NikolasMarkou

ML Audit — Security skill for Claude Code

Security community

Claude code skill and notes for auditing ML solutions.

How to install ML Audit

This entry records only its repository, not the path inside it, so there is no exact command to give. Open NikolasMarkou/ml_audit and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What ML Audit does

Claude code skill and notes for auditing ML solutions.

Alternatives in Security

Pal — Multi-model AI integration — chat, debugging, code review, planning, security audit 11.3k ★
Skills Audit Report — Date: 2026-02-15 Auditor: Automated Skill Quality Audit Scope: Recently added skills in business-growth/, fina 5.3k ★
Om Vault Audit — Vault Audit 4.6k ★

README

ml-audit

State-machine driven Claude Code skill for auditing ML systems. Sibling of [iterative-planner](https://github.com/anthropics/iterative-planner), specialized for audit rather than build.

What it does

Given a target ML system, `ml-audit` runs a structured audit grounded in a library of failure-mode catalogs shipped with the skill (`src/references/ml_*_failure_modes.md`, `src/references/ml_pipeline_guide.md`). It produces a report with evidence-backed findings, calibrated severities, and actionable recommendations.

**The phases:**

INTAKE → SCOPE → EXPLORE → AUDIT → REVIEW → REPORT → CLOSE
                                 ↑__________|
                                  (iterate per domain)

**The output** (in the target repo):

audits/audit_2026-04-20_a3f1b2c9/
├── scope.md                  # Audit contract (system, boundary, access)
├── findings.md               # Index of findings
├── findings/*.md             # Per-domain findings (data, training, deployment, monitoring)
├── evidence/E-*.md           # Captured code/log/config excerpts
├── risks.md                  # Prioritized risk register
├── report.md                 # Executive artifact
├── recommendations.md        # REC-NNN action list
└── decisions.md              # Severity calls, pivots, scope decisions

Installation

Install as a Claude Code skill: place `src/` under your skills directory, or reference it via your Claude Code configuration.

The skill is self-contained in `src/`:

SKILL.md — the state machine and protocol
agents/ — 9 specialized sub-agents (orchestrator, explorer, 3 auditors, specialist, reviewer, reporter, archivist)
references/ — audit methodology, evidence standards, severity rubric, file formats, report style, and the seven failure-mode catalogs (ml_*_failure_modes.md, ml_pipeline_guide.md)
scripts/bootstrap.mjs — CLI to create/resume/close audits
scripts/validate-audit.mjs — protocol-compliance checker