AI Psychiatry 0.5.0 — Complete Executive Control

AI Psychiatry 0.5.0

This release expands AI Psychiatry into a complete executive-control and reasoning-balance plugin for Claude Code and Codex.

New superpowers

• never-stop: explicit maximum-autonomy execution within existing permissions. It suppresses unnecessary questions, classifies routine/material/hard-gate decisions, uses an eight-level recovery ladder, continues independent work around blocked branches, validates blockers, reports real background progress, and stops only at proven completion or a genuine approval gate.
• all-the-medicine: a manifest-driven meta-superpower that loads and evaluates every registered public skill, applies only relevant controls, resolves conflicts deterministically, prevents self-recursion, and recompiles automatically when skills change.

Direct communication

• Added direct-communication for assertive, succinct, plain-language answers.
• Leads with the result, removes filler and repetition, preserves required evidence, and asks only one precise blocking question when input is genuinely required.
• Available as /ai-psychiatry:direct-communication in Claude Code and $direct-communication in Codex.
• Included automatically in all-the-medicine.

Reasoning and behavioral controls

• Balances underthinking and overthinking through sufficient-reasoning gates.
• Detects progress theater, premature completion, false blockers, hidden recursion, strategy laundering, scope laundering, context starvation, memory poisoning, shallow debugging, and unsupported assumptions.
• Preserves ADHD-, OCD-, hallucination-, overthinking-, perfectionism-, and nested-task behavioral analogies without making clinical claims about AI.
• Adds controlled executive overrides when critical evidence requires narrowly extending a budget.

Packaging and compatibility

• Shared Claude Code and Codex plugin release.
• 31 public skills, 49 runtime controls, and 58 focused rules.
• Codex branding assets and composer icon included.
• Codex default prompts satisfy the three-prompt and 128-character limits.
• Claude and Codex plugin validators pass.
• Portable full-plugin archive attached: ai-psychiatry-0.5.0.zip.

Validation

• 110 framework tests passed.
• Generated all-the-medicine inventory is current and self-recursion free.
• Framework, Claude Code, Codex, skill-schema, manifest, archive, and prompt-length validation passed.

Commands

Claude Code:

/ai-psychiatry:never-stop
/ai-psychiatry:all-the-medicine
/ai-psychiatry:direct-communication

Codex:

$never-stop
$all-the-medicine
$direct-communication

Install or update

Claude Code marketplace:

/plugin marketplace add ihabkhaled/AI-Psychiatry
/plugin install ai-psychiatry@ihabkhaled-ai

Existing Claude Code installations can update with:

claude plugin marketplace update ihabkhaled-ai
claude plugin update ai-psychiatry@ihabkhaled-ai

Restart Claude Code after updating.

AI Psychiatry 0.3.0

Adds semantic anti-bypass enforcement and sufficient-reasoning controls for false progress/completion/blockers, hidden recursion, strategy and scope laundering, memory/context integrity, underthinking, overthinking, and evidence-based delivery. Includes 28 public skills, 56 rules, a portable runtime, and validated Claude/Codex marketplace manifests.