Skip to content

AUDIT-PERFECTION-PLAN-v1.23.md

Latest commit

 

History

History
140 lines (115 loc) · 8.81 KB

AUDIT-PERFECTION-PLAN-v1.23.md

File metadata and controls

140 lines (115 loc) · 8.81 KB
Raw

WorldScript Studio — Audit / Correction / Perfection Plan (v1.23 cycle)

Living master artifact for the "Master Orchestration Prompt v1.1" engagement. Single tracked home for the 6-phase audit toward v1.23. Updated per batch. Engagement model: one PR per phase, stacked; CodeAnt re-reviews each push and all inline comments are cleared (0 unresolved) before a phase is declared done.

Owner: maintainer + Claude Code · Started: 2026-06-13 · Baseline version: 1.22.0 v1.23.0 shipped: 2026-06-16 · post-release documentation perfection pass: 2026-06-16 (corpus sync, metric/flag reconciliation, history archival, dependabot cooldown, GitHub release/package rebrand — see AUDIT.md § v1.23.0 Post-Release Documentation Perfection Pass). Phases 2–6 below remain forward work in the v1.24 / v2.0 cycle.

1. Corrected baseline (verified against live main, 2026-06-13)

An earlier prompt revision (v1.0) carried a stale gap list. Fresh inspection confirms the following are already shipped — do not re-plan them as gaps:

Area Status Evidence
ROADMAP/TODO/AUDIT consistency ✅ reconciled PR #118 (e8252b6, 6b3340d)
v1.22 Smoke-Test protocol ✅ exists docs/V1.22-SMOKE-TEST.md (7 sections)
Error Boundaries ✅ all 19+ views App.tsx + components/ui/ViewErrorBoundary.tsx
Plugin worker isolation ✅ delivered + tested workers/plugin.worker.ts, services/pluginRegistry.ts, 5 test files
Dependency audit (high + moderate) ✅ 0 vulnerabilities pnpm audit 2026-06-13; AUDIT.md overrides table

2. Phase roadmap & status

Phase Theme Status
0 v1.23 foundation: tracker sync, dependency hygiene, plugin sandbox validation ✅ Merged (PR #118 → 8f94178)
1 Reliability/Observability + Local-AI/Voice low-end hardening ✅ Complete (1.1, FU-1, 1.2, CP-i18n, Writer-errors + correlation-IDs; action-buttons = optional future polish)
2 Coverage elevation (L≥85 / B≥75 / F≥80) — multi-batch 🔄 In progress (2.1 services/fs/ done; threshold ratchet pending CI floor)
3 Tauri desktop: signing / notarization / updater pipeline + UX ⬜ Planned
4 WCAG 2.2 AA manual audit + i18n sustainability (<2% placeholders) ⬜ Planned
5 Performance / bundle / state-model clarity / ProForge isolation ⬜ Planned
6 Docs, ADRs, release checklist, onboarding ⬜ Planned

Each phase opens its own PR off main with an approved plan before execution.

3. Batch log

Batch 0.1 — Tracker reconciliation (PR #118, e8252b6+6b3340d)

ROADMAP/TODO/AUDIT brought into agreement; AUDIT Known Overrides table refreshed (esbuild row added, placeholder CVE refs replaced with verified advisory IDs); dependency audit re-verified 0 high/moderate. CHANGELOG [1.22.0] heading restored after a CodeAnt-caught regression.

Batch 0.2 — Plugin sandbox adversarial validation (PR #118)

  • tests/unit/workers/plugin.worker.test.ts: +5 adversarial tests — WebAssembly denial, GeneratorFunction + AsyncGeneratorFunction constructor-escape blocked, and self.* guard restoration on both success and error paths.
  • Verified tests/unit/pluginRegistry.test.ts already covers storage-key validation (prefix / length / .. traversal / path separators / empty suffix) and the 2 MiB value-size cap — no new registry tests needed.

Batch 1.1 — AI error taxonomy + fail-fast retry (Phase 1)

  • New services/ai/aiErrorTaxonomy.ts — pure classifyAiError(err) → {category, retryable, messageKey} over transient/rateLimit/auth/network/offline/policy/ invalidRequest/canceled/permanent (reuses the aiPolicy.ts "blocked" markers + HTTP status; AbortError cancellations fail fast — CodeAnt-flagged).
  • services/ai/aiRetry.ts withTransientRetry now fails fast on non-retryable errors (auth/policy/invalid-request/offline) via a shouldRetry default, and emits a structured createLogger('ai.retry') line per decision with a per-call correlation id. Backward compatible (unknown errors still retry). +33 tests (aiErrorTaxonomy.test.ts table-driven, fail-fast cases in aiRetry.test.ts).
  • i18n / UI mapping of messageKey deferred to Batch 1.2 (avoids 11-locale churn here).
  • FU-1 split out of this batch (fixed separately — see batch log).

Batch FU-1 — plugin-worker guard-restoration fix

  • Root cause (confirmed by probe): installRuntimeGuards reassigns the global self.Function = denied, so restoreRuntimeGuards's free Function.prototype.constructor reference resolved to denied.prototype (the stub), patching the wrong object and leaving the real Function.prototype.constructor denied. self.* restored fine because they use explicit self.X = assignments; the async/generator paths use module-captured consts, so only the bare Function identifier was poisoned.
  • Fix: use the module-captured GlobalFunction (native, never reassigned) for the three Function.prototype.constructor touch-points (snapshot capture, install, restore). One-file change in workers/plugin.worker.ts; isolation unchanged (denied stays active during runs).
  • Tests: re-enabled the Function.prototype.constructor round-trip assertions (success + error path) and added a cross-run isolation test in plugin.worker.test.ts (24 green).

Batch 1.2 — user-facing AI error messages (Phase 1)

  • Realigned the taxonomy messageKey to the existing singular error.ai.* convention and added 9 error.ai.<category> keys to all 11 locales (common.json; translated de/en/es/fr/it, English fallback Beta/RTL) — 2599 keys, i18n:check green.
  • getAiErrorMessage(err, t) helper + wired into useGlobalCopilot onError so the Copilot shows an actionable, localized message (e.g. "Invalid API key — open Settings → AI…") instead of the generic copilot.error. +helper test + 2 onError tests.
  • Out of scope (later): action buttons; Writer/other AI call sites (helper is ready).

Batch CP-i18n — Command Palette localization (i18n completeness)

  • Root cause: palette infra is correct (titleKey + t()), but ~20 palette.* keys per core locale were untranslated English fallback (AI-mode/editor/appearance/accessibility commands + 5 category headers). Keys existed → parity passed → values were never translated.
  • Translated for de/es/fr/it (values-only, no key churn); bundles rebuilt; i18n:check green.
  • Guard: tests/unit/i18n/paletteLocalization.test.ts fails if a palette.* key reverts to the English value in a core locale (allowlist for loanwords: Navigation/Editor). Beta/RTL unchanged.

Batch P1-Finish — Writer error UX + correlation IDs (closes Phase 1)

  • useWriterView.handleFailure now uses getAiErrorMessage(err, t) (classified, localized) instead of a hardcoded English string + localizes the [Cancelled] tag (writer.cancelledTag, 11 locales). Parity to the Copilot (Batch 1.2).
  • Correlation IDs: newCorrelationId() (services/logger.ts); useWorldScriptAI mints one per request, logs the lifecycle, and propagates it via the per-call body; worldScriptCompletionFetch reads it (schema) and logs it on failure (still never exposes err.message); withTransientRetry accepts a propagated id. No prompts/keys logged.
  • Phase 1 = ✅ complete. Optional future polish: error action-buttons; correlation IDs through the legacy aiProviderService/thunk path.

Batch 2.1 — Coverage: services/fs/ (Tauri FS backend)

  • services/fs/ (project/snapshot/asset/settings/codex stores + fsCore) was 0% tested (~1,000 lines). Now ~81% line / ~66% branch via 37 tests: fsCore.test.ts (pure utils) + fsStores.test.ts (an in-memory fake TauriApis driving real round-trips — compression, AES-GCM key encryption, JSON, import/export markdown/json). The whole store chain is exercised through one FsProjectStore (linear inheritance).
  • Threshold ratchet deferred to a follow-up keyed to the new CI-measured global floor (can't measure the global gate locally on this hardware).
  • Next: 2.2 ratchet + ProForge capability layer; 2.3+ storage/encryption stores, then branch-coverage deepening toward L85/B75/F80.

4. Decisions & follow-ups

  • FU-1 — ✅ RESOLVED (see Batch FU-1 above). The guard-restoration leak was a poisoned global-binding bug, not a property-attribute issue; fixed by routing the three Function.prototype.constructor touch-points through the captured GlobalFunction.

5. References

  • Engagement prompt: Master Orchestration Prompt v1.1 (corrected baseline).
  • ROADMAP.md (v1.23 section), TODO.md (P0 block), AUDIT.md (overrides + security), CHANGELOG.md ([Unreleased]).
  • Plugin sandbox: docs/PLUGINS-BETA.md, services/pluginRegistry.ts, workers/plugin.worker.ts.