claude/for
Search
Subscribe
Claude CodeUpdated

How do I run Claude Code `/doctor prompt-audit` on CLAUDE.md and skills?

Arthur Teboul
Arthur TeboulEditor, claude/for
UPDATED OCT 22 · 5 MIN

Run `/doctor prompt-audit` (Claude Code 2.1.283+) to audit CLAUDE.md, skills, agents, and commands for outdated or conflicting instructions. Alias `/checkup prompt-audit`. Not `/skill-doctor`.

Run /doctor prompt-audit (alias /checkup prompt-audit) on Claude Code ≥ 2.1.283 to audit CLAUDE.md files, skills, and related config for outdated or conflicting instructions. The CHANGELOG also names agents and commands, and patterns written for older models. Pass a path to audit one file or directory instead, and nothing changes until you ask Claude to apply the proposed edits. This is not /skill-doctor (skill context cost + usage) and not the default /doctor install checkup.

TL;DR: In a session on v2.1.283+, type /doctor prompt-audit or /checkup prompt-audit. Claude audits instruction files instead of running the setup checkup. Use /skill-doctor when you need unused-skill cost, not prompt-pattern audit.

Prerequisites

  1. Claude Code ≥ 2.1.283: the commands docs floor for the prompt-audit subcommand; the CHANGELOG adds it under ## 2.1.283. Check with claude --version.

  2. Interactive session: slash /doctor … runs inside a session. From the terminal, bare claude doctor is a different surface: read-only install diagnostics without starting a session (commands docs).

  3. Something worth auditing: project/user CLAUDE.md, skills, agents, or commands that may still teach older-model habits after a Sonnet/Opus upgrade.

Run /doctor prompt-audit

  1. Start (or attach to) a Claude Code session in the terminal.

  2. Type one of:

/doctor prompt-audit

/checkup prompt-audit
  1. Optional path, to audit one file or directory:
/doctor prompt-audit .claude/skills/deploy

Per the memory docs, the audit by default covers your CLAUDE.md, CLAUDE.local.md, and AGENTS.md files, plus the rules, skills, commands, subagents, and output styles under .claude/ and ~/.claude/. A path narrows it to that one file or directory.

What it audits (official wording):

SourceScope
Commands docsCLAUDE.md files, skills, and other configuration for outdated or conflicting instructions, instead of the checkup
CHANGELOG 2.1.283CLAUDE.md files, skills, agents and commands for prompting patterns written for older models

Report priorities (CHANGELOG improve under 2.1.283 history): stale paths, stale commands, and contradicting instruction files lead the report; thinking keywords that Claude Code documents are kept.

Contrast: plain /doctor / /checkup is a broader setup checkup that reports findings first and asks for confirmation before changing anything. prompt-audit also waits for you: the memory docs say you get "a report of findings with proposed edits, and nothing in your files changes until you ask Claude to apply them."

The Claude moves worth stealing, by email

Workflows tested on real jobs, sent only when they are worth it. Free.

/doctor prompt-audit vs /skill-doctor vs plain /doctor

CommandJob
/doctor prompt-audit / /checkup prompt-auditInstruction-pattern / conflict audit (older-model prompting, stale paths/commands, contradicting files)
/skill-doctorSkill context cost + usage → prune / turn off
/doctor / /checkup (no subcommand)Setup checkup: install health, unused skills/MCP/plugins vs context cost, CLAUDE.md trim/migrate, and more

One-line boundary: /claude-api … prompt-audit is a different bundled-skill path (flags older-model instructions in API prompts / skills / tool descriptions; floor v2.1.221+) and not covered here.

When to run it

  • After model upgrades when CLAUDE.md or skills still teach old thinking / tool_choice / computer-tool habits.
  • When skills, agents, or commands contradict each other or point at stale paths/commands.
  • When plain /doctor checkup is fine for install health, but you specifically need a prompt-pattern pass.

If permission / auto-mode defaults surprise you after a Claude Code bump, pair with How do I configure Claude Code permissions?. For sharing / nesting instruction files, see How should I share AGENTS.md / CLAUDE.md?.

Pitfalls

  • Unknown subcommand: upgrade to 2.1.283 or later.
  • Wrong doctor: /skill-doctor does not audit prompt patterns; plain /doctor does not replace prompt-audit.
  • Invented flags: the only documented form is /doctor [prompt-audit [path]]. No third-party blog flags.
  • Confusing with /claude-api prompt-audit: API-skill path, different job.
  • Turned-off bundled skill: the audit runs through the bundled /claude-api skill, so it's unavailable while that skill is off in skillOverrides or with disableBundledSkills.

FAQ

What version do I need?

Commands docs: prompt-audit requires Claude Code v2.1.283 or later. CHANGELOG adds /doctor prompt-audit (also /checkup prompt-audit) under 2.1.283.

Is /checkup prompt-audit the same?

Yes. /checkup is the documented alias of /doctor.

Does it replace /skill-doctor?

No. Skill-doctor is cost/usage prune; prompt-audit is instruction-pattern audit. Related: How do I use /skill-doctor?.

Does it edit files automatically?

No. Per the memory docs, you get a report of findings with proposed edits, and nothing in your files changes until you ask Claude to apply them.

Can I pass a path?

Yes. A path audits that one file or directory instead of the default set, for example /doctor prompt-audit .claude/skills/deploy.

Sources (checked 2026-10-22)

The Claude moves worth stealing, by email

Workflows tested on real jobs, sent only when they are worth it. Free.