feat: extract previous review feedback during regeneration

- Add step to extract previous review feedback into a markdown file
- Include quality score, strengths, weaknesses, and criteria checklist
- Update documentation to reflect mandatory reading for regenerations

Author: MarkusNeusinger

.github/workflows/impl-generate.yml

@@ -249,6 +249,65 @@ jobs:
             echo "is_regeneration=false" >> $GITHUB_OUTPUT
           fi
 
+      - name: Extract previous review feedback (regeneration)
+        id: prev_review
+        if: steps.existing.outputs.is_regeneration == 'true'
+        env:
+          SPEC_ID: ${{ steps.inputs.outputs.specification_id }}
+          LANGUAGE: ${{ steps.inputs.outputs.language }}
+          LIBRARY: ${{ steps.inputs.outputs.library }}
+        run: |
+          .venv/bin/python3 - <<'PY'
+          import os, pathlib, yaml
+
+          spec = os.environ["SPEC_ID"]
+          lang = os.environ["LANGUAGE"]
+          lib = os.environ["LIBRARY"]
+          meta_path = pathlib.Path(f"plots/{spec}/metadata/{lang}/{lib}.yaml")
+          data = yaml.safe_load(meta_path.read_text(encoding="utf-8")) or {}
+          review = data.get("review") or {}
+          quality = data.get("quality_score")
+
+          lines = [f"# Previous Review for {spec} / {lang} / {lib}", ""]
+          lines.append(f"**Previous quality score:** {quality if quality is not None else 'n/a'}")
+          lines.append("")
+
+          desc = review.get("image_description")
+          if desc:
+              lines += ["## Previous image description", str(desc).strip(), ""]
+
+          strengths = review.get("strengths") or []
+          if strengths:
+              lines.append("## Strengths (KEEP these)")
+              lines += [f"- {s}" for s in strengths]
+              lines.append("")
+
+          weaknesses = review.get("weaknesses") or []
+          if weaknesses:
+              lines.append("## Weaknesses (FIX these)")
+              lines += [f"- {w}" for w in weaknesses]
+              lines.append("")
+
+          checklist = review.get("criteria_checklist") or {}
+          if checklist:
+              lines.append("## Criteria checklist (focus on items that failed)")
+              for cat, payload in checklist.items():
+                  payload = payload or {}
+                  score = payload.get("score", "?")
+                  max_score = payload.get("max", "?")
+                  lines.append(f"### {cat}  ({score}/{max_score})")
+                  for item in payload.get("items") or []:
+                      item = item or {}
+                      mark = "✅" if item.get("passed") else "❌"
+                      lines.append(
+                          f"- {mark} {item.get('id', '?')} {item.get('name', '')}: {item.get('comment', '')}"
+                      )
+                  lines.append("")
+
+          pathlib.Path("/tmp/anyplot-prev-review.md").write_text("\n".join(lines), encoding="utf-8")
+          print(f"::notice::Extracted previous review (score={quality}) to /tmp/anyplot-prev-review.md")
+          PY
+
       - name: Ensure implementation directories exist
         env:
           SPEC_ID: ${{ steps.inputs.outputs.specification_id }}

prompts/workflow-prompts/impl-generate-claude.md

@@ -23,9 +23,20 @@ Read these files to understand the requirements:
 3. `prompts/library/{LIBRARY}.md` - Library-specific rules + theme-adaptive chrome mapping for this library
 4. `plots/{SPEC_ID}/specification.md` - What to visualize
 
-Optional (if regenerating):
-- `plots/{SPEC_ID}/metadata/{LANGUAGE}/{LIBRARY}.yaml` - Previous review feedback
-- `plots/{SPEC_ID}/implementations/{LANGUAGE}/{LIBRARY}.py` - Previous implementation
+### If regenerating (`IS_REGENERATION=true`) — MANDATORY
+
+When regenerating an existing implementation, you MUST read these BEFORE writing any code:
+
+1. `/tmp/anyplot-prev-review.md` — structured review from the previous attempt (image description, strengths, weaknesses, failed criteria checklist). The workflow extracts this automatically from the previous `metadata/{LANGUAGE}/{LIBRARY}.yaml`.
+2. `plots/{SPEC_ID}/implementations/{LANGUAGE}/{LIBRARY}.py` — the previous implementation.
+
+**Default regen mindset: incremental improvement, not rewrite.**
+
+- Preserve the bits listed under "Strengths" unchanged.
+- Address every bullet under "Weaknesses" and each ❌ item in the criteria checklist.
+- If a base style rule (palette, theme colors, chrome, etc. from `prompts/default-style-guide.md` or `prompts/library/{LIBRARY}.md`) conflicts with the previous implementation, update the previous code to match — base style wins.
+- Do NOT discard working structure / data generation / layout choices that the previous review did not flag.
+- Your deliverable is a refined version of the previous file, not a fresh rewrite from the spec.
 
 ### Feasibility Check (Static Libraries Only)
 
@@ -104,12 +115,25 @@ git commit -m "feat({LIBRARY}): implement {SPEC_ID}"
 git push -u origin implementation/{SPEC_ID}/{LIBRARY}
 ```
 
+If `IS_REGENERATION=true`, use an expanded commit body that names what you addressed. Example:
+
+```
+feat(matplotlib): implement scatter-basic
+
+Regen from quality 78. Addressed:
+- text legibility on dark background
+- grid contrast (VQ-03 failed → fixed)
+```
+
+Pass the multi-line message via `-F -` or a heredoc so git preserves the body.
+
 ## Final Check
 
 Before finishing, confirm:
 1. ✅ `plots/{SPEC_ID}/implementations/{LANGUAGE}/{LIBRARY}.py` exists
 2. ✅ `plot-light.png` AND `plot-dark.png` were generated successfully (plus `plot-light.html` / `plot-dark.html` for interactive libs)
 3. ✅ First categorical series renders in `#009E73` in both themes
 4. ✅ Changes were committed and pushed
+5. ✅ If regenerating: `/tmp/anyplot-prev-review.md` and the previous `.py` were read, and each weakness / failed criterion was either addressed or consciously kept (explained in the commit body)
 
 If any of these failed, DO NOT report success.