Summary

Closes #2057.

Hook-driven transcript mining is currently hardcoded to --wing sessions, even though the diary checkpoint path already derives a stable project wing from the transcript JSONL cwd/path. This adds a backward-compatible routing policy:

{
  "hooks": {
    "transcript_wing": "sessions"
  }
}

• sessions remains the default and preserves current behavior.
• project derives the stable wing through the existing _wing_from_transcript_path() logic.
• MEMPALACE_HOOK_TRANSCRIPT_WING overrides file config.
• invalid values fail safe to sessions rather than silently relocating memories.

Implementation

• Add MempalaceConfig.hook_transcript_wing with sessions|project validation.
• Resolve the transcript wing once for both direct and daemon hook write routes.
• Keep ambiguous project derivation on sessions rather than creating a surprise wing_sessions migration.
• Include the selected wing in daemon mine dedupe keys so the same source can be intentionally routed to a different wing without sharing an in-flight slot.
• Preserve the existing direct-spawn per-target guard, whose command hash already includes --wing.

Test-first verification

The repair branch was created from current upstream develop and used a temporary owner-controlled test-first workflow. The workflow:

1. added the regression tests and observed the pre-fix failures;
2. applied the minimal config/routing change;
3. ran the focused transcript-wing tests;
4. ran tests/test_hooks_cli.py and tests/test_hook_write_routing.py;
5. ran Ruff, compileall, and git diff --check;
6. removed its temporary repair script/workflow before committing.

The submitted branch contains only:

• mempalace/config.py
• mempalace/hooks_cli.py
• tests/test_hooks_cli.py

Risk

Low and opt-in. Existing installations remain on the historical sessions wing unless they explicitly select project. No storage schema, existing drawer, or default hook behavior is changed.