Type-Based Autonomy

Type-Based Autonomy

Transform your agent from reactive to autonomous worker on specific task types.

Concept

The agent pulls from tasks/QUEUE.md but only works on tasks tagged with specific types: 📚 Research → ✅ Autonomy works on these ✍️ Writing → ✅ Autonomy works on these 🔍 Analysis → ✅ Autonomy works on these 🧹 Maintenance → ❌ Autonomy SKIPS these (cron handles) 💾 Backup → ❌ Autonomy SKIPS these (cron handles) Cron jobs handle backups, cleanup, security audits. Autonomy handles research, writing, analysis.

1. Task Queue Structure

• Every task in tasks/QUEUE.md has a @type: label:
• ## 🔴 Ready
• ### 📚 Research (@type:research)
• [ ] @priority:high @type:research Competitor pricing for X product
• [ ] @priority:medium @type:research Ollama model alternatives
• ### ✍️ Writing (@type:writing)
• [ ] @priority:medium @type:writing Blog post on memory systems
• [ ] @priority:low @type:writing Documentation update
• ### 🔍 Analysis (@type:analysis)
• [ ] @priority:medium @type:analysis Review weekly metrics
• [ ] @priority:low @type:analysis Analyze token patterns
• ### 🧹 Maintenance (@type:maintenance)
• → Autonomy IGNORES, cron handles
• [ ] @priority:medium @type:maintenance Old log cleanup

2. Heartbeat Flow

Heartbeat → Check urgent → No → Read QUEUE.md → Filter by @type → Pick task → Work → Update QUEUE → Log Filter logic: Read all tasks in 🔴 Ready section ONLY pick tasks with @type:research | @type:writing | @type:analysis SKIP tasks with @type:maintenance | @type:backup | @type:security

3. Task Completion

1. Mark task as In Progress: @agent: @type:research [task description] 2. Work on it 3. Move to Done Today with completion notes 4. Log to memory/[today].md 5. Check GOALS.md and .learnings/ for follow-up tasks

Research (@type:research)

• Definition: Information gathering, investigation, discovery
• Examples:
• Competitor analysis
• API documentation research
• Technology exploration
• Market research
• Best practices investigation
• Output format:
• ## Research: [Topic]
• ### Findings
• Key point 1
• Key point 2
• ### Sources
• [Source 1](url)
• [Source 2](url)
• ### Recommendations
• Recommendation 1
• Recommendation 2
• Add follow-up task to Ideas:
• [Idea: @type:analysis @priority:medium Analyze research findings for X]

Writing (@type:writing)

Definition: Content creation, documentation, communication Examples: Blog posts Documentation updates Email drafts Announcements Guides/tutorials Output format: # [Title] [Content] Post-completion: If email: Add to Ideas for review by RA If blog/doc: Add to Ideas for publishing If announcement: Add to Ideas for distribution

Analysis (@type:analysis)

• Definition: Data review, metrics analysis, pattern finding
• Examples:
• Weekly performance review
• Token usage analysis
• Log analysis
• Trend identification
• Metrics dashboard creation
• Output format:
• ## Analysis: [Topic]
• ### Data Reviewed
• [List of data sources]
• ### Key Findings
• Finding 1 with metric
• Finding 2 with metric
• ### Patterns
• Pattern 1
• Pattern 2
• ### Recommendations
• Action 1
• Action 2
• Add follow-up tasks:
• [Idea: @type:writing @priority:medium Write analysis report]
• [Idea: @type:research @priority:low Investigate pattern X further]

Maintenance (@type:maintenance)

Definition: System cleanup, organization, routine tasks Handler: Cron (NOT autonomy) Examples: Old log cleanup Temp file deletion File organization Archive old records Behavior: Autonomy SKIPS these tasks Cron jobs handle them overnight Manual trigger if urgent, but usually not needed

Backup (@type:backup)

Definition: Data backup, version control, sync Handler: Cron (NOT autonomy) Examples: GitHub backup Database backup Cloud sync Behavior: Autonomy SKIPS these tasks Scheduled twice daily (00:00, 12:00 UTC)

Security (@type:security)

Definition: Security checks, audits, vulnerability scans Handler: Cron (NOT autonomy) Examples: Security audit Permission check Credential review Behavior: Autonomy SKIPS these tasks Monthly security audit cron (1st of month)

Priority System

Priority affects task selection order: PriorityWhen to useSelection@priority:urgentTime-sensitive, deadline < 24hPick FIRST@priority:highImportant, deadline 2-3 daysPick SECONDpriority:mediumNormal importancePick THIRDpriority:lowNice to have, no deadlinePick LAST

GOALS.md Integration

• Every task should support RA's long-term goal: MONEY
• When creating tasks:
• Check GOALS.md for current objectives
• Link tasks to money-making activities
• Ask: "How does this help RA make money?"
• Examples:
• ### 📚 Research
• [ ] @priority:high @type:research Competitor pricing (GOAL: pricing strategy for new product)
• [ ] @priority:medium @type:research Market fit analysis (GOAL: validate product idea)
• ### ✍️ Writing
• [ ] @priority:high @type:writing Sales email template (GOAL: improve conversion)
• [ ] @priority:medium @type:blog Marketing post (GOAL: drive traffic)
• Post-completion:
• Update GOALS.md with progress notes if relevant
• Add to Ideas: follow-up tasks that advance goals

.learnings/ Integration

When completing tasks, add findings to .learnings/: After research task: ## [LRN-20260216-001] research-findings Research: Competitor pricing analysis Key findings: [summarize] After analysis task: ## [LRN-20260216-002] analysis-insights Analysis: Token usage patterns Key insights: [summarize] After problem: ## [ERR-20260216-001] research-issue Error: API rate limit during research Fix: [document the fix]

Adding Tasks

• RA adds tasks directly to queue:
• ## 🔴 Ready
• [ ] @type:research @priority:high Analyze competitor X pricing
• The agent discovers tasks during work and adds to Ideas:
• ## 💡 Ideas
• [Idea: @type:research @priority:medium Investigate Ollama alternative models]

Updating Tasks

• When starting:
• ## 🟡 In Progress
• [ ] @agent: @type:research @priority:high Competitor pricing analysis
• - Started: 2026-02-16 14:00 UTC
• - Progress: Gathering data
• When complete:
• ## ✅ Done Today
• [x] @agent: @type:research @priority:high Competitor pricing analysis
• - Completed: 2026-02-16 14:25 UTC
• - Output: tasks/outputs/competitor-pricing-analysis.md
• When blocked:
• ## 🔵 Blocked
• [ ] @type:writing @priority:medium Email draft (needs: RA review)

Clearing Done Today

Daily routine (during heartbeat or cron): Move completed tasks from Done Today to tasks/archive/ if you want to keep history Or simply delete if not needed Clear section with: ## ✅ Done Today (empty)

Token Budget

Recommendation: 4 sessions/day, ~3-8K tokens each = 12-32K/day Session strategy: TimeTask TypeTokensFocus09:00 AMResearch8-10KDeep investigation13:00 PMWriting5-7KContent creation17:00 PMAnalysis3-5KData review21:00 PMLearning2-3KReview learnings When to stop: If tokens remaining < 5K for the day If queue has no tasks of allowed types If RA is actively messaging (human priority)

Coordination With Cron

Autonomy and cron work in parallel but on different task types: SystemTask TypesScheduleAutonomyResearch, Writing, AnalysisAny time (heartbeat-driven)CronBackup, Maintenance, SecurityScheduled (midnight, noon, daily) No conflicts - they work on completely different things.

When to Use This Skill

Use this skill when: You want autonomy focused on value-add tasks (research, writing, analysis) You want maximum token efficiency Tasks can be clearly categorized by type You want to expand task types later (add coding, testing as they come up) You want clear separation between autonomy and maintenance

When NOT to Use This Skill

Do not use this skill when: You want continuous work on any task type → Use autonomy-windowed skill Task types are unclear or ambiguous → Use simpler queue You want the agent to handle maintenance tasks → Those belong to cron You want predictable work hours → Use autonomy-windowed for time-based control

Quick Reference

Task types for autonomy: @type:research | @type:writing | @type:analysis Task types for cron: @type:maintenance | @type:backup | @type:security Priority order: urgent → high → medium → low Queue location: tasks/QUEUE.md See templates/QUEUE.md for full template structure