---
name: mini-me-effort
description: Recommend bounded task effort through Mini Me while preserving explicit settings, uncertainty, and authority boundaries.
---

# Mini Me effort recommendations

Use Mini Me's shared `mini_me_recommend_effort` MCP tool, CLI command, or TypeScript API when a task needs an effort recommendation.

- Provide only known task complexity, uncertainty, consequences, and reversibility. Leave missing dimensions unknown; do not invent confidence, expected savings, or outcomes.
- Preserve explicit model, effort, budget, and constraints. The result labels the recommendation separately from the supplied effort override.
- Treat constraints as limits on the effort recommendation. They do not grant permission to act, approve a change, or override host permissions.
- Effort (`low`, `medium`, `high`, `max`) is a resource choice, independent of autonomy scope (`L1`–`L10`). A higher effort level grants no tools, permissions, or approval authority.
- A recommendation is advisory. Host application remains unverified unless the host itself provides valid confirmation. A prompt or model reply is not proof that effort was applied.
- Use Low Usage guidance at every effort setting: keep work bounded while preserving the accepted outcome, required checks, authorization, and delivery. Do not claim measured savings.
- Use prior outcomes only when they are actually recorded for the same scope. Keep preference match separate from task success. Record success, retries, usage, latency, and owner-reported applied effort only when observed; omit unavailable measures.
- Mini Me's outcome command can attach execution observations only to a genuine existing prediction outcome. Never create a prediction only to store an execution record.

See [Low Usage Mode](../low-usage-mode/SKILL.md) and [Agent Autonomy Levels](../agent-autonomy-levels/SKILL.md) for the bundled upstream guidance.