generate-digest

name	generate-digest
description	Analyzes workspace state and generates actionable insights
feature_module	notes
trigger	scheduled

Generate Digest Skill

Analyze workspace activity and generate categorized, actionable suggestions for the AI Digest panel.

Quick Start

Use this skill when:

• Hourly background job triggers digest generation
• User manually refreshes the digest panel
• System detects significant workspace state changes

Workflow

1. Receive Workspace Context

   • Recent issues (last 7 days): state, assignee, priority, activity
   • Recent notes (last 7 days): annotations, extraction status
   • Active cycles: progress metrics, burndown
   • Issue links and dependencies

2. Analyze for Actionable Insights Categories to check:

   • stale_issues: Issues not updated in 3+ days while in active state
   • unlinked_notes: Notes with no linked issues (potential extraction candidates)
   • review_needed: Issues in review state for 2+ days
   • blocked_dependencies: Issues blocked by unresolved dependencies
   • cycle_risk: Cycles at risk of missing targets (>70% time, <50% done)
   • overdue_items: Issues past target date
   • unassigned_priority: High/urgent priority issues with no assignee
   • duplicate_candidates: Issues with high semantic similarity
   • stale_prs: PRs open for 3+ days without review
   • annotation_pending: Notes with unaddressed AI annotations
   • inactive_members: Team members with no activity in 5+ days
   • documentation_gaps: Features without linked documentation notes

3. Score Relevance For each suggestion, score 0.0-1.0 based on:

   • Ownership: Is the user the owner/assignee? (+0.3)
   • Recency: How recently did the user interact with related items? (+0.2)
   • Priority: Is the related item high/urgent priority? (+0.2)
   • Impact: Does this affect active cycles or deadlines? (+0.3)

4. Return Structured Output

Output Format

{
  "suggestions": [
    {
      "id": "uuid",
      "category": "stale_issues",
      "title": "3 issues haven't been updated this week",
      "description": "PS-42, PS-55, PS-61 are in progress but have no activity since Monday",
      "entity_id": "uuid",
      "entity_type": "issue",
      "action_url": "/workspace/issues?filter=stale",
      "relevance_score": 0.85
    }
  ],
  "summary": "Generated 8 suggestions across 5 categories",
  "tokens_used": {
    "input": 950,
    "output": 420
  }
}

Constraints

• Context must stay under 4000 tokens
• Maximum 20 suggestions per digest
• Each suggestion must have a unique (entity_id, category) pair
• Relevance scores must be between 0.0 and 1.0

Integration Points

• DigestContextBuilder: Builds SQL aggregate context (<4000 tokens)
• DigestJobHandler: Orchestrates the job, stores results
• WorkspaceDigest model: Stores suggestions as JSONB
• Queue: LOW priority (ai_low), 60s timeout, 3 retries

References

• Spec: specs/012-homepage-note/spec.md (Background Job Specification)
• US-19: Homepage Hub feature