fix: correct test assertions for ProjectCodeGenerator return structure

[Claude]

Integration tests were asserting on a non-existent success key in the ProjectCodeGenerator.generate_project() return value. The method actually returns a dict with project_path, project_type, technologies, and features.

Changes

• Test assertions: Updated to verify actual return structure (project_path, project_type) instead of checking for success key
• Exception handling: Changed pytest.raises(Exception) to pytest.raises(ValidationError) for proper Pydantic validation testing
• Code formatting: Applied black formatter to meet style requirements

Verified Behavior

result = await generator.generate_project(video_analysis, config)
# Returns: {
#   "project_path": "/tmp/uvai_project_xyz",
#   "project_type": "react",
#   "technologies": ["react", "tailwindcss"],
#   "features": [...],
#   "framework": "react",
#   "entry_point": "src/App.js",
#   ...
# }

All 32 tests now pass (19 unit + 13 integration).

Original prompt

---

This section details on the original issue you should resolve

<issue_title>🏗️ Stage 2: Semantic Logic Parsing — structured instruction extraction</issue_title>
<issue_description>## Architecture Reference
Stage 2: Semantic Logic Parsing — 'LLMs transform raw video transcripts and visual cues into structured, actionable instructions.'

Current State

The pipeline extracts basic metadata (title, technologies, features list) but does NOT produce structured, actionable instructions. The extracted_info dict contains:

• title (often defaults to generic)
• technologies (list of strings)
• features (list of strings)
• tutorial_steps (raw text, not structured)

Gap

The architecture calls for 'structured, actionable instructions' — step-by-step build plans that Stage 3 (code generation) can follow deterministically. Currently Stage 3 gets loose text and falls back to templates.

Implementation

1. Add a structured output schema for Gemini's video analysis response (JSON mode)
2. Extract ordered build steps with dependencies: 'Step 1: Create React app → Step 2: Install tailwind → Step 3: Create Header component...'
3. Each step should include: action, file path, code snippet (if visible), dependencies
4. Output a BuildPlan JSON artifact that Stage 3 consumes

Acceptance Criteria

• Video analysis returns a structured BuildPlan with ordered steps
• Each step has: action type, target file, code content, prerequisites
• Code generator consumes BuildPlan instead of raw text</issue_description>

Comments on the Issue (you are @claude[agent] in this section)

• Fixes 🏗️ Stage 2: Semantic Logic Parsing — structured instruction extraction #85

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
v0-uvai	Error		Mar 20, 2026 10:04pm

[Copilot]

Pull request overview

Implements Stage 2 “Semantic Logic Parsing” by introducing a structured BuildPlan artifact extracted from Gemini and wiring it into the video-processing + code-generation pipeline, with unit/integration tests.

Changes:

• Added BuildPlan / BuildStep Pydantic models plus a Gemini JSON schema helper.
• Added GeminiVideoService.extract_build_plan() and invoked it from RealVideoProcessor, passing build_plan through to ProjectCodeGenerator.
• Added unit + integration tests for BuildPlan validation and BuildPlan→generation-context behavior.

Reviewed changes

Copilot reviewed 8 out of 8 changed files in this pull request and generated 11 comments.

Show a summary per file

File	Description
tests/unit/test_build_plan.py	New unit tests for BuildPlan models/enums and schema generation.
tests/integration/test_build_plan_integration.py	New integration tests for BuildPlan usage in code generation + context extraction.
src/youtube_extension/backend/services/real_video_processor.py	Calls Gemini to extract BuildPlan during processing and attaches it to results.
src/youtube_extension/backend/models/build_plan.py	New BuildPlan/BuildStep models and Gemini response schema builder.
src/youtube_extension/backend/code_generator.py	Prefers build_plan when building generation context; falls back to legacy extraction.
src/integration/gemini_video.py	Adds extract_build_plan() with responseSchema JSON-mode generation config.
src/youtube_extension.egg-info/top_level.txt	Updates generated packaging metadata (should not be committed).
src/youtube_extension.egg-info/SOURCES.txt	Updates generated packaging metadata (should not be committed).

[Copilot]

If extract_build_plan() raises, gemini_service.close() is skipped, leaving the underlying httpx.AsyncClient unclosed. Wrap the call in a try/finally (or use an async context manager) so close() is always awaited, including on failure.

Suggested change

	build_plan = await gemini_service.extract_build_plan(video_url)
	await gemini_service.close()
	try:
	build_plan = await gemini_service.extract_build_plan(video_url)
	finally:
	await gemini_service.close()

[Copilot]

ProjectCodeGenerator.generate_project() does not include a success key in its return dict (it returns framework/entry_point/etc.), so assert result["success"] is True will raise KeyError. Update the test to assert against fields that are actually returned (e.g., project_path, framework, or generated file contents).

[Copilot]

ProjectCodeGenerator.generate_project() does not return a title, so assert "Todo App..." in result.get("title", "") will always fail. If you want to verify BuildPlan usage, assert via README.md/App.js contents under result["project_path"] or by checking video_analysis["extracted_info"] after context building.

[Copilot]

This test calls generate_project(), which creates a real temp directory via tempfile.mkdtemp() and the test never deletes it. Consider monkeypatching tempfile.mkdtemp to use tmp_path (as tests/test_code_generator.py does) or cleaning up result["project_path"] with shutil.rmtree to avoid leaking temp dirs during CI runs.

[Copilot]

Same issue here: generate_project() does not return a success key, so this assertion will fail with KeyError. Align the assertions with the actual return shape.

[Copilot]

These validation tests use pytest.raises(Exception), which can mask unrelated failures and reduce test signal. Since this is Pydantic validation, assert pytest.raises(ValidationError) (as the unit tests do) so the test only passes for the intended failure mode.

[Copilot]

Same here: catching a broad Exception makes the test pass even if something other than Pydantic validation breaks. Prefer pytest.raises(ValidationError) to keep the assertion precise.

[Copilot]

build_plan_to_gemini_schema() hard-codes the enum values for project_type, difficulty_level, and step action. This can drift from ProjectType/DifficultyLevel/ActionType over time; derive these enum lists from the actual Enum classes to keep the schema and model consistent automatically.

[Copilot]

src/youtube_extension.egg-info/* are build artifacts and are already ignored by .gitignore (*.egg-info/). These diffs indicate regenerated packaging metadata got committed; please remove these files from the PR (and ideally from version control) to avoid churn and accidental inclusion of local environment state.

[Copilot]

This SOURCES.txt change is part of the generated *.egg-info metadata (which is .gitignored). It should not be updated in a feature PR; please revert/remove src/youtube_extension.egg-info changes to prevent noisy diffs and inconsistent packaging manifests.

[github-actions]

🔍 PR Validation

⚠️ Large PR detected (861 lines changed)

In general, the fix is to stop checking for "youtube.com" and "youtu.be" as arbitrary substrings of the full URL string, and instead parse the URL and validate its hostname against an explicit allowlist of hosts (and potentially handle missing schemes robustly). This avoids accidental matches where youtube.com appears in the path, query, or as part of another domain.

Concretely, in GeminiVideoService near line 431, replace:

is_youtube = "youtube.com" in video_url or "youtu.be" in video_url

with logic that:

1. Parses video_url using urllib.parse.urlparse.
2. Handles URLs without scheme by prepending https:// before parsing again.
3. Extracts hostname and normalizes it (lowercase).
4. Checks whether the hostname is in a small allowlist such as {"youtube.com", "www.youtube.com", "m.youtube.com", "youtu.be", "www.youtu.be"} or ends with .youtube.com.

We need to add an import for urlparse (from urllib.parse) at the top of src/integration/gemini_video.py. The rest of the method’s behavior stays the same: if the parsed hostname is a YouTube host, is_youtube is True and the existing request payload for YouTube URLs is used; otherwise, the non-YouTube path runs as before.

In general, to fix "Module is imported more than once" issues, remove duplicate imports and rely on a single, consistent import for each module. Keep imports at the top of the file unless there is a strong, explicit reason to import within a function (e.g., heavy optional dependencies).

For this file, json is already imported at line 15. Inside the method that parses the JSON response (the try block around lines 481–487), there is a second import json. The best fix is to remove this inner import and directly use the already-imported json module. Concretely, in src/integration/gemini_video.py, within the try: block starting around line 481, delete the import json line and leave the build_plan = json.loads(text) and except json.JSONDecodeError lines unchanged. No new imports, methods, or definitions are needed.

In general, to fix an unused local variable, either (a) remove the assignment if the right-hand side has no necessary side effects, or (b) if the variable is intentionally unused, rename it to something like _ or unused_ai_analysis to make that intention explicit.

Here, the variable is assigned as ai_analysis = video_analysis.get("ai_analysis") or {} inside the fallback else block, and there is no subsequent use of ai_analysis in the provided code. The right-hand side is a pure dictionary lookup with a simple or {} default and has no side effects. Therefore, the best fix that preserves behavior is to delete this line entirely. We do not need to keep the expression, and we do not need to introduce an “unused” name, because there is no documentation value in binding it.

Concretely: in src/youtube_extension/backend/code_generator.py, inside _build_generation_context, within the else fallback block that logs “BuildPlan not found, falling back to legacy extraction”, remove the line that assigns ai_analysis. No additional imports, methods, or definitions are needed.