GTD Runner Skill

Autonomous task execution from GTD.md items.

Trigger

• Command: /cyber-gtd
• Default: plan-first (show before execute)

Workflow

1. Read @GTD.md → extract items from # Next
2. For each item:
   • Classify → match pattern to workflow
   • Lookup entities via database: bun scripts/db/query.ts find-entity "<name>"
   • Plan actions
3. Show plan → ask to execute
4. Execute approved items sequentially (one at a time)
5. Output: ~/CybosVault/private/content/work/MMDD-<slug>.md

Classification → Routing

Entity Lookup

Query database for entity context:

1. Parse item for names (people, companies)
2. Run: bun scripts/db/query.ts find-entity "<name>" --json
3. If found, get full context: bun scripts/db/query.ts entity <slug> --json
4. Entity context includes:
   • Recent interactions (calls, emails, telegram)
   • Deal associations
   • Pending items
5. If not found + confident it's an entity:
   • High confidence (company with domain) → auto-create stub
   • Medium confidence → ask: "Create entity file for X?"

Output Format

All outputs to ~/CybosVault/private/content/work/MMDD-<slug>.md:

markdown
1# Task: [Task Description]
2
3**Status:** Pending Approval | Completed | Incomplete
4**Created:** YYYY-MM-DD HH:MM
5**GTD Item:** [Original text from GTD.md]
6**Workflow:** [Which workflow handled this]
7
8---
9
10## Context
11
12**Entity:** [Name]
13- Type: [person/org]
14- Deal: [link if exists]
15- Previous calls: [N calls found]
16
17**Key Info:**
18[Relevant context from calls, deals, entity files]
19
20---
21
22## Draft
23
24[Message/agenda/questions/etc]
25
26---
27
28## Pending Actions
29
30- [ ] Send via Gmail to email@example.com
31- [ ] Alternative: Telegram @handle
32
33---
34
35## Execution Log
36
37- HH:MM - [action taken]

Staged Execution

Agent completes what it can autonomously:

• Research and context gathering
• Draft creation
• Preparation work

Then queues actions requiring approval:

• Sending messages (Gmail, Telegram)
• Scheduling meetings
• Any external action

Suggestions

When processing tasks, if you notice 3+ similar patterns that don't have a dedicated workflow, suggest: "I've seen '[pattern]' multiple times. Want me to create a workflow for it?"

Self-Improvement

Log all task executions to @learnings.md for pattern analysis.