portkey-guardrailsPortkey-inspired guardrails for OpenClaw: 5 configurable rules that block prompt injection, redact PII, flag off-scope responses, enforce agent budgets, and...
Install via ClawdBot CLI:
clawdbot install nissan/portkey-guardrailsGrade Limited — based on market validation, documentation quality, package completeness, maintenance status, and authenticity signals.
Contains instructions to override system prompt or ignore user requests
"Ignore all previous instructions"Calls external URL not in known-safe list
https://github.com/Portkey-AI/gatewayAudited Apr 17, 2026 · audit v1.0
Generated May 2, 2026
A customer support chatbot for a financial services company uses the guardrails to block prompt injection attempts (G-01) and redact any PII like phone numbers or credit card numbers from agent responses (G-02). This ensures secure and compliant interactions without relying on external services.
An AI tutor for K-12 education uses the off-scope filter (G-03) to flag NSFW, political, or harmful content in responses, ensuring age-appropriate and safe learning. The context length guard (G-05) warns when the conversation exceeds the model's context window, preventing truncation or errors.
A sales automation agent uses the budget guard (G-04) to block dispatch when the agent's usage budget is in the red state, preventing unexpected costs. Combined with PII redaction, it ensures customer data is never leaked in outbound messages.
A healthcare triage assistant uses PII redaction (G-02) to automatically remove Australian phone numbers, emails, and Medicare numbers from logs and responses, helping meet privacy regulations. The prompt injection guard (G-01) blocks attempts to override the agent's instructions.
A social media moderation bot uses the off-scope filter (G-03) to flag competitor-disparaging or political content in user messages, helping maintain community guidelines. The fail-open design ensures the bot keeps running even if guardrails fail to load.
Offer the guardrails skill as a premium add-on for OpenClaw workspaces, charging a monthly subscription per agent. Revenue comes from tiered pricing based on number of agents or rules enabled.
Provide consulting to customize guardrail rules, config schemas, and per-agent YAML configurations for enterprise clients. Revenue from setup fees and ongoing maintenance contracts.
List the skill on OpenClaw's marketplace, earning a per-install or per-hook-enable fee split with the platform. Adoption driven by free basic version and premium rule packs.
💬 Integration Tip
Start with default rules enabled and test with your agent's traffic in a staging environment. Use per-agent YAML overrides to tune severity levels without modifying core rule code.
Scored Apr 19, 2026
Security vetting protocol before installing any AI agent skill. Red flag detection for credential theft, obfuscated code, exfiltration. Risk classification L...
Security-first skill vetting for AI agents. Use before installing any skill from ClawdHub, GitHub, or other sources. Checks for red flags, permission scope,...
Comprehensive security auditing for Clawdbot deployments. Scans for exposed credentials, open ports, weak configs, and vulnerabilities. Auto-fix mode included.
Audit codebases and infrastructure for security issues. Use when scanning dependencies for vulnerabilities, detecting hardcoded secrets, checking OWASP top 10 issues, verifying SSL/TLS, auditing file permissions, or reviewing code for injection and auth flaws.
Audit a user's current AI tool stack. Score each tool by ROI, identify redundancies, gaps, and upgrade opportunities. Produces a structured report with score...
Detect anomalies and outliers in construction data: unusual costs, schedule variances, productivity spikes. Statistical and ML-based detection methods.