An Emergent Mission Architecture for Screeps

An Emergent Mission Architecture for Screeps

How I structure my bot to handle strategy, warfare, and large-scale decisions without drowning in micro-management.

---

Why I Rewrote My Bot

Like many Screeps players, my first bot grew organically: creeps made local decisions, special cases piled up, and every new feature felt harder than the last. Combat logic was especially painful—tactical decisions leaked everywhere, and it was difficult to answer simple questions like:

Is this war actually working?

When I rewrote my bot, I committed to two guiding ideas:

1. Top-down control where it matters Creeps shouldn’t decide why they act. Higher-level systems should.
2. Natural task decomposition Large goals should break down into smaller ones without relying on a rigid planner.

The result is a mission-based architecture, where Managers act as the brains of the empire and Missions act as the specialized tools they deploy.

---

High-Level Structure: Managers as the Highest Level

In this architecture, Managers are the highest-level actors. They are persistent, “always-on” entities that bridge the gap between global state and specific action.

Managers serve two primary functions:

1. Maintenance: They handle repetitive, baseline responsibilities (e.g., the Room Manager ensures the controller doesn’t downgrade).
2. Deployment: They act as the “commanders” who decide when the situation requires a specialized Mission.

Manager	Repetitive Responsibility	Mission Deployment Example
Room Manager	Local infrastructure, mining & upgrading	Remote Defense Mission
Combat Manager	Global threat assessment	Total War Mission
Expansion Manager	Room scouting & planning	Claim Mission

---

Missions: Goal-Oriented Execution

While Managers are the “who,” Missions are the “what.” A mission represents a concrete objective with a clear success or failure condition.

A useful mental model is chess:

• The Manager is the chess player.
• The Mission is the specific opening or strategy being executed.
• Creeps are the pieces.

Recursive Decomposition (The Mission Chain)

Missions are most powerful when they spawn sub-missions. This allows for complex behavior to emerge from simple, nested logic. A high-level mission defines the intent, while its children handle the tactics.

Example Mission Chain: Total War -> Siege -> Quad

1. Total War (deployed by Combat Manager): Tracks the overall economic drain on the enemy and decides which rooms to pressure.
2. Siege: Focuses on a specific room, managing staging points and wall-breaking progress.
3. Quad: A tactical unit mission that spawns and handles four creeps in formation to execute the siege’s goals.

---

How Missions Communicate

All coordination happens through mission memory. The guiding rule is simple:

A mission may read any mission’s memory, but may only write to its own.

Example Decision Flow

1. A Quad Mission loses a creep and records the loss in its memory.
2. A Siege Mission observes the creep loss flag in its child Quad’s memory and adapts future squad composition.
3. Continued poor results cause the siege to stagnate.
4. The Total War Mission observes the lack of progress across multiple Sieges and decides to pivot the empire’s resources elsewhere.

No explicit reporting or signaling is required—coordination emerges from shared observation of state.

---

Mission Memory Structure

All missions are stored in a double-layered structure: Memory.missions[type][id]. This allows for fast lookups and easy iteration.

Example

Memory.missions["siege"]["W1N1"] = {
  type: "siege",
  id: "W1N1",
  targetRoomName: "W1N1",
  childMissions: [
    { type: "quad", id: "W1N1_17253400" }
  ],
  status: {
    isEffective: false,
    netDamage: -1500
  }
}

---

The Execution Loop

The daily life of the bot follows a predictable cycle, moving from high-level observation down to individual action:

1. Analyze: Managers maintain the baseline bot functions and scan the world state to identify strategic needs.
2. Deploy: Based on the analysis, Managers spawn or update Missions. For example, the Combat Manager may deploy a Total War mission if a player is being mean. The Total War mission then recursively deploys several Siege missions toward that player’s rooms.
3. Resolve: High-level missions read the memory of their child missions to evaluate progress. For example, a Siege mission checks its Quad mission’s results to see if the room’s defenses are cracking.
4. Act: Missions update their own state and issue direct orders to their assigned Creeps. In this phase, a Quad mission control creeps and reports how much damage it is dealing, feeding the loop for the next decision making.

This sequence ensures that execution remains local, while evaluation remains strategic.