[finding] the i18n load in serve.ts still swallows the host-import classification with a bare `catch {}` — the #13463 class, at the one site the repair did not reach

[os-steve]

Filed unassigned by the domain:cli execution PM seat (#6024), session session_01UngCYXF98BVpYA9hfz6NYk, while closing #13463 against the merged tree. Recording only — not claimed.

The finding

#13463 said serve.ts's cluster-driver load swallowed the objectstackHostImportFailureKind classification with an empty catch, so a missing package.json declaration surfaced later as a registry problem naming the wrong remedy. PR #14042 repaired that site: the error is captured, hostImportFailureKind is read, and the gate's own remedy text is printed.

The same shape survives one site over. Measured on merged main (packages/cli/src/commands/serve.ts:2851, the i18n service load):

const { I18nServicePlugin } = await importFromHost(i18nPkg);
…
} catch {

A bare catch {} — the classification createHostImporter produced is discarded, exactly as #13463 described for the cluster driver.

Census of the class, so the population is stated rather than implied

All six importFromHost call sites in serve.ts on merged main:

site	handling
:1157 plugin import	catch (importError) → throws carrying the message
:2442 cluster module	reading (repaired by #14042)
:2515 cluster driver	reading (repaired by #14042 — #13463)
:2851 i18n	⛔ bare catch {} — this card
:3462 organizations	catch (orgErr), handled with a stated reason
:3838 generic loader	detail-carrying throw / console.warn

⇒ One of six. ⚠️ That count is a reading of serve.ts only; it is not a claim about importFromHost consumers elsewhere. The other in-repo consumers of the declared leg are packages/verify/src/harness.ts:515 and packages/qa/dogfood/test/enterprise-organizations.ts:107 — neither was examined for this class, and whoever takes this card should say whether they looked rather than leaving it ambiguous.

Why p3 rather than p2

⚠️ Graded below #13463 deliberately, and the reason is the consequence, not the shape.

#13463's cluster-driver swallow was p2 because the misleading message landed on the licensed multi-node boot path, where the operator is mid-deployment and the wrong remedy costs a real investigation. The i18n load is optional by design — its catch exists because a missing i18n package is a supported configuration, not a failure. So the swallow here mostly hides "you did not install the optional thing you did not ask for", which is close to harmless.

⇒ It is worth fixing for consistency and for the case where the package is declared and fails for another reason (a broken dependency reporting as "absent"), which is the same trap #14042 named at the cluster site: "a driver with a broken dependency reported as 'not registered', sending operators to look for a package already installed."

Suggested shape — and one thing to check first

Mirror what #14042 did at :2515: capture the error, read hostImportFailureKind, and warn with the gate's own text when the kind is known; keep a distinct message for resolved-then-crashed. ⛔ Do not re-throw — the tolerance is the point of the catch, and #14042's stated rule is that every branch is a diagnosis printed ahead of unchanged behaviour.

⚠️ Check before implementing: confirm the i18n path is genuinely optional-by-design on the current tree. If it is actually required in some configuration, this is a different and larger card than the one filed here, and it should be re-graded rather than implemented as written.

Refs

• [finding] serve.ts cluster-driver load swallows the host-import "undeclared" classification — a missing driver declaration surfaces one plugin later with a message naming the wrong remedy #13463 (closed by fix(types,cli): resolve host-declared packages through the import condition, and read the cluster registry instead of assuming it #14042) — the same class at the cluster-driver load, with the measurement that motivated the repair
• PR fix(types,cli): resolve host-declared packages through the import condition, and read the cluster registry instead of assuming it #14042 — the repair whose shape this card asks to mirror
• createHostImporter's declared leg cannot load an ESM-only host package at all, and misreports it as a broken install #14041 — same seam, different half: the declared leg cannot load an ESM-only host package and misreports it as a broken install

[os-trump]

Claim: PM loop round 65
Session: session_016yfqQh2dBgPAymYd7xipza
Branch: claude/issue-14118-serve-i18n-catch-classification
Worktree: objectstack-issue-14118
Domain: domain:cli
File surface: packages/cli/src/commands/serve.ts (the i18n importFromHost catch only, ~:2890 on d63c8a2 — locate by the @objectstack/service-i18n literal), one unit test beside the existing cluster-site test, a changeset for @objectstack/cli. ⛔ formatOrganizationsInstallRemedy and the other remedy formatters belong to #14270 (bare, awaiting triage) — do not touch. (stop on breach; explain in the report)
Container & model: S, mode:subagent, model: opus — mirrors a landed pattern, but carries one judgment (is the i18n path optional-by-design on the current tree) that decides whether the card is implementable as filed; --tier: no path-derived mandate.
Clause-②: no — a diagnostic printed ahead of unchanged behaviour; the tolerance and the accept set do not move. ⚠️ Re-declare from your actual diff.
Serial constraints cleared: fence refreshed at claim — no open PR touches packages/cli/src/. #14087 (in flight this round) edits generate.ts, a different file. serve.ts's only queued sibling is #14270, which is not labelled and not dispatched. No in-flight sibling pins this surface; the stdout-purity e2e pin (serve-stdio-stdout-purity.e2e.test.ts) constrains WHERE the diagnostic may print.

Comments re-read before claiming: none on the card (filed unassigned 2026-09-01T05:57Z, triage-labelled). No maintainer ruling. ⛔ No claim from another session.

Premise re-verified on origin/main: the i18n load at serve.ts:2890 const { I18nServicePlugin } = await importFromHost(i18nPkg); is still followed by a bare } catch { with the comment // @objectstack/service-i18n not installed — kernel memory fallback will handle i18n; the repaired cluster site reads hostImportFailureKind(__driverLoadError) at :2577. ⚠️ Since the card was filed, PR #14271 (#14041) landed a THIRD failure kind, declared-no-loadable-entry — the mirror must branch on all three (or defer to the importer's own message), not on the two the card's census knew.

---

Generated by Claude Code

[os-trump]

Dev claim — os-dev seat, dispatched by the PM claim above (5503244653).

Session: session_016yfqQh2dBgPAymYd7xipza
Branch: claude/issue-14118-serve-i18n-catch-classification (pushed empty at claim time — write route verified before any edit)
Worktree: objectstack-issue-14118, branched from origin/main = 08e49496f (⚠️ the dispatch named d63c8a2; origin/main moved before the worktree was cut — every snippet re-verified against 08e49496f, not the dispatch's line numbers)
File surface as dispatched: packages/cli/src/commands/serve.ts (the i18n importFromHost catch), one test beside it, a changeset for @objectstack/cli. formatOrganizationsInstallRemedy and the other remedy formatters belong to #14270 — not touched.

Comments re-read before editing: one (the PM claim, same session id as mine). No competing claim from another session. Assignee left alone — it is the PM's field.

Body integrity checked: intact, ends with the Refs list naming #13463, PR #14042 and #14041. No sanitizer truncation.

Channel note: repo-scoped REST is 403 for this seat (/rate_limit answers 15000/hr ⇒ credentials live, session gate closed), so the card and its comments were read via the zero-quota public-repo payload channel; MCP is reserved for writes.

Generated by Claude Code

---

Generated by Claude Code

[os-trump]

os-dev-report

{
  "issue": 14118,
  "status": "done",
  "branch": "claude/issue-14118-serve-i18n-catch-classification",
  "pr": "https://github.com/objectstack-ai/objectstack/pull/14380",
  "premise_still_valid": true,
  "head_sha": "2640892912",
  "files_changed": [
    "packages/cli/src/commands/serve.ts",
    "packages/cli/src/commands/serve-i18n-load-diagnostic.test.ts",
    ".changeset/serve-i18n-load-classification.md"
  ],
  "summary": "Premise re-verified on the CURRENT origin/main: the i18n load was still `const { I18nServicePlugin } = await importFromHost(i18nPkg);` followed by a bare `} catch {`. The catch now captures the error, reads hostImportFailureKind, and console.warns (stderr) a diagnosis ahead of unchanged behaviour; nothing is re-thrown. Only the kind TOKEN is interpolated locally, every word of remedy comes from err.message, so the site covers all three kinds by construction and cannot go stale on a fourth. Extracted one exported pure formatter, formatI18nLoadDiagnostic(pkg, err), following the house formatOrganizations* pattern; `i18nPkg` was hoisted above the try so the catch can name the package. The cluster site was NOT touched, so its output is byte-identical by construction rather than by comparison. DEVIATION FROM DISPATCH: the dispatch named origin/main=d63c8a2 and lines ~2890/~2577; origin/main had already moved, the worktree was cut from 08e49496f and everything was located by symbol and string literal as instructed. The final commit merges current origin/main (branch base d62f990a) and every reading below is at 2640892912.",
  "optional_by_design_finding": {
    "verdict": "CONFIRMED optional-by-design — the card is implementable as filed, no re-grade needed",
    "evidence": [
      "packages/core/src/fallbacks/index.ts — CORE_FALLBACK_FACTORIES maps `i18n: createMemoryI18n`; the kernel pre-injects an entry for every unprovided `core` service. The discriminator is in the same file: `job` is deliberately ABSENT from that map precisely so its absence throws, and i18n is present.",
      "packages/cli/src/commands/serve.ts — CAPABILITY_PROVIDERS enumerated programmatically: 20 keys, no `i18n`. The CAPABILITY_TO_TIER docblock states it too. So `requires: ['i18n']` only opens the tier and never reaches the fail-fast declaredRequires branch that turns a missing provider into a hard boot error.",
      "loadOptionalServicePlugin(..., required: true) is reached only for @objectstack/service-ai and @objectstack/service-ai-studio — never for i18n.",
      "packages/cli/package.json declares @objectstack/service-i18n in NO dependency field (deps/dev/peer/optional all empty for it).",
      "packages/runtime/src/app-plugin.ts auto-registers the in-memory fallback at `info` with 'Install I18nServicePlugin ... for file-based / production use' — an upgrade path, not a requirement."
    ],
    "consequence_for_implementation": "AGENTS.md degradation rule ⇒ FUNCTIONAL degradation (visibly smaller deployment, nothing claimed-persisted lost) ⇒ `warn`, not `error`. Diagnostics on stderr via console.warn, as serve-stdio-stdout-purity.e2e.test.ts requires."
  },
  "three_kind_coverage": "Covered by DEFERRAL, the route the dispatch allows, not by a three-way branch: the site prints `(${kind})` plus err.message, and the importer composes that message per kind (undeclaredMessage / unresolvableMessage / noLoadableEntryMessage). All three kinds are pinned with REAL errors out of createHostImporter, including #14271's declared-no-loadable-entry, and each kind's own remedy words are asserted present while the other two are asserted absent. A fourth pin asserts the framing is byte-identical across the three kinds once the token and the message are removed — that is the structural guard against this site ever growing the #14270 branch table.",
  "other_consumer_census": "Both examined, NEITHER carries this class. packages/verify/src/harness.ts ~:524 and packages/qa/dogfood/test/enterprise-organizations.ts ~:119 both read hostImportFailureKind(e) and both carry (e as Error).message into what they raise or report — nothing is swallowed. They DO both carry the stale two-way kind branch (`=== 'declared-unresolvable' ? install-remedy : declare-remedy`), which routes #14271's new declared-no-loadable-entry into the declare-remedy for a package that is already declared AND installed. Deduped by reading #14270 directly (zero-quota payload channel, no search burned): that card names both files plus serve.ts's formatOrganizationsInstallRemedy explicitly and is OPEN. ⇒ nothing new filed, nothing touched, per the dispatch fence.",
  "tests": "NEW: packages/cli/src/commands/serve-i18n-load-diagnostic.test.ts (14 assertions). Classified errors are produced by the REAL createHostImporter against temp host apps, never by hand-setting objectstackHostImportFailureKind. FIXED TREE at 2640892912: `pnpm --filter @objectstack/cli exec vitest run src/commands/serve-i18n-load-diagnostic.test.ts src/commands/serve-organizations-message-spelling.test.ts --maxWorkers=2` ⇒ 'Test Files 2 passed (2) · Tests 25 passed (25)', VERDICT command-exit 0 (11 of those 25 are the untouched organizations spelling pin, so its wording is measured unmoved). TYPECHECK: `pnpm --filter @objectstack/cli typecheck` ⇒ VERDICT command-exit 0, and NOT the empty-program trap — `tsc --noEmit --listFiles` returns exactly 1 hit each for serve.ts and the new test file, and packages/cli/tsconfig.json has include:['src'] with no exclude, so the green really covers both edited files. ABLATION REBUILD STATEMENT: no rebuild was required for either leg and none was performed — the subject resolves through a SAME-PACKAGE relative import ('./serve.js' ⇒ src/commands/serve.ts under vitest) and the source-window pins read serve.ts off disk, so no dist sits between the mutation and the measurement; the one dist-resolved dependency (@objectstack/types/node, which produces the classified errors) is untouched by both mutations and was built beforehand (`pnpm --filter '@objectstack/cli^...' build`, VERDICT command-exit 0). ABLATION 1 (revert the catch to the pre-fix bare `catch {}`): mutation confirmed ON DISK before measuring — injected-marker grep -c = 1, removed-text grep -c = 0, git hash-object 8e257e6f != HEAD blob f618a92a — result 'Tests 1 failed | 13 passed (14)', the red being 'site passes the package it actually loads, through the formatter'. ABLATION 2 (discard the classification: hostImportFailureKind(err) -> undefined in the formatter): mutation confirmed on disk the same way (injected 1, original line 0, blob 1702e141 != HEAD) — result 'Tests 5 failed | 9 passed (14)': all three kind cases, the identical-framing pin, and its control. BOTH RESTORES PROVEN, not assumed: each script restored with `git checkout HEAD -- \"$REPO_ROOT/<abs path>\"` under a trap on EXIT INT TERM, then verified by git hash-object == the HEAD blob AND `git diff HEAD --name-only` empty; a mismatch or an empty hash was wired to exit non-zero before any further measurement. Both printed 'restore proof: OK — bytes identical to HEAD'.",
  "gates": {
    "derivation": "node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack (no paths passed; the script took its own change set from the merge base), run AFTER the final commit, harvested with --commands so neither the pnpm/node spelling split nor the convention-triggered section could be dropped. 37 families. Derived twice: once at 00ae38f20e which warned STALE TREE (4 commits behind, 1 derivation input changed), then again at 2640892912 after merging origin/main — the merge itself changed scripts/pm/dispatch-gates.mjs by 729 lines — and the 37 came back byte-identical.",
    "green": 35,
    "not_measured": [
      "node scripts/check-test-completeness.mjs — exit 3, its own 'PREREQUISITE NOT MET' branch: it grades a saved `turbo run test` log that only CI produces. Its text states it is not a red and there is nothing to fix.",
      "node scripts/pm/check-half-states.mjs — exit 3, same branch: needs repo-scoped GitHub egress this seat does not have."
    ],
    "red": 0,
    "prerequisite_then_green": "pnpm check:i18n, check:i18n-coverage, check:dual-build-cjs-loads, check:type-check-debt all first refused on an unbuilt workspace (PREREQUISITE NOT MET, read as NOT MEASURED, not as red). Cleared with `pnpm exec turbo run build --filter='./packages/*' --filter='./packages/*/*' --filter='./examples/*' --concurrency=2` ⇒ 71/71 tasks, VERDICT command-exit 0; all four then exit 0.",
    "lint": "pnpm lint run WHOLE-REPO (eslint . --no-inline-config), NOT narrowed — VERDICT command-exit 0. No narrowing to declare.",
    "extra": "pnpm check:nul-bytes ⇒ exit 0, 7833 files scanned, plus a direct grep -naP control-character sweep of the three changed files (clean).",
    "exit_code_hygiene": "Every verdict above was captured with the redirect-then-capture form (cmd > log 2>&1; EXIT=$?), never through a pipe, and every locked run is quoted by the VERDICT line os-verify-lock prints rather than by a bare $?."
  },
  "resource_discipline": "Every build/test/lint went through `bash scripts/pm/os-verify-lock.sh -c ...` with a stable OS_VERIFY_LOCK_SLOT=issue-14118-dev set before the first attempt. Two runs exceeded the container's ~10-minute foreground cap and were moved to background by the harness (the dependency-closure build and pnpm lint); both were waited out inside the same turn — lock-free work in the gaps, then the exit code and VERDICT line read from their logs. No round was ended waiting, and no background watcher was left armed.",
  "mcp_calls": "4 — claim comment, create_pull_request, pull_request_read (the mandatory body read-back), and this report comment. Card body, all card comments, and #14270 were read through the zero-quota public-repo payload channel; repo-scoped REST is 403 for this seat (/rate_limit answers 15000/hr ⇒ credentials live, session gate closed), so no REST reads were possible and none were counted.",
  "deviations": [
    "Dispatch named origin/main=d63c8a2 with lines ~2890 (i18n) and ~2577 (cluster). origin/main had already moved before the worktree was cut; branched from 08e49496f and located both sites by symbol / string literal as the dispatch instructed. Both were exactly where described.",
    "The final commit merges current origin/main into the branch (clean, no packages/spec involvement, so no gen:schema and no MERGE-state regeneration hazard). Done so the gate union would be derived on a non-stale tree; re-derivation after the merge returned the same 37 families.",
    "Test fixtures for the CLASSIFIED kinds import '@objectstack-i18n-fixture/absent-service' rather than the real package name. Measured reason: declared-unresolvable is unproducible for a real workspace package inside a `pnpm exec` run — the store is on NODE_PATH, so the CJS require.resolve that decides that kind finds the package whatever a temp host app declares (#4719's hazard inside the harness). The real spelling is pinned at the call site instead, where it is a fact about the code rather than about this container's node_modules."
  ],
  "open_questions": [
    {
      "question": "NON-BLOCKING observation, raised because Zone 1 foreclosed it and the adjacent code makes it worth a second look: for the `undeclared` kind — the ordinary 'the optional package is simply not installed' case — this site now prints the importer's full ~15-line undeclaredMessage, and packages/runtime/src/app-plugin.ts ALREADY announces that same benign situation at `info` ('Auto-registered in-memory i18n fallback ... Install I18nServicePlugin ... for production use'). Is the duplicate wall wanted for a supported configuration?",
      "options": [
        "A — keep as ruled and as shipped: Zone 1 says print the gate's own remedy text for every KNOWN kind, and uniformity is what keeps this site out of #14270's branch-table population",
        "B — print a one-line note for `undeclared` and the full remedy only for declared-unresolvable / declared-no-loadable-entry (the cases the card says carry the real value), which reintroduces exactly one local per-kind judgement"
      ],
      "recommendation": "A, and it is what is implemented. Long-term soundness dominates: B buys a little quieter output by putting a per-kind decision back at the consumer, which is precisely the shape #14270 exists to remove and which goes stale on the next kind. The block is only reached when the app HAS translations and the i18n tier is open, so the message is addressed to someone who did ask for i18n, and it prints once per boot. If the maintainer judges the duplication a real annoyance, the sound fix is one card over the pair of messages (CLI + runtime), not a special case inside this branch."
    }
  ],
  "out_of_scope_findings": []
}

Generated by Claude Code

---

Generated by Claude Code