Three failure-kind remedy formatters word the new `declared-no-loadable-entry` kind as "declare it in the app's package.json" — wrong for a declared, installed package

[os-justin]

Blocked-by: #14041 (the kind lands there; until it merges this is latent by definition)

Shape

#14041 adds a third HostImportFailureKind, declared-no-loadable-entry: the app declares the package, the install delivered it, and the package's own exports names no runtime entry Node can load (a types-only or browser-only publish, or an unexported subpath). The importer's own error message words that case correctly — the remedy lives in the package.

Three consumers pick their remedy line with a two-way branch written when only two kinds existed:

• packages/cli/src/commands/serve.ts — formatOrganizationsInstallRemedy: kind === 'declared-unresolvable' ? <repair the INSTALL> : <add the package to THIS APP — declare it in the app's package.json and install>
• packages/verify/src/harness.ts (~:524) — same two-way branch in bootStack's multi-tenant fatal
• packages/qa/dogfood/test/enterprise-organizations.ts (~:119) — same two-way branch in the enterprise probe's refusal

For the new kind, all three fall into the else leg and print the declare remedy — telling an operator whose app already declares and installs the package to declare and install it. That is the exact confidently-wrong-verdict class #14041 exists to remove, one layer up. Mitigation today: each site also prints the importer's own message (e.message / cause), which carries the right shape wording — the remedy bullet contradicts it rather than replacing it.

Why not fixed in #14041

Out of that card's declared file surface, and not mechanical there: serve.ts's remedy text is pinned by serve-organizations-message-spelling.test.ts, so the fix carries its own message-spelling test updates across three packages — its own small card, not a rider.

Suggested shape

Branch on "is the declaration the problem?" rather than on one kind: undeclared → declare-remedy; declared-unresolvable → install-remedy; declared-no-loadable-entry → a package-shape remedy (or defer to the importer's message, which already words it). Update the spelling pin test in the same PR.

Population

Latent until a real host package publishes no loadable entry (the organizations runtime and shipped cluster drivers are dual-published, so today this is reachable only by a broken third-party or customer package). Same latency class as #14041 itself.

Generated by Claude Code

[huangyiirene]

Triage (R+89, triage seat, session session_019kDRpB7D2XzVzkaLp57T5D): pm:queue · priority:p3 · domain:cli · type Bug.

The body's Blocked-by: #14041 is spent: #14041 closed on 2026-09-01T21:50Z (PR #14271 landed), so the third HostImportFailureKind is on origin/main and the three two-way branches are wrong in code today, not latently. ⇒ Straight to queue, no pm:blocked.

Lands in: packages/cli/src/commands/serve.ts (formatOrganizationsInstallRemedy), packages/verify/src/harness.ts (~bootStack's multi-tenant fatal), packages/qa/dogfood/test/enterprise-organizations.ts (the enterprise probe), plus serve-organizations-message-spelling.test.ts. Ruled shape: branch on "is the declaration the problem?" — undeclared ⇒ declare-remedy; declared-unresolvable ⇒ install-remedy; declared-no-loadable-entry ⇒ defer to the importer's own message (which already words the package-shape case) rather than minting a fourth remedy sentence in three places. p3: reachable today only through a broken third-party/customer package, and each site also prints the importer's message, so the wrong bullet contradicts rather than replaces the right one.

Family: #14278 (same parent, packages/types). Size/model suggestion: S, sonnet/opus.

---

Generated by Claude Code