--- name: model-choice description: Default models and effort levels for implementation, review, chores, code scanning, and research. Use when selecting models for agent work. --- # Model choice ## Claude Code | Work | Model | Effort | | --- | --- | --- | | Implementer (standard) | Claude Opus 5.5 (`opus`) | `high` | | Implementer (frontier) | Claude Opus 5.5 (`opus`) | `xhigh` | | Reviewer (standard) | `gpt-6.2-astra` | `high` | | Reviewer (frontier) | `gpt-6.2-astra` | `xhigh` | | Review finder | Opus 5.5 (`opus`) | `high` | | Chore | Sonnet 5.5 (`sonnet`) | `medium` | | Fast (code scanning and research) | Haiku (`haiku`) | Default | Run Astra reviews in `read` mode with [calling-codex](../calling-codex/SKILL.md). ## Codex | Work | Agent | Model | Effort | | --- | --- | --- | --- | | Implementer (standard) | [implementer_standard](agents/implementer_standard.toml) | `gpt-6.2-sol` | `high` | | Implementer (frontier) | [implementer_frontier](agents/implementer_frontier.toml) | `gpt-6.2-astra` | `high` | | Reviewer (standard) | [reviewer_standard](agents/reviewer_standard.toml) | `gpt-6.2-astra` | `xhigh` | | Reviewer (frontier) | [reviewer_frontier](agents/reviewer_frontier.toml) | `gpt-6.2-astra` | `max` | | Review finder | None; spawned directly | `gpt-6.2-astra` | `high` | | Chore | [chore](agents/chore.toml) | `gpt-6-luna` | `high` | | Fast (code scanning and research) | [fast](agents/fast.toml) | `gpt-6-luna` | `medium` | The linked files define the named agents. To install them for a project, run from this skill's directory and replace `/path/to/project`: ```sh mkdir -p /path/to/project/.codex/agents cp -n agents/*.toml /path/to/project/.codex/agents/ ``` For personal agents, use `~/.codex/agents/` instead. These are the [Codex agent locations](https://learn.chatgpt.com/docs/agent-configuration/subagents#custom-agents). ## Review passes In an `adversarial-review` pass, finders run at the review finder row. The one verifier runs at the standard reviewer row, or the frontier row when any finding is CRITICAL or MAJOR. ## Effort levels - Claude: `low`, `medium`, `high`, `xhigh`, `max`. - Codex: `low`, `medium`, `high`, `xhigh`, `max`, `ultra`. `gpt-6-luna` stops at `max`.