Foreman
Bad agents always blame their tools. Foreman doesn't care.
One skill for agent-tool safety. Foreman inspects tool ideas, schemas, MCP servers, OpenAPI actions, and pull requests. If the model can invent authority, cross tenants, send messages, move money, delete records, run code, or hide behind payload: any, work stops until the boundary is safe.
npx skills add blairhudson/foreman --skill foreman
Defensive Tool Design
The key to building better AI agents.
Foreman was built from this book. It takes the lessons and turns them into one practical skill you can run against tool ideas, schemas, MCP servers, workflows, and PRs.
The book shows how to put safety into the tool itself: purpose, schema, permissions, side effects, approvals, retries, audit trails, observability, and evals. Prompts can ask for care. Boundaries make care enforceable.