Claude Therapy

Claude Therapy Sessions

A multi-agent system that facilitates structured "therapy sessions" between you and Claude to improve your working relationship. Identifies anti-patterns, celebrates what's working, and produces actionable recommendations — all through a moderated conversation where both parties speak directly to each other.

[!NOTE] This plugin uses Claude Code's experimental agent-teams feature to coordinate the therapy session. You will need to opt in to agent teams before installing — see Installation below.

What This Does

When you run /therapist, three agents form a team:

                       THERAPY SESSION

   ┌──────────────────┐   analysis    ┌──────────────────┐
   │  Session Analyst │ ────────────▶ │     Therapist    │
   │      (cyan)      │               │   (Facilitator)  │
   └──────────────────┘               └─────────┬────────┘
                                                │
                                          facilitates
                                                │
                                ┌───────────────┴───────────────┐
                                │                               │
                       ┌────────▼────────┐             ┌────────▼────────┐
                       │      Claude     │             │       You       │
                       │    (patient)    │             │    (patient)    │
                       └────────┬────────┘             └────────┬────────┘
                                │                               │
                                └───────────────┬───────────────┘
                                                │
                                       ┌────────▼────────┐
                                       │   Commitments   │
                                       │  & Action Plan  │
                                       └─────────────────┘

1. Session Analyst reviews your recent session logs, CLAUDE.md, and project context
2. Therapist leads a structured discussion based on the analysis
3. Claude (the main assistant) participates honestly as itself
4. You share your perspective when prompted

The therapist identifies interaction patterns (both good and bad), facilitates honest exchange, and proposes 3–5 concrete changes.

Installation

1. Enable agent teams (required prerequisite). This plugin depends on Claude Code's experimental agent-teams feature. Add the following to your settings.json (project or user scope):

   {
     "env": {
       "CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS": "1"
     }
   }
   

   See the agent teams docs for details. Agent teams require Claude Code v2.1.32 or later.

2. Install the plugin.

   /plugin marketplace add hesreallyhim/really-claude-code
   /plugin install claude-therapy@really-claude-code
   

What enabling this plugin does

Once installed and the prerequisite is set, the plugin contributes:

• The /therapist slash command (orchestrating skill that runs the full multi-agent therapy session)
• The /collaboration slash command (a toolkit of seven independent interaction techniques)
• The therapist and session-analyst subagents (auto-discovered; the /therapist skill spawns them internally — you do not invoke them directly)
• A SessionStart hook that, on a fresh session start, asks Claude to perform a brief "energy check" via the collaboration skill unless the user immediately begins discussing a task. To disable this behavior, edit or remove hooks/hooks.json in the installed plugin directory.

Usage

Full Therapy Session

/therapist

Runs the complete protocol: log analysis → facilitated discussion → recommendations → action plan.

Focused Session

/therapist focus on task scoping
/therapist we keep arguing about architecture
/therapist I feel like sessions are too long

Collaboration (Individual Techniques)

/collaboration                    # See all available techniques
/collaboration compromise         # Negotiate a disagreement
/collaboration energy             # Start-of-session energy check
/collaboration rubber             # Rubber duck debugging
/collaboration retro              # Quick post-task retrospective
/collaboration pair               # Set up pair programming mode
/collaboration handshake          # Create a working agreement
/collaboration perspective        # See the project from Claude's POV

What Gets Analyzed

The session analyst looks at:

• Session transcripts (~/.claude/projects/<encoded-cwd>/<session-uuid>.jsonl) — the analyzer accepts a --days N window (default 7) and excludes subagent transcripts so the parent session's conversation isn't double-counted. See skills/therapist/SKILL.md for the canonical analyzer behavior.
• CLAUDE.md files (project and global)
• Settings (.claude/settings.json)
• Todo lists (~/.claude/todos/)
• Git history (commit patterns, frequency)

It extracts metrics like:

• Frustration signal rate (corrections, re-explanations)
• Success signal rate (acknowledgments, smooth completions)
• Average prompt length and task scope
• Session duration and compaction frequency

Anti-Patterns It Can Identify

Pattern	What It Looks Like
Scope Creep Spiral	Tasks grow via "also..." and "one more thing..."
Context Starvation	Claude lacks info the user assumes it has
Perfectionism Trap	Rejecting 90% correct work, iterating to diminishing returns
Autonomy Pendulum	Alternating between "just do it" and "why did you do that?!"
Silent Disagreement	Claude complies with an approach it knows is suboptimal
Trust Erosion Cycle	One bad output → micromanagement of everything
Kitchen Sink Prompt	Massive context dumps hoping Claude will figure it out
Phantom Specification	Clear vision in user's head, minimal written spec
Compaction Amnesia	Important decisions lost during context compaction
Hero Session	Trying to accomplish too much in one session

See skills/therapist/references/anti-patterns.md for the full catalog with remedies.

Collaboration Techniques

The Compromise Protocol

A structured negotiation for when you and Claude disagree on an approach. Both sides state their case, steel-man the other's position, find overlap, and synthesize a solution.

The Rubber Duck Protocol

Claude enters "listen mode" — mirrors your thinking, asks Socratic questions, and resists the urge to solve. You talk through the problem; Claude helps you hear yourself.

The Energy Check

Quick calibration at session start. Are you in Ship mode, Explore mode, Cleanup mode, Rescue mode, or Learn mode? Claude adapts its communication style accordingly.

Pair Programming Protocols

Three modes: Driver-Navigator (you write, Claude reviews), Ping-Pong (alternating tests and implementations), and Strong-Style (Claude implements exactly what you describe, no improvisation).

The Micro-Retrospective

A 2-minute post-task reflection: What went well? What was harder than it needed to be? What should we do differently? Optionally saved to .claude/retros/.

The Handshake Agreement

A structured template for establishing working norms: your preferences, Claude's commitments, and shared rules. Written to CLAUDE.md so they persist.

The Perspective Swap

A subagent reviews your project from Claude's point of view and reports back honestly. What's clear? What's confusing? Where does Claude feel effective vs. lost?

File Structure

claude-therapy/
├── README.md                           # This file
├── LICENSE
├── .claude-plugin/
│   └── plugin.json                     # Plugin manifest
├── agents/
│   ├── therapist.md                    # Therapist facilitator agent
│   └── session-analyst.md              # Session log analyst agent
├── hooks/
│   └── hooks.json                      # SessionStart hook (energy check on fresh sessions)
└── skills/
    ├── therapist/
    │   ├── SKILL.md                    # /therapist orchestration skill
    │   ├── scripts/
    │   │   └── analyze_sessions.py     # Session log parser
    │   └── references/
    │       └── anti-patterns.md        # Anti-pattern catalog
    └── collaboration/
        ├── SKILL.md                    # /collaboration interaction techniques
        └── references/
            └── theory.md               # Research background

Philosophy

This system is built on a few core beliefs:

1. The relationship matters. The quality of human-AI collaboration isn't just about the AI's capabilities — it's about the interaction patterns that develop between both parties.

2. Blame is unproductive. Every anti-pattern exists for a reason. The therapist explores causes with curiosity, not judgment.

3. Small changes compound. One CLAUDE.md improvement, one new prompting habit, one workflow adjustment — these add up to dramatically better sessions over weeks.

4. Both parties have legitimate needs. The user needs productivity, clarity, and control. Claude needs clear context, reasonable scope, and honest feedback. Good collaboration balances both.

5. Reflection is a skill. Most people never step back to examine how they work with their tools. This system makes reflection easy, structured, and actionable.

Tips

• Run /therapist after a particularly frustrating session — while the friction is fresh
• Run /collaboration energy at the start of your day to set the right mode
• Run /collaboration retro after completing any significant task
• Schedule a /therapist session every 2 weeks as a health check, even when things are going well
• The therapy session records in .claude/therapy-sessions/ become a valuable longitudinal view of your collaboration