Diplomat-agent static scanner reveals 76% of tool calls in 16 open-source AI agent repos lack security guards

A Python tool called diplomat-agent scans codebases for unguarded AI agent tool calls—functions that write to databases, send emails, charge cards, or delete data. A scan of 16 open-source agent repositories found 76% of tool calls had zero checks (input validation, rate limiting, auth checks, confirmation steps, idempotency keys, or retry bounds).

The scanner runs in ~2 seconds on a 1,000-file repo using only Python's standard library AST module. It detects 40+ patterns across 8 categories (database writes/deletes, HTTP writes, payments, email/messaging, agent invocations, destructive commands, and publish/upload operations) and outputs results in Terminal, JSON, SARIF 2.1.0, CSAF 2.0, or Markdown formats.

The tool integrates into CI pipelines (to block unguarded PRs), IDEs (Copilot Chat, Claude Code, Cursor), pre-commit hooks, and GitHub Code Scanning. It also generates a toolcalls.yaml behavioral SBOM that documents what an agent can do, comparable to a requirements.txt file.