Daily Workflow

daily-workflow

Orchestrates the full content intelligence pipeline by discovering pending items, dispatching parallel worker subagents, merging AI suggestions, and triggering daily distillation and review.

Prerequisites: gws CLI, agent-browser, yt2doc. Refer to ../gws-shared/SKILL.md for auth. Dispatch Spec & Schemas: See dispatch_spec.md for classification rules, prompt templates, and output formats.

Orchestration Pipeline

[1. Discover & Classify] ──> [2. Pre-create Directories] ──> [3. Parallel Dispatch]
                                                                     │
[6. Review Suggestions] <── [5. Distill Knowledge] <── [4. Collect & Merge]

Step 1 — Discover and Classify Items

1. Discover Google Task Delegate list:

   gws tasks tasklists list
   

   Extract DELEGATE_LIST_ID for list matching title == "Delegate". List pending tasks:

   gws tasks tasks list --params '{"tasklist": "<DELEGATE_LIST_ID>", "showCompleted": false, "maxResults": 100}'
   

   Classify each task into threads_queue, youtube_queue, or website_queue based on URL matching rules in dispatch_spec.md.

2. Discover unread newsletters via Gmail API:

   gws gmail users messages list --params '{"userId": "me", "q": "label:newsletter is:unread", "maxResults": 1}'
   

   Check resultSizeEstimate for unread newsletter count.

3. Early exit: If all queues are empty and no unread newsletters exist, log "No content to process today." and exit.

Step 2 — Pre-create Directories

Create date-stamped output directories for non-empty queues and the suggestion staging area:

mkdir -p data/suggestions_pending
# Pre-create reports/<Type>_YYYY_MM_DD/ for active queues only (Newsletter, Threads, Website, YouTube)

Step 3 — Dispatch Parallel Subagents

Dispatch all content items concurrently in fire-and-forget mode. Do not block on individual completions. Invocations use path-free declarative sub-agent personas, passing runtime parameters in the first user message.

• Newsletters: Fetch email IDs in batches of 10 (q: "label:newsletter is:unread"). Spawn subagent newsletter_worker for each MESSAGE_ID.
• Threads / Website / YouTube: For each item in threads_queue, website_queue, and youtube_queue, spawn subagent threads_worker, website_worker, or youtube_worker.

Refer to dispatch_spec.md for prompt parameter templates and tracking specs.

Step 4 — Collect Results & Merge Suggestions

1. Synchronization Barrier: Wait for all dispatched worker subagents to report completion (or failure). Proceed to grading and merging as soon as all worker subagents finish.

2. Grade & Merge Suggestions: Run subagent rubric_grader to score each file matching data/suggestions_pending/suggestion_*.json:

   • Approved suggestions score $\ge 4/6$ and append to data/suggestions_pending.md.
   • Vetoed or low-scoring suggestions append to data/suggestions_filtered.md.
   • Remove processed pending JSON files.

Step 5 — Distill Knowledge

Invoke subagent distiller_reviewer (or follow ../daily-distiller/SKILL.md) to synthesize today's reports into reports/distillations/.

Step 6 — Review Suggestions

Invoke subagent distiller_reviewer (or follow ../review-suggestions/SKILL.md) to conduct interactive suggestion review and calibrate user preferences.

Step 7 — Final Summary

Print final summary adhering to the output format in dispatch_spec.md.