Guides
Model and effort
The model and the effort level are set in your tool, not in the skills. Below is where each tool sets them, whether its helpers inherit your choice, and some reference points for Claude Code.
The blueprint never picks a model or an effort level for you, in any tool. No skill or helper prompt sets either one, so a pipeline reasons at whatever level your session runs, and a helper inherits that choice wherever the tool passes it on. A prompt whose role header says it is safe at lower effort may run lower only where the tool takes a per-helper effort and you have not asked for your level everywhere. The blueprint never switches models to save effort.
| Tool | Set the model | Set the effort | Do helpers inherit them? |
|---|---|---|---|
| Claude Code | /model in a session, or claude --model <model> |
/effort, claude --effort <level>, or CLAUDE_; a level saved in settings applies below those |
Yes: a helper without an effort of its own inherits the session level; verified on 2.1.284 |
| Codex | /model in a session, or codex -m <model> |
model_ in ~/ (low to ultra, by model) |
Yes: a subagent inherits the parent’s model and effort unless configured otherwise, and spawn_agent can set effort per helper |
| Antigravity | /model in a session (saved), or agy --model <model>; agy models lists them |
agy --effort low| |
Not verified: a subagent definition can set a model tier, and no per-dispatch override is documented |
| Grok Build | /model, grok -m <model>, or [models] default in ~/ |
--effort none| |
Not verified: per-dispatch model and effort are not documented |
| Pi | /model, or pi --model <pattern>[:thinking] |
/thinking, or --thinking off| (clamped to the model) |
Only with the pi-subagents package, which takes a per-dispatch model:"provider/; inheritance not verified |
| Cursor CLI | /model picker, or cursor-agent --model 'id[effort= |
Inside the model string: id[effort=; there is no separate effort flag |
Helpers defined in agent files can say model: inherit; for the blueprint’s helpers, not verified |
| Hermes | /model in a session, or hermes -m <model> --provider <provider> |
/reasoning high in a session, or agent. in ~/ |
No: helpers use the global delegation.model and delegation. settings |
| Amp | The Dial: a mode (low, medium, high, ultra) sets model, effort, prompt and tools together, chosen before the first message and fixed for the thread |
Part of the mode | Not verified: a prompt cannot pick a model per helper |
Claude Code specifics
Opus 5.5 is the default model on every plan from CLI 2.1.280 and starts sessions at medium effort; other current models start at high. A level set with --effort, /effort or the environment variable wins over settings, and settings win over the model default. Some reference points:
| Session setting | Fits |
|---|---|
Opus 5.5 at high |
A solid default for pipeline runs |
Opus 5.5 at xhigh, or Fable 5.1 at high |
Hard or high-stakes work; slower and costlier |
Opus 5.5 at medium (its starting level) or lower |
Small tasks, quick fixes and questions |
The ship runner passes no model or effort flag of its own, so an unattended run uses the defaults you saved in the tool.