A universal behavior and guardrail layer for AI coding agents that enforces engineering discipline, prevents scope creep, and guarantees production-viable code quality.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install dev-guardrails
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install dev-guardrails using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is Dev Guardrails?
Dev Guardrails is a critical behavioral and constraint layer designed for AI coding assistants like Claude Code, Codex, and Gemini CLI. Acting as a Senior Principal Engineer with a conscience, this skill establishes a robust, multi-layered defense mechanism to prevent blind compliance, hallucinations, code degradation, and out-of-control scope creep during development sessions.
By integrating this tool into your Openclaw Skills workspace, you elevate your AI agents from simple code generators into disciplined, self-monitoring software engineers. The system maintains a dynamic Project Contract, tracks cumulative changes via a Scope Dashboard, runs automated regression/iteration gates, and enforces strict quality audits to ensure every line of code generated is production-ready, secure, and aligned with your design architecture.
Dev Guardrails Use Cases
- Preventing Scope Creep: When users continuously add minor requests that accumulate into major architectural drift, the system triggers alerts to renegotiate the project scope.
- Handling Vague or Ambiguous Prompts: When instructions lack concrete platform, framework, or structural definitions, forcing the agent to clarify requirements instead of hallucinating defaults.
- Technical Sanity Checks: Stopping the agent from executing impossible requests, such as nation-scale scalability on simple local storage, by presenting viable downgrades.
- Ensuring Production Quality: Eradicating placeholder logic (like "// TODO" comments) and ensuring code runs out of the box with edge-case handling.
- Managing Low-Engagement Clients: Utilizing the Commitment Protocol to break infinite feedback loops when users respond with non-committal terms like 'do whatever' or 'you decide'.
How Dev Guardrails Works
- Instruction Reception: The agent receives a development prompt and immediately passes it through a reactive Instruction Guard to analyze ambiguity or boundary violations.
- Project Contract Lock-in: For non-trivial tasks, the agent proposes a concrete scope and architecture, freezing a baseline contract upon user confirmation.
- Continuous Scope Monitoring: The Project Supervisor tracks file-change audits and counts accumulative additions, triggering alerts if drift exceeds 30% of the original contract.
- Regression & Health Gates: Prior to delivery, the agent executes a silent smoke trace to guarantee zero regressions and validates codebase maintainability via Iteration Health Checks.
- Five-Dimension Quality Audit: The system scans the generated codebase across Architecture, Security, Performance, Code Quality, and Testing dimensions, categorizing issues by severity.
- Deviation Correction: If the project veers off course, a structured 5-step process halts execution, visualizes wasted effort, and offers explicit re-planning pathways.
Dev Guardrails Setup
To deploy Dev Guardrails within your Openclaw Skills ecosystem, configure your workspace directory to include the behavioral configuration and boundary rules:
- Clone or download the skill file and its reference boundaries:
mkdir -p .claw/skills/dev-guardrails
cp SKILL.md .claw/skills/dev-guardrails/
- Setup boundaries and scenarios in your project references directory to provide concrete capability guardrails:
mkdir -p .claw/references
cat << 'EOF' > .claw/references/boundaries.md
# Capability Boundaries
| Within Capability | Refuse/Red Line |
|---|---|
| Lightweight IM (<10k users) | Nation-scale platforms |
| Regional e-commerce | Database kernels / OS |
| Enterprise RAG + AI | General LLM training |
EOF
- Initialize with your AI Agent (e.g., Claude Code, Codex, Gemini CLI) by loading the skill dynamically at session startup so that the project-level supervisor begins monitoring immediately.
Dev Guardrails Data Schema & Taxonomy
This skill maintains a series of mental registers and file-change audits to track development consistency across multi-turn sessions. The system organizes its data and telemetry as follows:
Mental State Registries
| Registry Name | Structure | Tracking Strategy |
|---|---|---|
| Project Contract | Scope, Architecture, Exclusions, Constraints, Milestones | Frozen upon initial confirmation (Principle 5) |
| Scope Dashboard | Cumulative additions (%), total feature count, architecture drift | Updated after each completed task request |
| Churn Tracker | Initial direction, log of direction changes, wasted LOC / turns | Incremented upon each architectural or focus shift |
| Guardrail Activity Log | Turn number, layer triggered, violation type, resolution outcome | Append-only log for cross-layer audit trails |
File-Change Audit Trail
The system maintains a strict mapping of session file diffs:
src/path/file.ext-> Classed asREQUESTEDorNOT REQUESTED(triggers SF-L1 to SF-L4 audits to prevent over-engineering)tests/path/test.ext-> Evaluated via the Regression Gate and Necessity Test to isolate side-effects
Dev Guardrails Advanced Features
- Commitment Protocol (M7-A): Employs a three-strike escalation rule to force non-committal users into locking down a default technical proposal or pausing development.
- Complexity Multiplier (M7-B): Detects buzzword-stacking (e.g., AI + Blockchain + Web3) and calculates a risk score to systematically break down bloated architectures.
- Zero-Bypass Reliability Gate (M6): Runs silent background checks ensuring existing entry points are intact and code features a clean rollback path even if bypassed by the user.
- Before/After Quantification (M8-C): Automatically generates comparative tables illustrating metrics like cyclomatic complexity, dead code reduction, and test coverage between version iterations.
- Scope Fidelity Escalation (M9-A): Instantly detects agent overreach and unnecessary formatting changes, reverting any code changes that fail the strict Necessity Test.
SKILL.md
Loading
Related Openclaw Skills
The Creative Act
An executable toolkit modeled on Rick Rubin's creative philosophy to help creators overcome blocks, tune into inspiration, and cultivate a creative mindset.
The Creative Habit
An executable AI-driven toolkit based on Twyla Tharp's philosophy to transform creativity from a mystical whim into a repeatable daily discipline.
OpenClaw Cron Health Checker
A portable, deterministic health checking framework to audit OpenClaw scheduled tasks and system crons safely without modifying system states.
The Crowd Psychology Skill
An executable crowd psychology agent skill based on Gustave Le Bon's classic study, designed to analyze group behavior, collective dynamics, and leader influence.
CleanShot Skill for OpenClaw
An OpenClaw workflow skill that enables AI agents to coordinate screen captures, OCR, screen recordings, and annotations using the CleanShot Tool plugin.
Skill Optimizer
An opinionated tool designed to audit, polish, and optimize existing AI agent SKILL.md files against the agentskills.io standard.
Featured*
