How do I install the Model Audit skill?

Run npx skillmds@latest add nitzan94/model-audit in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

Is Model Audit safe to use?

This skill has not completed SkillMD's automated safety review yet. Capability flags: makes network calls. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with Model Audit?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is Model Audit free to use?

Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.

Who published Model Audit?

Nitzan94 (@nitzan94) published this skill. Their other Agent Skills are listed on their SkillMD profile.

Model Audit by nitzan94 — Agent Skill

Name: Model Audit
Rating: 4.2 (5 reviews)
Author: Nitzan94

Model Audit

Audit Agent/Task model routing — reads the dispatch log written by the model-routing PreToolUse hook and breaks down spend by model, subagent type, and project. Use when the user asks 'how's the model routing', 'audit my model usage', 'check the dispatch log', 'is the routing hook working', '/model-audit', or before drafting any public-facing post about Claude Code usage. Requires the model-routing hook to be installed first (see hooks/model-routing in this repo).

Nitzan94 Updated Sep 10, 2026

File contents

/model-audit — Routing Discipline Reader

Reads ~/.claude/agent_model_log.jsonl (written by the agent-model-check.py PreToolUse hook) and shows whether your subagent dispatches are actually using the model tiers you intended.

Arguments: "$ARGUMENTS" — either a number of days (7, 30) or the literal word all. Default: 7.

Step 1: Verify prerequisites

Check that the audit script and log file both exist:

test -f ~/.claude/scripts/model_audit.py && echo "script: ok" || echo "script: MISSING — install hooks/model-routing from this repo first"
test -f ~/.claude/agent_model_log.jsonl && echo "log: ok" || echo "log: MISSING — hook hasn't fired yet, dispatch any Agent in a fresh session first"

If either is missing, tell the user to install hooks/model-routing/ from the same repo (https://github.com/Nitzan94/claude-code-skills/tree/master/hooks/model-routing) — the skill is the reader; the hook is what writes the data it reads.

Step 2: Run the audit

Parse the user's arguments. If they said a number, pass it; if they said "all", pass all; if they said nothing, default to 7:

python3 ~/.claude/scripts/model_audit.py "${ARGS:-7}"

Step 3: Interpret the output

The script's output already includes a drift signal at the bottom. Read it and tell the user in one sentence what it means for their routing discipline:

Many (unset) entries → the hook is catching frequent drift attempts. Good — that's the hook earning its keep.
Opus/Sonnet ratio above ~20% → routing has slipped. Suggest scanning the By subagent_type x model section for which subagent types are running Opus when they shouldn't.
Zero dispatches in window → either a quiet week, or check whether the hook is still wired (grep agent-model-check ~/.claude/settings.json).

Don't add new analysis the script didn't produce. The script is the source of truth; the skill's job is to surface it and frame the interpretation.

---name: model-auditdescription: Audit Agent/Task model routing — reads the dispatch log written by the model-routing PreToolUse hook and breaks down spend by model, subagent type, and project. Use when the user asks 'how's the model routing', 'audit my model usage', 'check the dispatch log', 'is the routing hook working', '/model-audit', or before drafting any public-facing post about Claude Code usage. Requires the model-routing hook to be installed first (see hooks/model-routing in this repo).---# /model-audit — Routing Discipline ReaderReads `~/.claude/agent_model_log.jsonl` (written by the `agent-model-check.py` PreToolUse hook) and shows whether your subagent dispatches are actually using the model tiers you intended.**Arguments:** "$ARGUMENTS" — either a number of days (`7`, `30`) or the literal word `all`. Default: `7`.## Step 1: Verify prerequisitesCheck that the audit script and log file both exist:```bashtest -f ~/.claude/scripts/model_audit.py && echo "script: ok" || echo "script: MISSING — install hooks/model-routing from this repo first"test -f ~/.claude/agent_model_log.jsonl && echo "log: ok" || echo "log: MISSING — hook hasn't fired yet, dispatch any Agent in a fresh session first"```If either is missing, tell the user to install `hooks/model-routing/` from the same repo (https://github.com/Nitzan94/claude-code-skills/tree/master/hooks/model-routing) — the skill is the reader; the hook is what writes the data it reads.## Step 2: Run the auditParse the user's arguments. If they said a number, pass it; if they said "all", pass `all`; if they said nothing, default to `7`:```bashpython3 ~/.claude/scripts/model_audit.py "${ARGS:-7}"```## Step 3: Interpret the outputThe script's output already includes a drift signal at the bottom. Read it and tell the user in one sentence what it means for their routing discipline:- **Many `(unset)` entries** → the hook is catching frequent drift attempts. Good — that's the hook earning its keep.- **Opus/Sonnet ratio above ~20%** → routing has slipped. Suggest scanning the `By subagent_type x model` section for which subagent types are running Opus when they shouldn't.- **Zero dispatches in window** → either a quiet week, or check whether the hook is still wired (`grep agent-model-check ~/.claude/settings.json`).Don't add new analysis the script didn't produce. The script is the source of truth; the skill's job is to surface it and frame the interpretation.

Frequently asked questions