skills.rest
burningportra
burningportraCOMMUNITY
Explore all skills by this author →
GitHub

slb

Enforces two-person peer review for destructive shell commands from AI agents.

3|1|Updated Apr 15, 2026
Search Tags:#audit logging#peer review#command approval#ai agent safety#destructive commands#risk tiers#two-person rule
One-click install
npx skills add https://github.com/burningportra/agent-flywheel-plugin --skill slb-burningportra
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: slb
Source: https://github.com/burningportra/agent-flywheel-plugin/tree/main/skills/slb
Command: npx skills add https://github.com/burningportra/agent-flywheel-plugin --skill slb-burningportra

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve? AI coding agents can hallucinate or misunderstand context and run destructive commands like rm -rf, git push --force, or DROP TABLE without oversight. SLB applies a two-person rule so dangerous commands require peer approval before execution. ## Core Features & Use Cases - Risk-tier classification: Commands are categorized as CRITICAL, DANGEROUS, CAUTION, or SAFE, with approval requirements scaled to the tier. - Peer review workflow: Agents submit requests, reviewers approve or reject, and execution is gated by five verification checks including SHA-256 hash binding and approval TTL. - Claude Code hook integration: A PreToolUse hook automatically allows, prompts, or blocks commands based on their classified tier. - Use Case: An agent wants to run terraform destroy on staging infrastructure. SLB blocks it, routes the request to a second agent for review, and only executes after explicit approval with a full audit trail. ## Quick Start Ask the agent to run a risky command through slb run with a reason, then have a second agent session approve the pending request before it executes.

Frequently Asked Questions about slb

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I require approval before running destructive commands from an AI agent?▼

Use slb run with the command and a reason, which creates a request that blocks until a peer agent approves it. CRITICAL tier commands like terraform destroy need two or more approvals, while DANGEROUS tier commands need one.

How does SLB classify command risk tiers?▼

SLB parses commands with shell-aware tokenization, splits compound commands, and matches each segment against tiered patterns, with the highest-risk segment winning. If parsing fails, the tier is upgraded one level as a fail-safe.

Does SLB work with Claude Code hooks?▼

Yes, slb hook install adds a PreToolUse hook that classifies every command before execution. SAFE commands are allowed, CAUTION commands prompt the user, and DANGEROUS or CRITICAL commands are blocked until approved through slb request.

What happens if the SLB daemon is not running?▼

SLB falls back to file-based polling so requests still work without the daemon, though real-time notifications are unavailable. The hook fails closed, blocking dangerous commands when the daemon is unreachable.

Can a human bypass SLB in an emergency?▼

Yes, slb emergency-execute allows human override with a mandatory reason and extensive logging. Non-interactive use additionally requires acknowledging the SHA-256 hash of the command being executed.

Related Skills

Explore More Software Engineering Skills→
Dunc4nJDunc4nJ

slb

Coordinate two-person peer review for destructive commands in multi-agent workflows.

Community
Advanced
MassimilianoPiliMassimilianoPili

audit-manager

Aggregate JSONL audit event logs into human-readable Markdown reports.

Community
Intermediate
Kingly-AgencyKingly-Agency

bd-issue-tracking

Track multi-session issues with dependency graphs using beads.

Official
Advanced
ArthurZizumboArthurZizumbo

portal-terraform-gcp

Defines minimal GCP infrastructure with Terraform for Cloud Run, GCS, and Secret Manager.

Community
Intermediate