Agent Clinic — Development skill for Claude Code
Diagnostic plugins for Claude Code and Codex: why a session went wrong, and whether a proposed change is too much or too little for the outcome.
How to install Agent Clinic
This entry records only its repository, not the path inside it, so there is no
exact command to give. Open shinpr/agent-clinic and copy the folder into
~/.claude/skills/, or the file into ~/.claude/agents/.
What Agent Clinic does
Diagnostic plugins for Claude Code and Codex: why a session went wrong, and whether a proposed change is too much or too little for the outcome.
Alternatives in Development
- Dental Clinic — Page builder: Elementor 648 ★
- Directional Prompting — Outcome-first plus directional language 138 ★
- Coder Eval Create Plan — Create a structured, phased implementation plan for a feature or change in the coder_eval codebase, executable 116 ★
README
Agent Clinic

Diagnostics for agent-assisted work. Each one is a separate plugin, and you install only the ones you want.
| Plugin | What it is for | How it runs | Needs |
|---|---|---|---|
| Session Doctor | A session went badly and you want to know why. | You ask for it | Nothing |
| Outcome Doctor | A plan, design, or diff may be more or less than the job needs. | On its own, while your agent works | A TypeSafe API key |
Both report what they found and leave the fixes to you. Neither one edits your code.
Add the marketplace once:
/plugin marketplace add shinpr/agent-clinic
codex plugin marketplace add shinpr/agent-clinic
Session Doctor
Looks back at one of your past Claude Code or Codex sessions and tells you what to change.
Sessions go wrong for boring reasons. A request with no "done" condition. A task that quietly vanished at a compaction. A skill that overrode one of your own rules. Session Doctor reads the saved transcript and looks for problems like these:
- What you asked for: where the request left the outcome, scope, or "done" undefined.
- How the work was run: work the agent started but never finished, state lost during compaction, subagent results that were never checked.
- What was instructing the agent: your skills,
AGENTS.mdandCLAUDE.mdas they actually combined in that session, including rules that conflict or duplicate each other when they apply at the same time.
The three passes run separately, so a hunch from the first does not bias the other two. The report shows what went wrong, where it happened, and the smallest change likely to prevent it next time.
Install
/plugin install session-doctor@agent-clinic
codex plugin add session-doctor@agent-clinic
Start a new session afterwards so the skill is loaded.
Use
/recipe-diagnose # Claude Code
$recipe-diagnose # Codex
T
Related Skills
Fables
Four Claude Code / Codex plugins for spending tokens where they change the outcome
Incident Capture
Capture a structured post-mortem when something went wrong. Walks operator through 4 questions, writes ~/.clau
Forensic
Investigate what went wrong in a failed build.
Gsd Forensics
Post-mortem investigation for failed GSD workflows — diagnoses what went wrong.
Explain Error
You are executing the /explain-error command. Your job is to diagnose a Coq type error: parse it, gather conte
Context Files Coding Agents
Do Context Files Help Coding Agents? Two-agent (Claude Code + Codex) ablation on whether AGENTS.md/CLAUDE.md c
Related Agents
Retro
Retrospective facilitator that captures learnings from sprints and sessions, surfaces past insights, and compo
Log Analyzer
Delegate here to analyze stderr output, error logs, stack traces, and debug command failures. Fast triage of w
Lean Main
Default main-session agent for token-efficient work. Behaves like normal Claude Code, but reads as little as p