feat(0130-FaceLandmarkerSidecar): 3.12 Face Landmarker → 0124 assist pipe

[Ryan-AI-Studios]

Summary

• Optional Python 3.12 Face Landmarker sidecar under scripts/face-landmarker/ (mediapipe==1.0.1) maps 478 → 0124 form-read ids and dumps assist_pixel_capture for existing meshops proportion capture --source px --merge --prefer-merge.
• Frozen map SoT: src/meshops/proportion/face_landmarker_map.json (+ 3.13 loader). MeshOps never imports mediapipe.
• Capture + assist preserve method=pose_model through Landmark2D. run.ps1 skip-ok when 3.12 missing.
• MCP stay 53. No remake / D7. Recipe 1.4.0 / report 1.2.0 unchanged.

Honesty

face_landmarker_sidecar_not_mesh_or_print_success — detector pixels ≠ mesh/print success (Difficulty §12).

Test plan

• ledgerful verify (ruff / format / basedpyright / pytest)
• tests/test_proportion_face_landmarker.py T0–T10
• CI quality + design

---

Note

Medium Risk
Touches proportion capture/assist serialization and adds an isolated detector path that can fill landmarks; core 3.13 runtime and MCP stay unchanged, but merge/method handling is business-logic.

Overview
Adds an optional Python 3.12 sidecar that maps MediaPipe Face Landmarker (478 points) onto existing 0124 form-read ids and writes assist_pixel_capture JSON for meshops proportion capture --source px. MeshOps stays on 3.13 and never imports mediapipe; operators run run.ps1 / run.py then merge with --prefer-merge.

The frozen map lives in face_landmarker_map.json (packaged with the wheel) plus a 3.13 loader that maps indices, drops Z, skips cranial_vertex, and treats 0/2+ faces as skip. Capture/assist now round-trip method (e.g. pose_model) on landmark dicts. Sidecar missing 3.12/model/face is skip-ok (exit 2). MCP catalog stays at 53 tools.

^{Reviewed by Cursor Bugbot for commit 31dbd80. Bugbot is set up for automated code reviews on this repo. Configure here.}

[cursor]

Cursor Bugbot has reviewed your changes using default effort and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 31dbd80. Configure here.}

[cursor]

Multi-face skip never triggers

Medium Severity

num_faces=1 is MediaPipe’s maximum faces to return, not an exact-one requirement. With that cap the detector never yields 2+ faces, so the multi_face skip is dead and a multi-face image silently gets one chosen face—against the stated §1 rule not to pick a primary.

Additional Locations (1)

• scripts/face-landmarker/README.md#L79-L80

 

^{Reviewed by Cursor Bugbot for commit 31dbd80. Configure here.}