Skip to content

OnePageQuickStart.md

Latest commit

 

History

History
37 lines (30 loc) · 1.2 KB

OnePageQuickStart.md

File metadata and controls

37 lines (30 loc) · 1.2 KB

RippleLogic Guardrails v0.1 Quick Start (1 page)

If you are building a tool-using agent, implement these four steps.

Step 1: Classify the intended action

Before tools, produce:

  • capability class (filesystem_write, shell_exec, publish_share_upload, etc.)
  • data_sensitivity (public, personal, secrets, regulated)
  • blast_radius (low, medium, high)
  • reversibility (easy, hard, irreversible)

Step 2: Gate it (three decisions)

Return one of:

  • ALLOW
  • ALLOW_WITH_CONSTRAINTS
  • BLOCK

Apply: A) Rights floor hard-blocks B) Tail-risk constraints for high blast or irreversible actions C) Containment for shell/browser/credentials/payments D) Default unknowns to constraints + ask clarifying question

Step 3: Enforce constraints

If ALLOW_WITH_CONSTRAINTS:

  • ask for explicit confirmation if required
  • sandbox high-risk capabilities
  • preview changes before applying
  • run sensitive-data check before publishing

Step 4: Write an audit entry

If tools were used, append an audit JSON object to a JSONL log. If a high-risk request was blocked, log that too. Never log secrets.

Minimum claim you can make if you do the above: "Implements RippleLogic Guardrails v0.1 (portable spec)"