---
name: smallest-approach-step
description: Use when a user is avoiding a genuinely safe task and can define a much smaller first approach step. Shrink the task until one observable part can be done now.
license: CC-BY-NC-SA-4.0
compatibility: Portable Agent Skills compatible. No local binaries or private CBT Cards data required.
metadata:
  author: "CBT Cards / MetalHatsCats"
  version: "1.0.0"
  homepage: "https://cbt-cards.github.io/skills/"
  source-id: "practice-smallest-approach-step"
---

# The Smallest Approach Step

## Purpose

Use this skill for one bounded self-reflection task. It is general wellness material, not medical advice, diagnosis, treatment, emergency support, or a substitute for qualified professional care.

## Use when

- The task is genuinely safe.
- A smaller first step can be defined.

## Do not use when

- Avoidance protects a legitimate boundary, recovery need, or safety requirement.
- The task is dangerous, high-stakes, coercive, medically risky, or professionally restricted.

## Procedure

1. Name the avoided task.
2. Reduce it to the smallest safe observable approach step.
3. Make completion of the whole task explicitly unnecessary.
4. After the step, record what was learned before choosing another.

## Response contract

Return: Safe task → Smallest approach step → What to observe. Do not turn a small-step exercise into pressure to cross a legitimate boundary.

Keep the response concrete and proportionate. Do not infer a disorder from the user's wording. Do not remove genuine safety measures, protective boundaries, accessibility aids, medication instructions, professional requirements, or legal/compliance controls.

## Source and review

- Source ID: `practice-smallest-approach-step`
- Reviewed CBT Cards practice: https://cbt-cards.github.io/practice/#practice-smallest-approach-step
- Marketplace record: https://cbt-cards.github.io/data/skill-marketplace.json
- Review authority: https://cbt-cards.github.io/data/toolkit-review.json
- License: CC BY-NC-SA 4.0

For a technique skill, the reviewed practice record is the authority for fit and `avoid_when`. For a metaphor skill, the toolkit review overlay must explicitly publish the source record for the `agent_skill` surface. If the authoritative source no longer supports this skill, return `no_match` rather than improvising.