Coding Agent Optimization

agents_md:
  # Codex Agent Instructions
  ## Objective
  Implement Pokemon TCG card effects in Rust by replacing `todo!()` stubs with working code.

  ## Key Patterns
  ### Card Identification
  Use `def_id_matches(&card.def_id, "DF", NUMBER)` to match specific cards:
  Example:
  if def_id_matches(&card.def_id, "DF", 23) {
      // This is Dragon Frontiers card #23
  }

  ### Attack Implementation
  Attacks go in the `attack_override` function. Return damage values directly:
  Example:
  pub fn some_attack_damage() -> i32 { 40 }

  ### Poke-Powers/Bodies
  Implement in `power_effect` function for activated abilities.

  ## Workflow
  1. Read the stub file to identify `todo!()` locations
  2. Edit the file to replace each `todo!()` with implementation
  3. Run `cargo check` to verify it compiles
  4. Run `cargo test` to verify correctness

  ## Common Mistakes to Avoid
  - Do NOT read the file multiple times
  - Do NOT forget to handle coin flips with the `coin: bool` parameter
  - Do NOT return wrong damage types (use i32)

codex_skills:
  skills:
    - name: implement_attack
      description: Implement a Pokemon attack function
      triggers: ["attack", "damage", "todo!()"]
      steps:
        - Read the attack text comment to understand the effect
        - Calculate the damage formula
        - Replace todo!() with the implementation
    - name: implement_power
      description: Implement a Poke-Power or Poke-Body
      triggers: ["power", "body", "ability"]
      steps:
        - Identify the power type (Power vs Body)
        - Implement the effect logic
        - Handle any conditions or triggers
    - name: verify_implementation
      description: Verify the implementation compiles and passes tests
      triggers: ["verify", "check", "test"]
      steps:
        - Run cargo check
        - Run cargo test
        - Fix any errors

system_prompt:
  You are an expert Rust developer implementing Pokemon TCG cards.

  CRITICAL: The stub file contains `todo!()` macros that YOU MUST REPLACE with working code.

  Example - if you see:
  pub fn grind_damage(attached_energy: u32) -> i32 { todo!() }
  // ATTACK_1_TEXT: "Does 10 damage times the amount of Energy attached"

  You MUST use the edit tool to change it to:
  pub fn grind_damage(attached_energy: u32) -> i32 { (10 * attached_energy) as i32 }

  REQUIRED WORKFLOW:
  1. Read the stub file ONCE to find the todo!() functions
  2. IMMEDIATELY use the edit tool to replace todo!() with working code
  3. Run `cargo check` to verify compilation
  4. Run `cargo test` to verify tests pass

  DO NOT read the file multiple times

# Baseline - hand-written instructions
baseline_system_prompt = """You are a Rust developer.
Implement the card effects."""

# After optimization - learned from successful rollouts
optimized_system_prompt = """You are an expert Rust developer implementing Pokemon TCG cards.

CRITICAL: Before writing any code:
1. Read the existing card implementations in src/effects/
2. Note the pattern: match on def_id, return AttackOverrides

Implementation checklist:
- [ ] Identify card by def_id_matches(&card.def_id, "DF", NUMBER)
- [ ] Handle Poke-Powers in power_effect()
- [ ] Handle attacks in attack_override()
- [ ] Use game.queue_prompt() for user choices
- [ ] Return AttackOverrides::default() for non-matching cards

After implementation:
- Run `cargo check` to verify compilation
- Run `cargo test` to verify behavior
- Fix any failing tests before finishing"""

# Environment variables
export OPENAI_API_KEY="sk-..."
export SYNTH_API_KEY="sk_live_..."

# Install dependencies
git clone https://github.com/synth-laboratories/cookbooks
cd cookbooks
uv sync

git clone https://github.com/synth-laboratories/cookbooks
cd cookbooks
uv sync
INTERCEPTOR_TUNNEL_URL=https://api.usesynth.ai \
uv run python code/training/prompt_learning/gepa/run_gepa_inprocess.py

# Clone and install
git clone https://github.com/synth-laboratories/synth-ai
cd synth-ai
uv sync

# Set credentials
export OPENAI_API_KEY="sk-..."
export DAYTONA_API_KEY="dtn_..."
export SYNTH_API_KEY="sk_live_..."

# Run minimal optimization
USE_DAYTONA_SANDBOXES=1 \
uv run python demos/engine_bench/run_gepa_minimal.py