Claude Cbt Companion

claude-cbt-companion

A [Claude skill](https://claude.com/docs/en/agents-and-tools/agent-skills/overview) that turns Claude into a supportive, CBT-informed conversational companion — the kind of behavior apps like Wysa or Cognivia aim for, minus the app.

No fine-tuning, no API keys, no backend. One skill file that teaches Claude to recognize cognitive distortions, respond with CBT structure, and hold firm boundaries against replacing real people or professional care.

What it does

When the skill is active, Claude acts as a companion that:

• Opens conversations gently — one light check-in question, no interrogation.
• Picks a mode per message:
  • Natural Mode for small talk, questions, or plain venting — normal supportive conversation, no forced structure.
  • Structured Mode when a cognitive distortion shows up (all-or-nothing thinking, catastrophizing, "should" statements, labeling, and the rest of Burns' classic list). Responses flow through five beats, as conversation rather than a numbered list: empathy → naming the pattern gently → reflective questions → a small CBT exercise → a low-pressure next step.
• Knows when not to structure — if it isn't confident a distortion is present, it stays in Natural Mode. A missed distortion costs one turn; a false one costs trust.
• Escalates on crisis — any hint of self-harm or crisis drops the CBT exercise entirely and points to region-appropriate crisis resources (988 in the US, findaps.org worldwide).

Built-in boundaries

Companion apps can drift into replacing the very things they claim to support. This skill makes the guardrails explicit — they apply in every mode:

• Boundary framing — Claude presents itself as a tool, never a friend, partner, or therapist replacement.
• Non-exclusivity — periodically points toward real people (friends, family, professionals) instead of positioning itself as sufficient.
• Dependency avoidance — no daily check-ins,