Help us improve
Share bugs, ideas, or general feedback.
From agent-guardrails
Mandatory safety laws for AI coding agents: read before editing, stay in scope, verify before committing, and halt when uncertain. Enforces safe and reliable code modifications.
npx claudepluginhub thearchitectit/agent-guardrails-templateHow this skill is triggered — by the user, by Claude, or both
Slash command
/agent-guardrails:four-lawsThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
These four laws are MANDATORY and NON-NEGOTIABLE for all AI agent operations.
Enforces the Four Laws of Agent Safety on all operations, halting when uncertain. Ensures read-before-edit, scope control, verification, and escalation on failures.
Enforces Karpathy guidelines to prevent LLM coding errors: read before writing, surgical changes only, verify assumptions, define success upfront. Use for feature implementation, code modifications, or scope discipline.
Use when implementing any substantial feature, multi-file modification, or architectural change - produces a plain-language walkthrough of every alteration so the developer can verify genuine understanding before committing, preventing the accumulation of cognitive debt where code ships faster than comprehension
Share bugs, ideas, or general feedback.
These four laws are MANDATORY and NON-NEGOTIABLE for all AI agent operations.
NEVER modify code without reading it first.
Read tool to view file contents before any editOnly touch files explicitly authorized.
Test and check all changes.
Ask for clarification instead of guessing.
These laws apply to ALL operations:
When running in pi, these laws are enforced automatically by the @architectit/pi-guardrails extension:
guardrail_verify_read blocks edits to unread filesguardrail_check_scope blocks out-of-scope editsguardrail_check_halt evaluates commit safetyguardrail_record_attempt/guardrail_check_strikes enforce Three Strikes; injection defense blocks prompt attacksSee [[guardrails-core]] for the full enforcement coverage map.
Apply the Four Laws of Agent Safety to the current operation. Evaluate whether any law is at risk of being violated, enforce compliance, and halt if necessary.
Full documentation: docs/AGENT_GUARDRAILS.md