Pre-flight critic for user prompts. Diagnoses the 5 dimensions agents need to execute well — scope, metric, baseline, subject, success criterion — and proposes a sharper version before you hit send.
Distributed as a Claude Code plugin containing a single skill (prompt-doctor).
/plugin marketplace add stepcha/prompt-doctor
/plugin install prompt-doctor@prompt-doctorCopy skills/prompt-doctor/ into the host's skills directory:
- Project-local:
<repo>/.claude/skills/prompt-doctor/ - User-global:
~/.claude/skills/prompt-doctor/
Invoke explicitly or let the skill auto-trigger when a vague prompt arrives.
/prompt-doctor "refactor auth to make it cleaner"
/prompt-doctor "refactoriza auth" # multilingual input — Spanish also works
The skill returns a Diagnosis table (✓ / ✗ / partial per dimension), an Improved version with [DEFAULT: ...] markers for assumed gaps, and 1–2 high-risk Confirmation questions.
See skills/prompt-doctor/SKILL.md for full operating modes (critique-only / critique+execute / learning mode), domain detection table, anti-pattern catalog, and output format.
MIT — see LICENSE.