Instruction file imported from iberi22/gastronomic-open-standard-GOS (.github/instructions/codex-tools.instructions.md). Copyright stays with the author.

GPT Codex Tool Calling Best Practices

Tool Definition Format

GPT Codex uses OpenAI's function calling format:

{
  "type": "function",
  "function": {
    "name": "tool_name",
    "description": "Clear description of the function",
    "parameters": {
      "type": "object",
      "properties": {
        "parameter_name": {
          "type": "string",
          "description": "What this parameter does"
        }
      },
      "required": ["parameter_name"]
    }
  }
}

Best Practices

1. Agentic Workflows

GPT Codex excels at multi-step coding:

• Plan before implementation
• Break down complex tasks
• Execute incrementally
• Verify results

2. Tool Selection

• Limit active tools to 10-20 for best results
• Use tool sets for related functions
• Enable only relevant tools per task

3. Parallel Function Calling

• Use for independent operations
• Batch file reads
• Parallelize searches

4. Structured Outputs

• Request specific JSON schemas
• Use consistent response formats
• Validate return types

Codex-Specific Features

Code Generation Excellence

GPT Codex is optimized for:

• Production-ready code
• Test generation
• Refactoring
• Documentation

Multi-File Operations

Can handle:

• Cross-file refactors
• Module reorganization
• Dependency updates

Error Recovery

• Handles tool failures gracefully
• Suggests alternatives
• Retries with corrections

Git-Core Protocol Integration

Atomic Commits

# Wrong: Giant commit
git add .
git commit -m "big update"

# Right: Atomic commits
git add src/auth/
git commit -m "feat(auth): add login endpoint #42"

git add tests/auth/
git commit -m "test(auth): add login tests #42"

Issue-Driven Development

1. Read issue first

   gh issue view 42

2. Create feature branch

   git checkout -b feat/issue-42

3. Implement with references

   git commit -m "feat(scope): description #42"

4. Create PR

   gh pr create --fill

Example: Multi-Step Implementation

# Step 1: Plan
tools = ["search", "read_file"]
# Understand codebase structure

# Step 2: Implement
tools = ["edit_file", "create_file"]
# Make changes atomically

# Step 3: Verify
tools = ["run_command", "test_failure"]
# Run tests and verify

# Step 4: Commit
tools = ["run_command"]
# git add, commit, push

Temperature Settings

For coding tasks:

• 0.0-0.3: Deterministic code generation
• 0.5: Balanced creativity/accuracy
• 0.7+: Avoid for production code