feat(hooks): import reviewed Claude and Codex commands - #1840

Summary

• Add explicit preview/apply/update management for the supported synchronous command Hook subset from Claude Code and Codex across root CLI, interactive TUI, and Desktop settings.
• Keep AgentHookEngine as the only executor. Ecosystem adapters discover and prepare; core owns plan/apply fencing; the service store owns private managed snapshots.
• Unify native, discovered, and imported Hook status under bitfun hooks / /hooks, with old external-Hook commands retained only as aliases.

Product and architecture boundaries

• Import is always explicit and command-visible before apply; source files are never modified.
• User and workspace imports remain independently managed with enable, disable, update, remove, and corrupt-store reset actions.
• OpenCode Hooks, non-command or asynchronous handlers, generic Hook Runtime, Plugin Runtime integration, startup import, file watching, Remote/Peer execution, and scheduler/agent-loop changes are intentionally excluded.
• Imported layers reuse the existing native Hook parser, execution engine, event ordering, timeout, audit, and next-event cache reload behavior.
• Process-only docs/superpowers/** material is excluded; durable behavior and ownership constraints stay in the existing architecture and feature documents.

User experience

• Desktop uses one review dialog and existing settings controls; TUI uses numbered /hooks actions; root CLI retains the exact fingerprint only where cross-process non-interactive confirmation requires it.
• GUI/TUI do not expose the internal fingerprint. Initial import and update use distinct confirmation copy, and completion messages reflect disabled imports and the global Hook switch accurately.
• The project-file gate is labeled as the project Hook file so it is not confused with independently reviewed project imports.
• Refresh is explicit and asynchronous. No startup import, file watcher, or automatic replay is introduced.

Safety and failure handling

• Plans are bound to the redacted catalog revision and exact prepared command/assets fingerprint; stale actions fail without automatic replay.
• Private snapshots enforce file/count/depth/byte budgets, no-follow managed paths, per-store cross-process locking, content-digest verification, and fail-closed corrupt or tampered bundles.
• Runtime engine rebuilds verify each exact managed bundle once and pass the same verified hooks.json bytes to the native parser; an earlier validity bit cannot authorize newly read commands.
• Bundle publication stages and validates first, rolls back on index publication failure, and removes retired snapshots after a successful commit on a best-effort path.
• Updates preserve the authoritative enabled/disabled state, and all user-facing surfaces report that state instead of implying that a disabled import became active.
• Unsupported or ambiguous source semantics remain visible as skipped/unsupported diagnostics and never silently become executable.

Validation

• cargo check --workspace
• cargo test -p bitfun-services-integrations --no-default-features --features hook-import hook_import (13 passed)
• cargo test -p bitfun-cli hook --no-default-features (19 passed)
• pnpm --dir src/web-ui run test:run src/infrastructure/config/components/HooksConfig.test.tsx (6 passed)
• Focused Claude Code, Codex, parser, coordinator, core/native Hook, desktop API, CLI/TUI, and Web UI tests
• pnpm run type-check:web
• pnpm run i18n:contract:test
• pnpm run i18n:audit
• pnpm run lint:web
• node scripts/check-core-boundaries.mjs
• pnpm run check:repo-hygiene
• git diff --check
• Context-isolated adversarial review findings fixed, followed by a final scope and UX audit of the updated paths

CI is intentionally not awaited for this draft.