…er e2e — baseline 6 → 0 (#2352)

Bite B of #2354 umbrella (drive stub-debt baseline to zero). Resolves the
6 `@ts-expect-error -- fork stub typing mismatch` suppressions in
src/auto-reply/reply/agent-runner.runreplyagent.e2e.test.ts by properly
typing the local `modelFallbackModule` spy host with `vi.fn<ModelFallbackFn>()`
instead of `{ runWithModelFallback: vi.fn() } as any`. When the host was
`any`, `vi.spyOn` resolved to the untyped overload `(...args: unknown[]) => any`
and the callback impls' strictly-typed `{ run }` parameters were contravariantly
incompatible — removing `as any` and binding the mock's function signature
makes the spies accept the existing callback shapes without suppression.

Changes:
- `src/auto-reply/reply/agent-runner.runreplyagent.e2e.test.ts`:
  - Replace `const modelFallbackModule = { runWithModelFallback: vi.fn() } as any;`
    (with `eslint-disable-next-line @typescript-eslint/no-explicit-any`) with a
    local `ModelFallbackFn` function type and `vi.fn<ModelFallbackFn>()`.
  - Delete 6 `// @ts-expect-error -- fork stub typing mismatch` comments.
  - Formatter (oxfmt) reflows 5/6 `vi.spyOn(...).mockImplementation(...)`
    chains into multi-line builder-chain style — cosmetic only, no behavior
    change. Callback bodies (mock return shape, `expect()` assertions) are
    byte-identical modulo whitespace.
- `.stub-debt-baseline`: `6` → `0`.

The baseline file and `scripts/check-stub-debt.mjs` gate simplification
(Bite C) remain the next step to close #2354 — a separate follow-up PR.

Verification:
- `node scripts/check-stub-debt.mjs` → `stub-debt check passed: 0 == baseline 0.`
  (H8 fork-boundary-mock counter unchanged: `132 == baseline 132`).
- `pnpm check` (format + tsgo + lint + project-specific lints) → exit 0.
- E2E file runtime parity: 37 failed / 5 passed / 42 total — identical to
  pre-change. The 37 failures are pre-existing (resolveFallbackTransition
  gutted per Middleware Boundary Principle); not touched by this PR and
  not run by CI's `pnpm test` (which excludes e2e via test-parallel.mjs).
- Adversarial validation (fresh-context subclaude): CLEAN verdict on 7
  AC + 9 adversarial checks (no bypass patterns, no stealth normalization,
  no H8 regression, no formatter side effects, Bite C gate-simplification
  ready).

Refs: #2354, #2457, ADR 0005 H5

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

…er e2e — baseline 6 → 0 (#2352) (#2458)

Bite B of #2354 umbrella (drive stub-debt baseline to zero). Resolves the
6 `@ts-expect-error -- fork stub typing mismatch` suppressions in
src/auto-reply/reply/agent-runner.runreplyagent.e2e.test.ts by properly
typing the local `modelFallbackModule` spy host with `vi.fn<ModelFallbackFn>()`
instead of `{ runWithModelFallback: vi.fn() } as any`. When the host was
`any`, `vi.spyOn` resolved to the untyped overload `(...args: unknown[]) => any`
and the callback impls' strictly-typed `{ run }` parameters were contravariantly
incompatible — removing `as any` and binding the mock's function signature
makes the spies accept the existing callback shapes without suppression.

Changes:
- `src/auto-reply/reply/agent-runner.runreplyagent.e2e.test.ts`:
  - Replace `const modelFallbackModule = { runWithModelFallback: vi.fn() } as any;`
    (with `eslint-disable-next-line @typescript-eslint/no-explicit-any`) with a
    local `ModelFallbackFn` function type and `vi.fn<ModelFallbackFn>()`.
  - Delete 6 `// @ts-expect-error -- fork stub typing mismatch` comments.
  - Formatter (oxfmt) reflows 5/6 `vi.spyOn(...).mockImplementation(...)`
    chains into multi-line builder-chain style — cosmetic only, no behavior
    change. Callback bodies (mock return shape, `expect()` assertions) are
    byte-identical modulo whitespace.
- `.stub-debt-baseline`: `6` → `0`.

The baseline file and `scripts/check-stub-debt.mjs` gate simplification
(Bite C) remain the next step to close #2354 — a separate follow-up PR.

Verification:
- `node scripts/check-stub-debt.mjs` → `stub-debt check passed: 0 == baseline 0.`
  (H8 fork-boundary-mock counter unchanged: `132 == baseline 132`).
- `pnpm check` (format + tsgo + lint + project-specific lints) → exit 0.
- E2E file runtime parity: 37 failed / 5 passed / 42 total — identical to
  pre-change. The 37 failures are pre-existing (resolveFallbackTransition
  gutted per Middleware Boundary Principle); not touched by this PR and
  not run by CI's `pnpm test` (which excludes e2e via test-parallel.mjs).
- Adversarial validation (fresh-context subclaude): CLEAN verdict on 7
  AC + 9 adversarial checks (no bypass patterns, no stealth normalization,
  no H8 regression, no formatter side effects, Bite C gate-simplification
  ready).

Refs: #2354, #2457, ADR 0005 H5

Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

… — Bite C (#2354)

**Bite C** of the #2354 umbrella (drive stub-debt baseline to zero and
remove it). Bites A (#2457, baseline 12 → 6) and B (#2458, baseline 6 → 0)
drove the H5 `@ts-expect-error` count to zero across `src/`, `extensions/`,
and `ui/`. This PR retires the H5 baseline entirely and flips the gate to
zero-tolerance.

The H8 fork-boundary-mock counter (`.fork-boundary-mock-baseline = 132`)
is untouched — still baselined, still ratchetable.

Changes:
- `scripts/check-stub-debt.mjs`: inline the H5 counter as a zero-tolerance
  check (was baseline-gated via `readBaseline`+`reportCounter`). Any
  `@ts-expect-error` in a gated file fails the script with an inventory
  and a remediation pointer to Bite A (`as unknown as T` cast pattern,
  PR #2457) and Bite B (`vi.fn<Fn>()` typed-mock pattern, PR #2458). The
  H8 branch — `readBaseline` + `reportCounter` for `.fork-boundary-mock-baseline`
  — is byte-identical to pre-change; inline-for-H5 / helper-for-H8
  asymmetry is intentional since the two counters are now fundamentally
  different patterns.
- `.stub-debt-baseline`: DELETED (last value was `0`).
- `CLAUDE.md:121`: rewrote the `stub-debt-gate` bullet to reflect
  zero-tolerance H5 + link to `CONTRIBUTING.md § Fork-boundary mocks`
  for H8 detail.
- `CONTRIBUTING.md:217`: row label for the sync-pr-audit composite
  summary updated from `H8 stub-debt + fork-boundary-mock baselines`
  to `H5 stub-debt (strict) + H8 fork-boundary-mock` — also corrects
  pre-existing mislabel of H5 as H8. Script reference unchanged.
- `.github/workflows/sync-pr-audit.yml:116`: same row-label update;
  script invocation and exit-code variable `H8` intentionally unchanged
  to minimize churn.

Verification:
- `node scripts/check-stub-debt.mjs` → `stub-debt check passed: 0
  @ts-expect-error suppressions.` + `fork-boundary-mock check passed:
  132 == baseline 132.` (exit 0).
- Failure-path probe (temp `@ts-expect-error` under `src/`) →
  script prints inventory, remediation guide, and exits 1. Confirms
  the zero-tolerance branch.
- `pnpm check` (format + tsgo + lint + project-specific lints) →
  exit 0.
- `pnpm test` (full unit+extensions+gateway suite) → 7010 passed /
  3 skipped / 7013 total.
- Rescan: `git ls-files | xargs grep -l "\.stub-debt-baseline"` →
  zero hits. No stale references remain.
- Adversarial validation (fresh-context subclaude): 5 AC PASS +
  13 adversarial checks PASS; 1 MINOR cross-repo finding (HQ ADR
  0005 H5 staleness) tracked as post-merge HQ follow-up, does NOT
  block this PR.

Closes #2354

Refs: #2457, #2458, ADR 0005 H5

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

… — Bite C (#2354) (#2459)

**Bite C** of the #2354 umbrella (drive stub-debt baseline to zero and
remove it). Bites A (#2457, baseline 12 → 6) and B (#2458, baseline 6 → 0)
drove the H5 `@ts-expect-error` count to zero across `src/`, `extensions/`,
and `ui/`. This PR retires the H5 baseline entirely and flips the gate to
zero-tolerance.

The H8 fork-boundary-mock counter (`.fork-boundary-mock-baseline = 132`)
is untouched — still baselined, still ratchetable.

Changes:
- `scripts/check-stub-debt.mjs`: inline the H5 counter as a zero-tolerance
  check (was baseline-gated via `readBaseline`+`reportCounter`). Any
  `@ts-expect-error` in a gated file fails the script with an inventory
  and a remediation pointer to Bite A (`as unknown as T` cast pattern,
  PR #2457) and Bite B (`vi.fn<Fn>()` typed-mock pattern, PR #2458). The
  H8 branch — `readBaseline` + `reportCounter` for `.fork-boundary-mock-baseline`
  — is byte-identical to pre-change; inline-for-H5 / helper-for-H8
  asymmetry is intentional since the two counters are now fundamentally
  different patterns.
- `.stub-debt-baseline`: DELETED (last value was `0`).
- `CLAUDE.md:121`: rewrote the `stub-debt-gate` bullet to reflect
  zero-tolerance H5 + link to `CONTRIBUTING.md § Fork-boundary mocks`
  for H8 detail.
- `CONTRIBUTING.md:217`: row label for the sync-pr-audit composite
  summary updated from `H8 stub-debt + fork-boundary-mock baselines`
  to `H5 stub-debt (strict) + H8 fork-boundary-mock` — also corrects
  pre-existing mislabel of H5 as H8. Script reference unchanged.
- `.github/workflows/sync-pr-audit.yml:116`: same row-label update;
  script invocation and exit-code variable `H8` intentionally unchanged
  to minimize churn.

Verification:
- `node scripts/check-stub-debt.mjs` → `stub-debt check passed: 0
  @ts-expect-error suppressions.` + `fork-boundary-mock check passed:
  132 == baseline 132.` (exit 0).
- Failure-path probe (temp `@ts-expect-error` under `src/`) →
  script prints inventory, remediation guide, and exits 1. Confirms
  the zero-tolerance branch.
- `pnpm check` (format + tsgo + lint + project-specific lints) →
  exit 0.
- `pnpm test` (full unit+extensions+gateway suite) → 7010 passed /
  3 skipped / 7013 total.
- Rescan: `git ls-files | xargs grep -l "\.stub-debt-baseline"` →
  zero hits. No stale references remain.
- Adversarial validation (fresh-context subclaude): 5 AC PASS +
  13 adversarial checks PASS; 1 MINOR cross-repo finding (HQ ADR
  0005 H5 staleness) tracked as post-merge HQ follow-up, does NOT
  block this PR.

Closes #2354

Refs: #2457, #2458, ADR 0005 H5

Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>