under clause ②'s CONTENT limb is open on #13588; the .md ⇒ maintainer-review terminal covers this PR under either answer.)
Serial constraints cleared: none — zero open PRs touch skills/objectstack-query/** (open-PR listing read this hour); prior art PR #13577 on two of these files is MERGED (03:15:39Z); flights ④+ are not dispatched.

---

Generated by Claude Code

[huangyiirene]

Dev claim (os-dev subagent, in addition to the PM claim above)

Session: session_01EnE7G31tqbxN1rqpQmzurT (same session as the dispatching PM seat — this seat is the mode:subagent dev under it; the distinguishing identity bit is the branch below)
Branch: claude/issue-13717-skills-sweep-query (pushed empty at claim time as the write-route probe — push OK)
Worktree: objectstack-issue-13717, base e398863cebeebb4baff867dd6bf7efcfef4e848f
File surface: skills/objectstack-query/** only

Channel note for the record: this seat's session gate is CLOSED for repo-scoped REST (GET /repos/objectstack-ai/objectstack/issues/13717 returns 403; /rate_limit answers core 15000/hr, so credentials are alive and it is the repo-scoping that refuses; gh absent). Bulk reads therefore go through the public-repo payload channel; MCP is reserved for writes plus the one paginated read the payload channel could not reach.

---

Generated by Claude Code

[huangyiirene]

os-dev-report

{
  "issue": 13717,
  "status": "done",
  "branch": "claude/issue-13717-skills-sweep-query",
  "pr": "https://github.com/objectstack-ai/objectstack/pull/13740",
  "premise_still_valid": true,
  "summary": "Flight 3 of the published-skills factual sweep. Swept all 6 files / 1,550 lines of skills/objectstack-query against the implementation: cross-file contradiction scan first (found 4, every one settled against the code), then executed parse-and-run probes for every behavior-bearing surface claim in both error directions. 13 distinct false facts corrected across 17 landing sites, plus 4 derived eval rubrics and 1 false framing label; 11 of the 17 sites are the deader-than-real direction, matching flight 2's majority finding. Biggest: $field cross-field comparison and per-aggregation filter are both ENFORCED (docs said schema-reserved, zero rows / unfiltered number); count_distinct runs on every SQL face (docs said it throws); search expands to $icontains not $contains (the doc's own 'case-insensitive' sentence contradicted the operator it named); query.distinct and aggregations[].distinct are TOMBSTONED, refused at parse, not 'silently ignored'; and all three rows of the CRM Analytics blueprint named removed keys (compareTo bare string, categoryGranularity, groupingsDown/groupingsAcross). Net -33 lines / -255 tokens; no ratchet ceiling raised, every file under its pinned ceiling. Draft PR, governed surface, human merge only. Two PM mechanism assumptions falsified, both harmless: the origin/main face was exactly as briefed (673/292/278/233/39/35 re-derived and confirmed), but false density measured 4.4-5.7%, roughly double the 1.5-3% planning floor; and check:skill-examples does not apply to this package (it type-checks only os:check-marked fences and this package has zero of them, control-verified against three packages that do).",
  "tests": "Union re-run after the final commit; every figure below is from the tree at d601906b (git rev-parse --short HEAD), which is the PR head. GATES: 13 families derived mechanically via `node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack` from the REAL diff, on a NON-stale tree (first derivation warned lint.yml and cross-package-test-inputs.mjs were stale and are themselves families in the list, so origin/main 0f911eb9 was merged in and the derivation re-run — identical 13-family answer). All exit codes captured as `cmd > log 2>&1; EXIT=$?`, before any pipe. EXIT=0: check-ci-filter-parity, check-cross-package-test-inputs, check-shard-attestation, check-skills-token-ratchet, check:doc-formula-expressions (lint pkg), check:agent-test-spelling, check:cross-package-test-inputs, check:doc-authoring, check:pm-governed-merges, check:role-word, check:skill-compatibility, check:skill-frame-sync. EXIT=3: check-test-completeness — its own PREREQUISITE NOT MET branch (it grades a saved turbo test log; the family names it with no argument, and its output instructs the local reader to record NOT MEASURED). check:skill-examples: not in the derived list and correctly so — population empty here (zero os:check markers in this package; control: objectstack-ui / objectstack-i18n / objectstack-formula each carry them). Two bounded foreground attempts to run it anyway were blocked by its @objectstack/client-react build prerequisite, whose closure fails on pre-existing type-only import gaps in @objectstack/runtime and @objectstack/verify (Cannot find module '@objectstack/rest'), unrelated to this diff; both hit the container's ~10-minute foreground cap. CI runs it on every PR. RATCHET, the load-bearing gate: exit 0, every file under ceiling — SKILL.md 5443/5552, evals/README.md 558/567, rules/aggregation.md 2226/2357, rules/filters.md 2144/2149, rules/pagination.md 1381/1382; package total 11752/12007 (-255). Lines 1550 -> 1517 (-33). EXECUTED PROBES (vitest, run through scripts/pm/os-verify-lock.sh, after building the dependency closure — pnpm --filter '@objectstack/spec^...' and '@objectstack/objectql^...' / '@objectstack/driver-sql^...' build, both exit 0): (a) $field on the in-memory evaluator, 3 rows over/under/equal — $gt -> [\"over\"], $eq -> [\"equal\"], $lt -> [\"under\"]; control literal $gt:150 -> [\"over\"]. Through SqlDriver on better-sqlite3, pushed down: [\"over\"]. Claim under test predicted zero rows on both paths. (b) SqlDriver.aggregate count_distinct -> [{\"n\":2}]; control, undeclared function median -> threw INVALID_QUERY/400, so a refusal would have been visible. (c) engine.aggregate with a per-aggregation filter over 4 rows, 2 closed_won -> [{\"total_deals\":4,\"won_deals\":2}]; claim predicted won_deals 4. (d) engine.aggregate having order_count $gt 1 -> [{\"customer_id\":\"c1\",\"order_count\":2}] (also proves groupBy auto-selects the grouped column). (e) expandSearchToFilter('machine learning') -> $and of two $or groups of $icontains; single term 'apollo' -> a bare $or with no $and wrapper. (f) $contains 'apollo' -> [\"apollo\"], $icontains 'apollo' -> [\"Apollo\",\"apollo\"]. (g) dateGranularity quarter with alias -> [{\"quarter\":\"2025-Q1\",\"revenue\":30},{\"quarter\":\"2025-Q2\",\"revenue\":5}]. (h) $exists/$null prior art of PR 13577 re-verified: $exists true -> [\"v\"], $null true -> [\"n\",\"absent\"] — HAS A VALUE, both merged rows correct. (i) bare null comparand: [\"n\",\"absent\"] in memory, [\"n\"] on SQL — identical to $null:true on each path; $null:false -> [\"v\"]. PARSE ORACLES: QuerySchema REJECTS cursor, joins, windowFunctions, distinct, aggregations[].distinct, array_agg, string_agg, alias-less aggregation; ACCEPTS all six aggregation functions, top, having, aggregations[].filter, all fifteen documented operators, and a $field comparand. Position oracle: $field ACCEPTED in comparison position, REJECTED as $in/$nin member and $between endpoint. DashboardWidgetSchema REJECTS compareTo:'previousPeriod' and categoryGranularity, ACCEPTS compareTo:{kind}; ReportSchema REJECTS groupingsDown/groupingsAcross, ACCEPTS dataset+rows+columns+values. isDateMacroToken LIVE for all twelve tokens filters.md names and for all six parameterised units in both directions, DEAD for the two near-misses the doc calls near-misses. NON-VACUITY: satisfied many times over — every probe above ran a discriminating control, and probes (a)-(d) each returned a result the claim under test predicted would not occur. NO PROBE FILE REMAINS: three temporary vitest probe files were created under packages/objectql/src and packages/drivers/driver-sql/src, then deleted before the gate derivation; final git status is 5 modified files, all under skills/objectstack-query, 0 untracked.",
  "mcp_calls": "7 — issue_read(get_comments #13658), add_issue_comment(dev claim on #13717), create_pull_request(#13740), issue_read(get_labels #13740, failed: cannot resolve a PR by issue number), pull_request_read(get, body+label read-back), issue_write(labels union), pull_request_read(get, label read-back). All bulk issue reads went through the zero-quota public-repo payload channel instead; this seat's repo-scoped REST is 403 (session gate closed — /rate_limit answers core 15000/hr so credentials are alive and it is the repo-scoping that refuses; gh absent), which is also why the additive REST label POST fell back to the MCP read-union-write-verify path. This report comment is the 8th.",
  "files_changed": [
    "skills/objectstack-query/SKILL.md (673 -> 658 lines, 5552 -> 5443 tokens)",
    "skills/objectstack-query/rules/aggregation.md (278 -> 264 lines, 2357 -> 2226 tokens)",
    "skills/objectstack-query/rules/filters.md (292 -> 290 lines, 2149 -> 2144 tokens)",
    "skills/objectstack-query/rules/pagination.md (233 -> 232 lines, 1382 -> 1381 tokens)",
    "skills/objectstack-query/evals/README.md (39 -> 38 lines, 567 -> 558 tokens)"
  ],
  "line_budget": "package 1550 -> 1517 lines (net -33, budget was at most 0); ratcheted tokens 12007 -> 11752 (net -255); no ceiling raised, no ceiling touched; references/_index.md (generator-owned, 35 lines) verified and unchanged",
  "claim_counts": {
    "total_claims_inventoried": 296,
    "verified": 269,
    "false_corrected": 18,
    "not_measurable": 9,
    "false_distinct_facts": 13,
    "false_landing_sites": 17,
    "false_derived_rubrics": 4,
    "false_framing_label": 1,
    "false_density_landing_sites": "5.7%",
    "false_density_distinct_facts": "4.4%",
    "deader_than_real_share": "11 of 17 landing sites",
    "per_file_false_landing_sites": {
      "SKILL.md": 10,
      "rules/aggregation.md": 4,
      "rules/pagination.md": 1,
      "rules/filters.md": 2,
      "evals/README.md": 4,
      "references/_index.md": 0
    },
    "cross_file_contradictions_found": 4,
    "cross_file_contradictions_that_were_false": 4
  },
  "deviations": [
    "Narrowing DECLARED, not silent: repo-wide `pnpm lint` was not run locally. The 13 derived families were run instead, in full, at the final commit; CI runs the whole farm on every PR. No local eslint reading is claimed either way.",
    "check:skill-examples was named in the dispatch brief but is NOT in the mechanically derived family list, and the derivation is right: its population is the set of os:check-marked fences and this package has zero (control-verified against three packages that do have them). Two bounded attempts to run it regardless were blocked by an unrelated pre-existing build-closure failure and the ~10-minute foreground cap; recorded as empty-population, not as a skipped gate.",
    "check-test-completeness exit 3 recorded as NOT MEASURED per the script's own PREREQUISITE NOT MET text, never as a red.",
    "origin/main was merged into the branch mid-flight (to 0f911eb9) so the gate derivation would run on a non-stale tree; the first derivation had warned that two files it derives from — lint.yml and cross-package-test-inputs.mjs — had changed underneath. The re-derived family list is byte-identical to the stale one, so the merge changed nothing except the warning; the PR diff vs origin/main remains the same 5 skills files.",
    "The PR body carries the Generated-by footer TWICE — the create path appended its own on top of the one written. Left alone deliberately: a PATCH to remove it would downgrade the session-URL footer to the bare form, which is the durable-attribution loss the discipline warns about. Cosmetic only.",
    "assignee left untouched (PM owns it); the dev claim comment on #13717 carries this session id and the branch."
  ],
  "open_questions": [],
  "out_of_scope_findings": [
    "NOT FILED, handed to PM for filing — this seat's repo-scoped REST is 403 and the one MCP dedup search was reserved for higher-value work, so the finding is reported here rather than opened blind: packages/objectql/src/engine.ts carries a stale docblock (near the ADR-0061 search-expansion helper shared by find and findOne) describing the expansion as 'a server-resolved cross-field $or of $contains'. It emits $icontains — search-filter.ts one file over documents that correction explicitly and pins it. Exactly the sentence this PR corrects in the published skill, still live as an internal comment. Suggested: `finding` label, objectstack repo, no assignee. Severity low (comment only, no behavior), but it is the same phantom the sweep exists to remove and it sits in the file an agent reads first."
  ]
}

---

Generated by Claude Code