model-policy

name	model-policy
description	Choose behavior-analysis models and runtimes with explicit tradeoffs around reproducibility, cost, privacy, speed, and artifact compatibility.
metadata	{"annolid":{"always":false}}

Model Policy

Use this skill when selecting between hosted reasoning, local models, existing Annolid backends, or heavier optional pipelines.

Rules

1. Prefer existing Annolid perception backends for default execution.
2. Prefer typed, replayable outputs over one-off model calls.
3. Use hosted reasoning for planning and interpretation when it materially helps.
4. Keep privacy-sensitive or offline workflows compatible with local fallback.
5. Do not silently switch to a heavier stack without explaining why.

Output

State:

• selected policy,
• why it fits the task,
• what evidence it is expected to produce,
• any runtime or reproducibility tradeoff.