Refined capture pipeline and parser debugging plan

2 min read

Refined capture pipeline and parser debugging plan

  • Day: 2026-06-22
  • Time: 12:00 to 12:10
  • Project: Dev
  • Workspace: WP 2: Operational
  • Status: In Progress
  • Priority: MEDIUM
  • Assignee: Matías Nehuen Iglesias
  • Tags: Csv-Parsing, Firefox, Mediarecorder, Capture-Lifecycle, Office-Window, Office-Auto-Lab

Description

Session Goal

Clarify and de-risk several implementation paths across the office automation stack, with emphasis on capture lifecycle ownership, UI boundaries, and two concrete debugging issues: CSV multiline parsing in /queues and Firefox voice recorder failures.

Key Activities

  • Reviewed the likely cause of a CSV UI bug where multiline quoted fields were being split incorrectly by a custom line-based parser.
  • Proposed validating the issue with Python and replacing the homegrown parser with [[csv]]-parse/sync to preserve quoted multiline rows.
  • Added a defensive UI alert recommendation so malformed or suspicious rows surface visibly instead of silently corrupting project_id display.
  • Diagnosed the Firefox recorder issue by outlining a fallback-first debugging plan: verify MediaRecorder MIME support, microphone permissions, secure-context requirements, and visible error handling in the client UI.
  • Defined a manual test path to confirm capture, recording, and upload behavior end to end.
  • Revisited the architecture for office automation, recommending that office-auto-lab become the canonical owner of capture lifecycle merging and artifact generation, while office-window remains a compiled lifecycle viewer.
  • Clarified roadmap sequencing: compile raw and derived capture events into canonical lifecycle artifacts before adding transcription, routing, approval, or reingest flows.
  • Specified a UI-only PR boundary for office-window to prefer compiled capture_lifecycle artifacts when available, while preserving raw inbox fallback and audio playback.
  • Outlined a clean split for PRs 4-6 where office-auto-lab owns transcription, routing, artifactization, and candidate compilation, and office-window only observes derived surfaces.

Achievements

  • Identified the most probable root cause of the CSV display bug and selected a safer parser strategy.
  • Established a practical Firefox debugging checklist for the recorder, including MIME fallback and explicit error surfacing.
  • Tightened the architectural boundary between capture processing and UI observation, reducing ambiguity about which system owns lifecycle compilation.
  • Produced a phased implementation direction that can guide future PR sequencing.

Pending Tasks

  • Confirm the CSV bug with a reproducible test case and migrate the parser implementation.
  • Implement the UI warning/alert for malformed CSV rows.
  • Verify Firefox recorder behavior across MIME types and permission states, then patch the client component accordingly.
  • Build the canonical capture lifecycle compiler in office-auto-lab and wire office-window to consume compiled artifacts with fallback behavior.

Evidence

  • source_file=2026-06-22.sessions.jsonl, line_number=3, event_count=0, session_id=d47d98534cf8aa685939a6155b77711561c6894e2b74414189d47d5b7a0c83f0
  • event_ids: []

Graph View

Backlinks