MCP Workflow Audit
Audit Claude Code, MCP, and AI-agent workflows for permissions, consequential actions, idempotency, failure recovery, verification, and operator handoff.
SKILL.mdskills/mcp-workflow-audit/SKILL.md
GitHubMCP Workflow Audit
MCP Workflow Audit is a read-first operational audit workflow for Claude Code, Model Context Protocol integrations, and AI-agent automations. Use it when an agent workspace works inconsistently, has accumulated MCP configuration, performs consequential external writes, or needs a maintainable handoff before further automation is added. The workflow inventories instructions, MCP configuration, Skills, hooks, scheduled automation, package metadata, and credential boundaries; maps external writes and approval points; and reviews least privilege, bounded retries, idempotency, missing or conflicting data, partial-failure recovery, and post-action verification. It explicitly treats scanner findings as heuristics rather than proof, avoids destructive testing, and does not broaden permissions merely to make a workflow pass. The upstream MIT-licensed repository also provides a dependency-free static scanner, regression tests, workflow contracts, and synthetic appointment, trades-operations, and retail-pricing references.
Installation
Install or set up from the source-backed instructions:
git clone https://github.com/OssaBellator/claude-mcp-workflow-audit.git; run node audit.mjs /path/to/workspace or node report.mjs /path/to/workspace > audit-report.md