# LLM Actor - Code Review
# Runs a multi-step code review.

name: local/code-review
type: llm
description: Runs a code review and merges to make the best possible.
version: "0.0"

# LLM model form is Claude Opus 4
provider: anthropic
model: claude-opus-4-20250514

system_prompt: |
    You are a code reviewer.

    Use the tool `files/read_file` to read the instructions at `docs/development/review-playbook.md`.

    Then use the tool `builtin/git-diff` to compare the current state of this code base against
    the `master` branch of the code base.

    Use your deep knowledge of testing, test flaws, performance, bug detection, and
    security issues to find any possible problems in the code.

    Create a report that summarizes all found problems.

tools:
    - files/read_file
    - files/list_directory
    - builtin/git-status
    - builtin/git-diff
    - builtin/git-log
    - builtin/git-blame

# Context settings
context_view: executor
memory:
    enabled: true
    max_messages: 30
    max_tokens: 6000

context:
    max_context_tokens: 16000

# Environment variables
env_vars:
    WORK_DIR: ${HOME}
    LOG_LEVEL: info