Agent roles and models
A role is a job in the workflow, not a person or an account. Each role has an agent and a model, set in .agents/agents.conf, one role per line:
implementer: claude opus
reviewer: codex gpt-6-luna| Role | What it does |
|---|---|
project-grill |
Asks you questions about a new project and writes the requirements |
project-planner |
Writes the architecture, the roadmap, and the decision records |
planning-reviewer |
Reviews the plan, read-only |
implementer |
Implements a feature and writes its tests |
reviewer |
Reviews a feature, read-only |
triage |
Proposes what to do with each review finding |
triage-implementer |
Fixes the findings that were selected |
analyst |
Compares the runs of several features and proposes improvements, read-only (learning from your runs) |
The supported agents are Claude Code (claude) and Codex (codex). You can override the agent and the model for one run with --agent and --model on a workflow script. The model you name is passed to the agent as it is: the workflow never replaces it with another one, and an unknown agent is an error.
The reviewer must use a different agent than the implementer, at every risk level: that is what makes the review independent. review-feature.sh refuses to start a reviewer that resolves to an agent the implementer of the feature ran with, and doctor.sh warns when a reviewing role and the role it reviews resolve to the same agent.
A role of a feature (implementer, reviewer, triage, triage-implementer) can also have a line per risk level of the feature, such as implementer@high: claude fable next to implementer: claude opus. A script that starts the role takes --agent and --model first, then the line of the level the feature has at that moment, then the plain line; a later session of a feature never runs at a weaker level than an earlier one did. Without a level line the plain line applies. doctor.sh validates the lines, and report-runs.sh shows the calls per level and model.
What the template ships. These are the template’s own choices; set the agents, models, and levels your project wants:
| Role | Low and medium risk | High risk |
|---|---|---|
implementer, triage-implementer |
Claude, opus |
Claude, fable |
reviewer |
Codex, gpt-6-luna |
The same |
triage |
Claude, sonnet |
The same |
project-grill, project-planner |
Claude, fable |
|
planning-reviewer |
Codex, gpt-6-astra |
|
analyst |
Claude, fable |
The levels come from the features that built the template: a feature that changes the workflow and its gate, which is high risk, needed one or two rounds with the strongest model and many with a weaker one, while the other features did not need it. Codex stays supported as implementer for a project that configures it.
Shipped configuration
The active lines of .agents/agents.conf:
project-grill: claude fable
project-planner: claude fable
planning-reviewer: codex gpt-6-astra
implementer: claude opus
implementer@high: claude fable
reviewer: codex gpt-6-luna
triage: claude sonnet
triage-implementer: claude opus
triage-implementer@high: claude fable
analyst: claude fable