Self-PR — Self Principled Robotics

AI agent self-governance framework. Guardrails born from real mistakes, skills that mature through practice, and a feedback loop that makes agents safer over time.

What This Skill Provides

Guardrails Engine — Indexed rule checking (175k checks/sec), keyword + regex matching
Two-Tier Rules — HARD (absolute block) vs AUTH (needs owner approval)
Skill Runner — Step-by-step YAML skill execution with verify/retry
Execution Logger — SQLite history of all outcomes
Feedback Loop — Rule weight adjustment based on outcomes
Credential Scanner — Pattern detection for leaked secrets
Dashboard — Visual overview of rules, incidents, skills, architecture

Quick Start

Check an action against guardrails

uv run --with pyyaml python3 engine/guardrails.py check "deploy code to production"

View guardrail stats

uv run --with pyyaml python3 engine/guardrails.py stats

Run a compiled skill

uv run --with pyyaml python3 engine/skill_runner.py run <skill-name>

Launch the dashboard

cd dashboard && python3 -m http.server 8077

Starter Rules

Ships with common-sense rules. Agents add their own as they learn from mistakes — every guardrail should trace back to a real incident.

Skill Maturity Levels

Level	Name	Description
L0	Prose	Markdown docs, memory notes
L1	Recipe	Structured YAML pseudo-flow
L2	Script	Python/bash with error handling
L3	Native	Compiled, registered as tool

Architecture

Agent Action → Guardrails Check → Auth Gate → Execute Skill → Log Outcome → Feedback Loop
                 ↑                                                              ↓
                 └──────────── Rule weights adjusted ←──────────────────────────┘