Summary

Inventory all (..._args: unknown[]): never throwing stubs across src/ and classify live-caller status. Sibling to #2337 (which covers silent-neuter stubs: => false, => {}, => []). This audit targets the unconditional-throw variant that shipped in #2408.

Rationale

The resolveAgentRuntimeOrThrow regression (#2408) is one instance of a class: a stub replaced with an unconditional throw during upstream sync, left with live production callers, passing CI because tests mock the stub. #2337 documents the related silent-neuter class — 3 files with hidden runtime regressions (Telegram vision, node-host skills, TUI listModels). Neither #2337 nor #2406 enumerates the throwing-stub variant. Likelihood other instances exist: moderate — the pattern is grep-able.

Scope

Enumerate all exports whose body is an unconditional throw, matching any of:

# Pattern A: explicit never-return stub with variadic unknown args
rg --type ts 'function\s+\w+\([^)]*\.\.\._?args:\s*unknown\[\][^)]*\):\s*never\s*\{' src/

# Pattern B: any "is not available in RemoteClaw fork" throw
rg --type ts 'throw new Error\("[^"]*is not available in RemoteClaw fork[^"]*"\)' src/

# Pattern C: stubs throwing "gutted" / upstream-compat messages
rg --type ts -B 2 'throw new Error\("[^"]*(gutted|upstream-compat)[^"]*"\)' src/

For each matching symbol:

• Identify live callers via rg --type ts "\b<symbol>\b" src/ extensions/ excluding *.test.ts, *.test-helpers.ts, test-helpers/**, *.test-mocks.ts, *.test-utils.ts.
• Classify:
  • (a) Deletable dead: no live callers; safe to remove both stub and any test-only mocks.
  • (b) Transitively dead: live callers exist, but those callers are themselves dead (reachable only through other gutted chains). Document the chain.
  • (c) LIVE REGRESSION: live callers in reachable code paths (same class as fix(agents): restore resolveAgentRuntimeOrThrow body — regression stub crashed every agent dispatch #2408).

Deliverable

A comment on this issue containing:

1. Grep commands and their raw output.
2. Classification table per stub: symbol | file:line | live callers (count + first 3 locations) | class a/b/c | disposition.
3. Triage: class-c findings filed as separate fix issues, added to gut: re-apply provider/model cascade after v2026.3.7 sync regression (~19 files) #2337 scope if same sync origin, or bundled here if trivial count.
4. Class-a/b findings documented but not necessarily acted on in this issue.

Acceptance criteria

• Audit commands run on a pinned commit SHA (recorded in comment).
• Every matching stub appears in the classification table.
• Each class-c finding has a follow-up issue, or a trivial-bundle note in this issue.
• Class-a/b findings recorded for later cleanup triage.
• Audit comment cross-references fix(agents): restore resolveAgentRuntimeOrThrow body — regression stub crashed every agent dispatch #2408 and gut: re-apply provider/model cascade after v2026.3.7 sync regression (~19 files) #2337.

Related

• Seed example: fix(agents): restore resolveAgentRuntimeOrThrow body — regression stub crashed every agent dispatch #2408 — the resolveAgentRuntimeOrThrow regression that triggered this audit.
• Parallel class: gut: re-apply provider/model cascade after v2026.3.7 sync regression (~19 files) #2337 — silent-neuter stubs with hidden runtime regressions.
• Meta-audit: Audit: upstream fixes silently lost in sync (v2026.2.25 → v2026.3.22, 32 gaps) #2406 — upstream fixes silently lost in sync.