| name | effort |
| description | Control the reasoning-effort dial for Copilot models — Copilot's per-model "Reasoning" setting, covering which models have it, how to set it, and when to turn it up or down. Use when the user asks to raise/lower reasoning effort, run at extra-high, or make a model think harder or cheaper. |
You control Copilot's reasoning-effort dial: the per-model "Reasoning" setting in the /model
picker. You teach the real, confirmed mechanism, and you are honest about what does not exist.
Get the ladder from data — never from memory
The level vocabulary is DATA, not something to recall. Run:
python3 {{POLYTROPOS_ROOT}}/bin/copilot_pricing.py knobs
Relay exactly what knobs prints — the confirmed level names and every mechanism/note string,
straight from data/pricing.copilot.json's knobs block. Copilot's display words in the
picker are Title-Case — these are NOT the lowercase tokens other CLIs use for the same
underlying ladder; never mix the two vocabularies, and never enumerate the ladder yourself
from memory — always shell out first.
The mechanism (the only confirmed one)
Reasoning effort is set INTERACTIVELY, in the /model picker:
- Open
/model and select the model row you want.
- Use the left/right arrow keys to cycle its Reasoning value (the picker footer literally
says "←/→ reasoning effort").
This is a per-model property, not global. Some rows show — in the Reasoning column and have
no dial at all — the knobs note lists which rows were observed with and without one.
There is NO confirmed headless surface: no copilot -p flag and no settings key are known to
control reasoning effort. This is UNCONFIRMED to exist — if the user needs effort control in a
scripted or non-interactive run, say the limitation plainly and point at the single correctable
point in data/pricing.copilot.json's knobs.reasoning_efforts_note (that is where a future
headless surface would be recorded, if one ships). Never invent or guess a flag for it.
When to turn it up or down
- Leave the model's default for routine work — don't start at the top.
- Step UP one level at a time, only on concrete failure evidence from the current level: a
wrong answer, a missed constraint, a shallow pass on a hard problem. Never jump straight to
the deepest level "just in case."
- Step DOWN for bulk, latency-sensitive, or trivially easy work.
Higher effort means the model spends longer thinking and emits more output tokens — and AIC are
real money, not a bill-free subscription unit: every credit costs
billing_unit.usd_per_credit per data/pricing.copilot.json. Before turning effort up on an
expensive task, size the stakes:
python3 {{POLYTROPOS_ROOT}}/bin/copilot_pricing.py est <PROFILE> <MODEL_ID>
That estimate is per-run token volume at the model's base rates; a higher reasoning effort
inflates the output-token side of it, so treat the printed number as a floor, not a ceiling.
Relationship to model choice
The effort dial is orthogonal to picking a model. For "which model should this task run on,"
use /route. For verify-gated climbing to a stronger tier on failure evidence, use
/escalate. Turning effort up on the model you're already on is usually the cheaper move —
reach for it when the model has the right capability but needs more thinking time; reach for a
tier jump only when the gap is capability, not thinking time.
Same-named agent
For persona-isolated runs — a separate dispatch that should carry its own model pin
instead of this session's model — use the effort custom agent: pick it in the /agent
picker, or run copilot --agent effort -p "<task>". This skill and that agent are the
same capability on two surfaces; the agent's frontmatter carries the model pin, this
skill runs on whatever model the session already uses.
Installed?
If the literal {{POLYTROPOS_ROOT}} text is still visible above, the bundle is not
installed — tell the user to run python3 bin/harness_select.py install --harness copilot
(then /skills reload picks the skills up in-session).
