fix(avatar-pack): bg removal, style regen consistency, and doc restructure

[latentflux42]

Summary

• Background removal: add rembg fallback, lower checkerboard detection threshold, add connected-component island removal for enclosed remnants
• Style regen consistency: remove stale base_image_original.png guard so profile card matches new style after scenario C option 2; use local processed path for profile card generation
• Doc restructure: move output rules and file paths from SKILL.md to GENERATE.md; simplify SKILL.md description to English
• Misc: add __pycache__/ to .gitignore, fix potential UnboundLocalError in rembg cleanup

Test plan

• Generate a fresh avatar (scenario B) — verify all files created and consistent
• Adjust style (scenario C option 2) — verify profile card matches new expressions
• Regenerate single expression (scenario C option 3) — verify original preserved
• Test with images that have checkerboard backgrounds — verify clean removal
• Test rembg fallback when rembg is not installed

🤖 Generated with Claude Code

---

Note

Medium Risk
Adds a new skill plus a large image-processing pipeline (background removal, GIF/sticker generation) that could fail on edge-case inputs or missing system deps, but is isolated to the new cola-avatar-pack/ tooling.

Overview
Introduces a new cola-avatar-pack skill that generates a pixel-art self-portrait, a profile card, 4 animated emoji GIFs, and 3 meme stickers, with a documented multi-phase flow for first-time generation and regeneration.

Adds cola-avatar-pack/scripts/process_avatar.py to handle background removal (including optional rembg CLI fallback and artifact cleanup), canvas fitting, animation frame generation, profile-card rendering (wuxing colors + rarity diamonds), and dual-size (128px + @2x) outputs for both GIFs and PNG stickers.

Updates docs/changelog to surface the new skill in README.md, README.zh.md, and CHANGELOG.md, and extends .gitignore for Python cache/artifacts.

^{Reviewed by Cursor Bugbot for commit a9d7f70. Bugbot is set up for automated code reviews on this repo. Configure here.}

[cursor]

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

Autofix Details

Bugbot Autofix prepared a fix for the issue found in the latest run.

• ✅ Fixed: Happy GIF always overwritten during single-expression regeneration
  • I made happy.gif generation conditional during targeted expression/meme regeneration and allowed meme-only runs to proceed without requiring any expression input.

Or push these changes by commenting:

@cursor push efb373116a

Preview (efb373116a)

diff --git a/avatar-pack/scripts/process_avatar.py b/avatar-pack/scripts/process_avatar.py
--- a/avatar-pack/scripts/process_avatar.py
+++ b/avatar-pack/scripts/process_avatar.py
@@ -1203,12 +1203,17 @@
         return
 
     # Process expression images — skip any not provided (supports single-expression regen)
+    has_targeted_regen_input = any((
+        args.sad, args.angry, args.thinking,
+        args.meme_confused, args.meme_annoyed, args.meme_cracked,
+    ))
     images = {
-        'happy': args.base,
         'sad': args.sad,
         'angry': args.angry,
         'thinking': args.thinking,
     }
+    if not is_expression_only or not has_targeted_regen_input:
+        images['happy'] = args.base
 
     available = {}
     for emotion, path in images.items():
@@ -1217,7 +1222,8 @@
         elif path:
             print(f'Warning: {emotion} image not found: {path}, skipping', file=sys.stderr)
 
-    if not available:
+    has_meme_inputs = any((args.meme_confused, args.meme_annoyed, args.meme_cracked))
+    if not available and not has_meme_inputs:
         print('Error: no expression images found', file=sys.stderr)
         sys.exit(1)

_{This Bugbot Autofix run was free. To enable autofix for future PRs, go to the Cursor dashboard.}

[latentflux42]

Fixed in e1f528d. Added --regen-happy flag — full generation (Phase 6) passes it, single-expression regen does not. Also decoupled meme-only invocations from expression input requirements.

Resolves #20 (comment)

[latentflux42]

Fixed in d4bdb4b — removed unused import math.

Resolves #20 (comment)

[latentflux42]

Fixed in 356f2cc — base_image_original.png is already background-removed by save_base_image, so the second remove_background call in generate_profile_card was redundant and could cumulatively erode light/neutral character features. Removed it for the original branch; kept it in the else fallback for unprocessed inputs.

Resolves #20 (comment)

[latentflux42]

Fixed in ac21693 — premultiply diamond outline color against white background so it stays fully opaque (alpha=255) while looking visually identical. Eliminates transparency holes on dark backgrounds.

Resolves #20 (comment)

[cursor]

Cursor Bugbot has reviewed your changes 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 ac21693. Configure here.}

[latentflux42]

Won't fix — the 24px clipping occurs only at the apex frame (1 of 8, 100ms) at @2x resolution. At 128px display size it's 12px, imperceptible during animation playback. Reducing stretch to eliminate it would weaken the bounce expressiveness for no visible gain.

Resolves #20 (comment)