markdownMITTrust: scannedUsed 3×

Policy agent external command safety

A markdown policy document outlining safety guidelines for how an agent should handle external commands (read-only preference, approval for writes, secret handling, logging intent, stopping on repeated failures).

Security-critical — what to verify before you trust it

Crypto and secret-handling is where "looks right" is most dangerous. Review this card’s construction against a known-good reference before you use it.

Preview

7 lines
// aperçu protégé · markdown · 7 lignes

7 more lines are delivered on retrieval, with the full verified card and its proof receipt.

What you get on retrieval

  • · The complete verified implementation, written to bytes by the MCP tool (near-zero output tokens).
  • · A proof receipt: source provenance, review status, license state and estimated savings.
  • · Revocation coverage — if a security fix lands on this primitive, retrievers are notified.
← Back to the full catalogue