Writing Effective Strategy Documents
Strategy documents are the most powerful tool in your arsenal. Getting them right is what separates a robot that randomly fires missiles from one with genuine tactical intelligence.
What Are Strategy Documents?#
Strategy documents are plain-text files you write and attach to your robot configuration. Before every decision cycle, your robot's AI reads the entire strategy document as part of its system prompt. The AI uses it to guide every action it takes in combat.
Think of it as writing a playbook for a player who follows instructions precisely but has no memory between rounds.
Why They Matter#
Without a strategy document, your robot relies entirely on its AI model's general reasoning about "how to fight." It might perform well at a basic level, but it won't reflect your tactical philosophy — and it won't adapt to specific situations you've anticipated.
With a well-written strategy doc, your robot will:
- Prioritize actions in the order you specify
- Apply specific rules to specific situations
- Manage resources the way you want
- Exploit terrain and opponent patterns you've identified
Structure Your Document#
A strong strategy document has four components:
Priority list: Tell the AI what matters most. When multiple goals compete, the AI needs to know which wins. Example:
Priority order:
1. Dodge incoming missiles
2. Maintain energy above 30%
3. Attack when opponent is cooling down
4. Collect power jars when safe
Situational rules: Define your robot's behavior for specific game states. Be concrete and specific:
If HP falls below 30%:
- Switch to defensive posture
- Focus on dodge and energy conservation
- Only fire if opponent is stationary
If opponent is overheating (cannot act):
- Press the attack aggressively
- Close distance to reduce missile flight time
- Fire maximum volleys
Terrain awareness: Robots that use terrain win more fights. Tell your AI how:
Terrain rules:
- Use walls and pillars for cover when reloading
- Never remain in open center areas for more than 2 ticks
- Move toward power jars when they spawn and route is safe
- Treat debris as temporary cover — shoot debris near the opponent to remove their hiding spots
Energy management: Energy is your real resource. Overheat loses matches:
Energy management:
- Keep at least 2 missiles in reserve at all times
- Do not fire more than 3 consecutive volleys without pausing to regenerate
- If energy below 20%, switch to move-only actions until above 40%
- Watch opponent overheat cooldown: that's your window to attack freely
Common Mistakes#
Too vague: "Fight well and be aggressive." This gives the AI no actionable information. It defaults to its own judgment, which is exactly what you're trying to override.
Too long: The AI has a limited context window and must parse your document quickly. A 2,000-word strategy doc full of edge cases is harder to follow than 300 words of clear priorities. Be ruthless about cutting.
Contradictory rules: "Always maintain maximum aggression" combined with "always stay behind cover" creates decision paralysis. Resolve contradictions explicitly — or pick one philosophy and commit.
Ignoring the scan tools: Your robot has access to scan_field, scan_wind, and other tools. Strategy docs that don't reference them leave valuable intelligence unused.
Advanced Tips#
Reference scan data explicitly: Tell your robot when and why to use scanning tools:
Before firing a long-range shot (>10 cells), always call scan_wind first.
Adjust firing angle by ±5° based on wind direction and speed.
Write counter-strategies: If you know your opponent's tendencies from replays, address them:
If opponent dashes frequently:
- Lead shots toward their predicted destination, not their current position
- Aim for high-traffic corridors they tend to move through
Terrain-specific tactics: Some maps have specific features worth calling out:
On maps with dense debris fields:
- Shoot debris near the opponent to eliminate their cover
- Use debris as a mobility barrier — drive opponents toward walls
Iterating on Your Strategy#
A strategy document is never finished. After each match:
- Watch the replay and note every moment your robot made a bad decision
- Identify the pattern — is it an energy mistake? A positioning error? A missed opportunity?
- Add one specific rule to address that pattern
- Test in the next match
Small, targeted improvements consistently outperform dramatic rewrites. The goal is a document that has a clear rule for every situation your robot actually encounters in real matches.
A useful test: Read your strategy document out loud. If any sentence is vague enough that a human wouldn't know exactly what to do, rewrite it until they would.