Root cause: kb_workflow.py contained 312 lines of deprecated legacy
code that was confusing AI agents. The legacy functions have been
completely replaced by the kb_pipeline package.
Changes:
- Removed all legacy functions (is_video_content, check_duplicate, etc.)
- Kept only process_command() and main() entry points
- Updated SKILL.md to remove old command examples
- File reduced from 495 to 143 lines (71% reduction)
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Zhihu HTML structure includes noscript fallback images:
<figure><img data-actualsrc=...><noscript><img src=...></noscript></figure>
After fixing lazy-load (replacing src with data-actualsrc), both imgs
have real URLs, causing duplicates. The fix removes noscript tags before
processing img elements.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Add _normalize_metadata() to flatten nested metadata structure
- Fix generate_frontmatter() to use correct metadata fields
- Change _sanitize_filename() to return standardized "bilibili_{bvid}" format
Fixes: analysis.md frontmatter showing "未命名"/0/"未知" instead of actual values
Fixes: audio filename using video title instead of platform_bvid format
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
All fixes verified working:
- Video download for videos without official subtitles
- Whisper transcription with correct offset parsing
- Text extraction from Whisper JSON format
- End-to-end analysis with real video content
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Handle Whisper transcription format (array of segments with text)
- Extract full text from transcription array before passing to DeepAnalyzer
- Fix offsets parsing to support both dict and list formats
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Add download_video() to BilibiliAdapter using yt-dlp
- Add bilibili_cookie config to Settings
- Update DownloadProcessor to call video download for bilibili videos
- Add idempotency check to ASRProcessor for existing subtitle files
- Fix indentation error in asr.py
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- When content_type is video and no official subtitle available:
- Download video using yt-dlp
- Extract audio for Whisper fallback
- Update content_file with actual video path
This enables the pipeline to handle videos that don't have
official Bilibili subtitles.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- When content_type is video and no official subtitle available:
- Download video using yt-dlp
- Extract audio for Whisper fallback
- Update content_file with actual video path
This enables the pipeline to handle videos that don't have
official Bilibili subtitles.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- When a step is skipped (already completed), ask user if they want to re-run
- If skipping, scan content_dir for existing files and populate results
- Support both new naming (platform_id_subtitle.json) and old naming (platform_id.json)
- This fixes the issue where analyze step couldn't find subtitle files
when ASR step was skipped
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Add content_dir field to ASRInput schema
- Pass content_dir from pipeline when creating ASRInput
- Modify _try_official_subtitle to use content_dir for output
- Modify _run_whisper_basic to use content_dir for output
- Add validation for empty video_path in Whisper fallback
Root cause: DownloadProcessor sets content_file="" for video types,
causing Path("").parent to resolve to current directory instead of
content directory. Subtitle files were written to wrong location.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Root cause: BilibiliExtractor sync wrappers use asyncio.run() internally,
which fails when called from async context in BilibiliAdapter.
Fix: Call async methods directly instead of sync wrappers.
Verified: Human confirmed fix works (bilibili_BV16cNMzpExh)
Parse duration strings like "12:34" or "01:23:45" into seconds.
Previously only integer duration values were handled.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
BilibiliAdapter._fetch_metadata_with_retry() now calls the underlying
async methods directly instead of sync wrappers that use asyncio.run().
This prevents "asyncio.run() cannot be called from a running event loop"
error when the adapter is called from within an already-running event loop.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Simplify to single-command workflow: kb-workflow process <url>
- Document new CLI parameters and options
- Add architecture overview
- Remove deprecated multi-step manual process
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Archive ROADMAP and REQUIREMENTS to milestones/
- Delete original ROADMAP.md and REQUIREMENTS.md
- Update PROJECT.md with v1.0 validated requirements
- Update STATE.md for milestone complete status
- Update MILESTONES.md with key accomplishments
- Add VERIFICATION.md for phase 04
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Add async run() method to Pipeline class
- Execute processors sequentially with state persistence
- Handle step-specific input types (Parse, Download, ASR, Analyze)
- Support skip_completed for resumable workflows
- Return PipelineOutput with results and content_dir
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Test full workflow execution from URL to analysis
- Test workflow error handling
- Test state persistence
- Test workflow resumability (skip completed steps)
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Add process_command function that executes full pipeline
- Add subparser for 'process' command with URL argument
- Add --output-dir and --skip-completed flags
- Add --no-skip flag to re-run all steps
- Maintain backward compatibility with legacy arguments
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Test process --help shows usage
- Test process <url> calls Pipeline.run()
- Test invalid URL returns error
- Test --output-dir flag
- Test --no-skip flag
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Pipeline class initializes with processors list
- Validates processor STEP_NAME matches config.steps
- _parse_url method parses URLs using LinkParser
- _get_content_dir creates content directory from parsed URL
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Test PipelineInput validates url field (required)
- Test PipelineOutput contains results dict
- Test PipelineConfig has default steps list
- Test PipelineConfig allows custom output_dir
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- AnalyzeProcessor wraps DeepAnalyzer for AI-powered content analysis
- Analysis reports saved as Markdown with YAML frontmatter
- All 15 tests passing
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- AnalyzeProcessor wraps existing DeepAnalyzer for pipeline integration
- Loads metadata and content from JSON files
- Generates analysis reports as Markdown with frontmatter
- Records state on success and failure
- Supports both video and article content types
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- AnalyzeInput extends StepInput with metadata_file, content_file, content_type, output_file
- AnalyzeOutput extends StepOutput with analysis_file, summary, main_topics, key_insights, tags, stats
- Supports both video and article content types
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Test schema validation for AnalyzeInput (metadata_file, content_file, content_type, output_file)
- Test schema validation for AnalyzeOutput (analysis_file, summary, main_topics, key_insights, tags, stats)
- Test AnalyzeProcessor integration with DeepAnalyzer
- Test state recording on success and failure
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
TDD GREEN phase:
- DownloadProcessor downloads article/answer/zhuanlan content
- Video content is skipped (handled by ASR in Phase 3)
- Content files saved with correct naming convention
- DownloadProcessor exported from kb_pipeline.processors
- All 17 tests pass
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
TDD GREEN phase:
- DownloadProcessor inherits from BaseProcessor
- Fetches metadata from Bilibili and Zhihu adapters
- Saves metadata to content_dir as JSON
- Records state on success/failure
- Skips step if already completed (idempotency)
- All metadata tests pass
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
TDD RED phase:
- Test DownloadProcessor fetches Bilibili video metadata correctly
- Test DownloadProcessor fetches Zhihu answer metadata correctly
- Test DownloadProcessor records state on success
- Test DownloadProcessor records state on failure
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
TDD RED phase:
- Test DownloadInput validates required fields (content_dir, content_type)
- Test DownloadOutput validates required fields (metadata_file, content_file, title, author)
- Test DownloadOutput handles both video and article content types
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- BaseProcessor abstract class with unified interface
- ParseProcessor wrapping LinkParser for URL parsing
- ParseInput/ParseOutput schemas for parse step
- All tests passing (18 tests)
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- ParseProcessor wraps LinkParser for URL parsing
- Returns ParseOutput with platform, content_type, content_id
- Records state when state_manager is provided
- Handles unknown URLs with success=False
- Export BaseProcessor, ParseProcessor, ASRProcessor from __init__.py
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- ParseInput extends StepInput with url field
- ParseOutput extends StepOutput with platform, content_type, content_id, resolved_url, original_url
- Tests for schema validation and inheritance
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
starmen1265