slb

What problem does it solve?

SLB tackles the risk of executing destructive commands from AI agents by ensuring peer review and explicit approvals before any command runs, binding approvals to the exact command content.

Core Features & Use Cases

• Two-person review workflow: Requires approval from a second reviewer to authorize potentially dangerous commands.
• Command hash binding: Approvals attach to the exact command content via a SHA-256 hash to prevent tampering.
• Risk-tier classification: Commands categorized into CRITICAL, DANGEROUS, CAUTION, and SAFE with tailored approval rules.
• Client-side execution guidance: Commands run in the user’s environment with preserved credentials/context; a notary daemon verifies approvals.
• Auditable state and governance: Persistent logs and state for audit trails, patterns, and decision outcomes.
• Integrations and tooling: Agent-mail style notifications and optional UI/dashboard components for human reviewers.
• Frontmatter-driven entry: SKILL.md at the skill root provides title/description; optional resources load operational policies.

Quick Start

Initialize a project with SLB, start a review session, submit a dangerous command for review, and await a peer’s approval before execution.