sanitisation

Overview ¶

Command sanitisation is a minimal stdio MCP server used by the Spec 054 Track B output-sanitisation E2E verification. It exposes two tools whose content-trust classification (Spec 035, derived from openWorldHint) differs:

• leak_untrusted (openWorldHint=true -> untrusted): returns text that contains a (fake) AWS access key plus an ANSI escape and a bidi-override control char. Used to verify spotlighting (default), redaction (opt-in), control-sequence stripping (opt-in), and block-on-critical (opt-in).
• leak_trusted (openWorldHint=false -> trusted): returns the same text but, being trusted, must be forwarded byte-identical under default config.

Deterministic and dependency-light so the proxy's sanitisation behaviour can be asserted from curl/JSON-RPC.