fix: autocomplete search crashes on ANSI category badges

docs/UIUX_BRANDING_HANDOFF.md

@@ -108,23 +108,19 @@ From `cpa.config.json` or catalog `customOptions`:
 - `text` for string options
 - Password/invisible types are skipped with a yellow warning
 
-### Category badge colors
+### Category badges
 
-Template list badges use deterministic ANSI colors from `_CATEGORY_PALETTE` in `catalog.py`:
+Interactive template choices use a plain fixed-width badge from
+`short_category_label()` (strips "Applications", "Application", "Boilerplate";
+abbreviates long names to initials). Titles stay plain text because
+`questionary.autocomplete` wraps choices in HTML for match highlighting — ANSI
+or other markup raises XML parse errors in prompt_toolkit.
 
-| Index | ANSI color | Typical use |
-|-------|------------|-------------|
-| 0 | yellow (`\033[33m`) | category badge |
-| 1 | green (`\033[32m`) | category badge |
-| 2 | cyan (`\033[36m`) | category badge |
-| 3 | magenta (`\033[35m`) | category badge |
-| 4 | blue (`\033[34m`) | category badge |
+`--list-templates` uses Rich tables for color, not ANSI in choice strings.
 
-Color selection: `sum(ord(char) for char in category_slug) % 5`. Respects `NO_COLOR` (plain text, no ANSI).
-
-Badge label: compact form from `short_category_label()` (strips "Applications", "Application", "Boilerplate"; abbreviates long names to initials).
-
-Design implication: terminal category colors are slug-hash-driven, not semantically mapped (e.g. "backend" is not always blue). A future brand system may want stable category-to-color mapping for docs and website cards.
+Design implication: if terminal category color returns, prefer Rich styling or
+a prompt library that does not HTML-parse choice titles (CNA uses `prompts` +
+picocolors).
 
 ### Rich semantic color usage
 
@@ -195,7 +191,7 @@ The CLI experience is functional and CNA-aligned for catalog flows, but the broa
 Current CLI aesthetic:
 
 - Rich semantic colors (red/yellow/green/cyan/dim).
-- Hash-based category badge colors in terminal.
+- Plain fixed-width category badges in autocomplete titles (HTML-safe for questionary).
 - Minimal hero SVG (slate + teal).
 - No startup banner or branded prompt chrome beyond questionary defaults.
 
@@ -294,7 +290,7 @@ Before implementation, produce a complete audit answering:
 - What should the brand personality become?
 - Does the CLI first run explain the product clearly in the first 5 seconds?
 - Do autocomplete and checkbox flows feel premium and discoverable?
-- Should category badge colors become semantic instead of hash-based?
+- Should website/docs category badges use semantic colors (CLI titles stay plain for questionary)?
 - Does the PyPI package README convert visitors into users?
 - Does the root GitHub README convert visitors into contributors?
 - Are templates presented in a way that feels premium and trustworthy?
@@ -367,7 +363,7 @@ Visual identity:
 - Define illustration/hero style.
 - Define voice and tone (match CLI error copy guidelines).
 - Define how "cozy" and "developer infrastructure" coexist.
-- Map `_CATEGORY_PALETTE` colors to branded hex values for non-terminal surfaces.
+- Define category badge colors for website/docs cards (CLI autocomplete titles stay plain text).
 
 ## Constraints And Standards
 
@@ -410,7 +406,7 @@ Environment variables affecting UX:
 | Variable | Effect |
 |----------|--------|
 | `CI` | Disables interactive prompts unless `--interactive` |
-| `NO_COLOR` | Disables category badge ANSI colors |
+| `NO_COLOR` | Honored by Rich output (list tables, messages) |
 | `CPA_CATALOG_URL` | Override catalog source |
 | `CPA_NO_CATALOG_CACHE` | Force catalog refresh |
 | `CPA_STRICT_VERSION` | Treat version mismatch as error |
@@ -424,7 +420,10 @@ We need to do a full UI/UX and branding review of the Create Python App ecosyste
 
 Please start with discovery and audit before implementing. Review the root create-python-app repo, the package README, cpa-templates, and docs/BRAND.md. The goal is to improve engagement, attraction, branding, cozy developer experience, visual consistency, and conversion across GitHub, PyPI, docs, CLI, and generated starter UIs.
 
-The CLI already uses Rich on stderr, questionary autocomplete/checkbox flows, and hash-based category badge colors from catalog.py. Evaluate whether those defaults should evolve into a cohesive brand system. Pay attention to error message tone, interactive vs CI non-interactive behavior, and the minimal hero SVG.
+The CLI already uses Rich on stderr and questionary autocomplete/checkbox flows
+with plain-text category badges (HTML-safe for prompt_toolkit). Evaluate whether
+those defaults should evolve into a cohesive brand system. Pay attention to error
+message tone, interactive vs CI non-interactive behavior, and the minimal hero SVG.
 
 Previous work established basic READMEs and BRAND.md notes, but now I want a broader review and a stronger cohesive brand direction. Do not assume the current teal-on-slate hero or terminal colors are final.
 

packages/create-awesome-python-app/src/create_awesome_python_app/catalog.py

@@ -20,14 +20,6 @@
 console = Console(stderr=True)
 
 CUSTOM_TEMPLATE_SENTINEL = "__custom_template__"
-_ANSI_RESET = "\033[0m"
-_CATEGORY_PALETTE = (
- "\033[33m", # yellow
- "\033[32m", # green
- "\033[36m", # cyan
- "\033[35m", # magenta
- "\033[34m", # blue
-)
 
 
 @dataclass(frozen=True)
@@ -172,13 +164,6 @@ def short_category_label(category_name: str) -> str:
  return " ".join(words[:2]) or category_name
 
 
-def _color_category(slug: str, label: str) -> str:
- if os.environ.get("NO_COLOR"):
- return label
- idx = sum(ord(char) for char in slug) % len(_CATEGORY_PALETTE)
- return f"{_CATEGORY_PALETTE[idx]}{label}{_ANSI_RESET}"
-
-
 def _category_map(data: dict[str, Any]) -> dict[str, str]:
  return {
  str(category.get("slug", "")): str(category.get("name", ""))
@@ -256,8 +241,10 @@ def build_template_choices(data: dict[str, Any]) -> list[TemplateChoice]:
  label_suffix = " · " + ", ".join(str(label) for label in labels[:3])
  description = str(template.get("description", "")).strip()
  description_suffix = f" — {description}" if description else ""
+ # Plain text only: questionary.autocomplete wraps choices in HTML for
+ # match highlighting, so ANSI / markup here raises XML parse errors.
  title = (
- f"{_color_category(category_slug, badge)} "
+ f"{badge} "
  f"{template.get('name', slug)} ({slug})"
  f"{label_suffix}{description_suffix}"
  )

packages/create-awesome-python-app/tests/test_catalog_resolve.py

@@ -102,12 +102,37 @@ def test_build_template_choices_are_searchable() -> None:
  assert "FastAPI Starter" in first.title
  assert "OpenAPI" in first.title
  assert "uv" in first.title
+ assert "\033" not in first.title
  assert "openapi" in first.search
  assert "backend" in first.search
  assert "uv" in first.search
  assert choices[-1].value == CUSTOM_TEMPLATE_SENTINEL
 
 
+def test_template_choice_titles_are_html_safe_for_questionary() -> None:
+ """questionary.autocomplete formats choice text as HTML (match underline)."""
+ from prompt_toolkit.formatted_text import HTML
+
+ catalog = {
+ "categories": [
+ {"slug": "backend-applications", "name": "Backend Applications"}
+ ],
+ "templates": [
+ {
+ "slug": "fastapi-starter",
+ "name": "FastAPI Starter",
+ "description": "Async API with OpenAPI docs",
+ "url": "file:///templates/fastapi",
+ "category": "backend-applications",
+ "labels": ["FastAPI"],
+ }
+ ],
+ }
+ title = build_template_choices(catalog)[0].title
+ # Must not raise "not well-formed (invalid token)" from ANSI escapes.
+ HTML("{}<b><u>{}</u></b>{}").format(title[:3], title[3:6], title[6:])
+
+
 def test_build_extension_choices_filters_by_template_type() -> None:
  catalog = {
  "categories": [