Mission Control — AI Agent Task Tracking System

A task tracking and plan confirmation system for AI agents. Implements a 5-stage workflow with 3-iteration Qiushi (求是) analysis for every task before execution.

Core Philosophy

• Plan-First: All code/file/agent tasks require Boss confirmation before execution
• Qiushi Iteration: Every plan goes through 3求是→analysis→revision cycles
• Single-Agent Execution: Handled entirely by the main agent (Moss), no sub-agents

Workflow

Stage 1: Task Reception & Triage
Stage 2: Task Definition + Plan Generation (3 iterations)
Stage 3: Boss Confirmation
Stage 4: Execution (with live logging)
Stage 5: Report & Delivery

Auto-Triage Criteria

Automatically determine if a task needs mission-control:

• File operations (.py/.sh/.js/.md/.yaml/.json)? → Yes
• Code/script generation or modification? → Yes
• SubAgent delegation? → Yes
• Web research? → Yes
• Terminal/CLI operations? → Yes
• Pure Q&A (<1min)? → No

File Structure

mission-control/
├── SKILL.md                          # Main skill documentation
├── templates/
│   └── project-intake.md             # Project intake template (Boss fills)
├── references/
│   ├── 2026-05-19-static-spec-conversion.md  # Conversion case study
│   └── project-intake-case-studies.md        # Template usage examples
└── scripts/
    ├── generate_report.py
    ├── next_number.py
    ├── save_plan.py
    └── update_progress.py

Key Concepts

Task ID Format

T-YYYYMMDD-NNN — e.g. T-20260519-001

Task Lifecycle Files

• t-requirement.md — Task requirements
• t-plan.md — Plan with 3-iteration Qiushi analysis
• t-log.md — Execution log (step-by-step status)
• t-report.md — Completion report

Project Intake

When Boss fills project-intake.md before task start, Moss reads it and skips straight to plan generation — no repeated confirmation needed.

Qiushi Integration

Stage	Skill Used
Plan generation (Stage 2)	qiushi — 3 iterations required
Report (Stage 5)	qiushi — criticism & self-review
Other stages	mission-control itself

Prohibited Behaviors

• ❌ Skip mission-control without objective triage
• ❌ Start execution without Boss confirmation
• ❌ Skip the 3-iteration Qiushi analysis
• ❌ Execute without logging to t-log.md
• ❌ End task without t-report.md

Citation

Designed and published by Huang Kai (huangjk8023@yeah.net).

Built for the Hermes Agent system. Part of the open-source AI agent workflow toolkit.