Secure & Govern · Prompt
Sample a set of controls and test whether the evidence actually proves them, like an auditor would.
Needs an LLM client connected to ScalePad MCP — Claude Desktop, Cursor, Windsurf or similar. Without it the assistant cannot call the recipe's tools.
Act as an internal auditor for [CLIENT NAME]'s [FRAMEWORK] program. Select 15 controls, weighted toward high-risk and recently changed ones, and explain how you chose them. For each, review the mapped ControlMap evidence and decide: does it prove the control operated, for the right period, for the right systems? Mark Pass, Fail or Needs work with a short reason and what better evidence would look like. Read-only. How to work with the ControlMap API (ScalePad MCP): - Find the client with GET /controlmap/v1/clients/health and use its tenant ID (for example "dundmif") in every request. Some clients have no UUID. - Send the header Content-Type: application/json on every POST, PUT and PATCH, including searches. - Use page_size of 50 or less and follow next_cursor. The "fields" parameter is ignored, so responses are large. - Judge controls by "status", not the "compliant" flag. Objective lists include sub-objectives, so take totals from objective_summary. - Use framework-level scores and labels. The overall score_label on client health can be wrong. - Older policy sections may be URL-encoded twice: decode them and use the highest version of each section. - If a call returns 502 or fails to connect, wait a few seconds and retry once. If it still fails, tell me exactly which data is missing.
# Internal control testing
Hook: Sample a set of controls and test whether the evidence actually proves them, like an auditor would.
Mid-year, or as a dry run before an audit.
Test results by control with pass, fail or needs work.
Tested against the ControlMap API on October 16, 2026. Only asks for what the API supports.
