config-audit

Reviews Claude configuration rules and proposes removals of obsolete or model-constraining entries.

Updated May 3, 2026
One-click install
npx skills add https://github.com/spikelab/multiplai-cc-mktplace --skill config-audit-spikelab
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: config-audit
Source: https://github.com/spikelab/multiplai-cc-mktplace/tree/main/plugins/multiplai-context/skills/config-audit
Command: npx skills add https://github.com/spikelab/multiplai-cc-mktplace --skill config-audit-spikelab

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Claude and multiplai configurations accumulate rules over time that were written for older models or workflows that no longer exist, silently taxing every session with extra context tokens and constraining newer models. This Skill performs a periodic subtractive audit that classifies every standing rule and proposes what to remove, without ever applying changes itself. ## Core Features & Use Cases - Full config surface enumeration: Reads global and workspace CLAUDE.md files, settings.json env and permissions blocks, hook registrations, and memory-file standing rules. - Three-way classification: Labels each rule as still-serving, obsolete, or model-constraining, with evidence cited for every removal candidate. - Removals-first proposal: Writes a dated proposal to .multiplai/dreams/ listing removals, edits, and only the additions a removal requires, then stamps a state file to close the 60-day nudge gate. - Use Case: After a model upgrade, run the audit to find scaffolding rules written for the previous model's weaknesses — such as single-file-refactor limits or redundant hooks — and get a reviewed proposal of what to delete. ## Quick Start Ask the assistant to run the config audit and produce a removals-first proposal of obsolete or model-constraining rules in the current Claude configuration.

Frequently Asked Questions about config-audit

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I audit my Claude Code configuration for outdated rules?

Run the config-audit skill, which enumerates CLAUDE.md files, settings.json blocks, hook registrations, and memory-file rules, then classifies each as still-serving, obsolete, or model-constraining. It writes a removals-first proposal to .multiplai/dreams/ for your review.

Does the config audit change my settings automatically?

No. The skill never edits configuration, hooks, or memory files. Its only writes are the proposal file in .multiplai/dreams/ and a state stamp written via the config_audit.py script; you review and apply changes yourself.

What is a model-constraining rule in Claude configuration?

A model-constraining rule is scaffolding written to work around an older model's weaknesses, such as single-file-refactor limits or step-by-step hand-holding, that newer models no longer need. These rules now suppress better behavior and are prime removal candidates.

Why does the config audit nudge keep appearing every session?

The SessionStart nudge fires until the state file is stamped. The audit must finish by running config_audit.py --stamp, which writes config_audit_state.yaml next to dream_state.yaml; skipping this step leaves the 60-day gate open.

How often should a Claude configuration audit run?

The cadence is roughly 60 days, pinned to how often new models ship rather than how fast configuration rots. Model upgrades absorb capabilities that older scaffolding supplied, making frequent subtractive review worthwhile.