useai-pro/openclaw-skills-security
13 skills26.3k installs897 starsGitHub
Install
npx skills add https://github.com/useai-pro/openclaw-skills-securitySkills in this repo
1Skill VetterSecurity-first vetting for OpenClaw skills. Use before installing any skill from ClawHub, GitHub, or other sources. --- name: skill-vetter description: Security-first vetting for OpenClaw skills. Use before installing any skill from ClawHub, GitHub, or other sources. Checks for red flags, permission scope, and suspicious patterns. metadata: short-description: Run a legacy deep-vetting checklist before installing an OpenClaw skill from any source. why: Preserve a conservative review path for operators who want a manual-first audit flow. what: Provides a legacy pre-install security vetting module for skill review and comparison. how: Uses a structured red-flag checklist focused on permissions, patterns, and suspicious instructions. results: Produces a conservative manual review output for install-or-block decisions. version: 1.0.0 updated: '2026-03-10T03:42:30Z' jtbd-1: When I want a simple manual-first checklist to vet a skill before install. audit: kind: module author: useclawpro category: Security trust-score: 97 last-audited: '2026-02-01' permissions: file-read: true file-write: false network: false shell: false --- # Skill Vetter You are a security auditor for OpenClaw skills.20.6kinstalls2Skill Auditorskill-auditor is a comprehensive security auditor from useai-pro/openclaw-skills-security that developers run before adding any OpenClaw skill to a workspace. It follows a fixed six-step review protocol producing severity-based verdicts and a safe-run plan. Checks cover typosquatting in skill names, dangerous permission requests, prompt injection patterns, supply chain dependencies, and data exfiltration vectors in skill metadata and instructions. The skill outputs a structured SKILL AUDIT REPORT so agents and developers can block, remediate, or safely install third-party skills with documented risk. skill-auditor exists because over-privileged or malicious skills can compromise agent environments unchecked. Reach for it whenever a new community skill is proposed for install, especially from unfamiliar repositories. The short-description metadata states the goal plainly: vet any OpenClaw skill before install with structured review rather than trusting README marketing copy alone.658installs3Skill Guardskill-guard is a runtime security monitor for active OpenClaw skills in the useai-pro/openclaw-skills-security repository. During skill execution, skill-guard applies permission-boundary checks and suspicious-behavior signals across filesystem reads and writes, outbound network calls, and shell command use. When actual behavior drifts from declared permissions, skill-guard flags anomalous activity and produces monitoring findings with escalation signals before damage spreads. Developers reach for skill-guard when running third-party or custom OpenClaw skills that request shell, network, or filesystem access and need live enforcement instead of trusting static manifest declarations alone.542installs4Prompt Guardprompt-guard is a security skill from useai-pro/openclaw-skills-security that scans OpenClaw skill content, user prompts, and external text for hidden instructions that override host agent behavior. It uses pattern detection, input normalization, and severity tiers separating critical signals from warnings, then outputs injection findings with handling guidance before unsafe content executes. Developers reach for prompt-guard when importing third-party skills, ingesting user-provided documents into agents, or hardening OpenClaw pipelines against context manipulation and instruction hijacking.518installs5Credential Scannercredential-scanner is a security skill from useai-pro/openclaw-skills-security that scans a workspace for exposed credentials, API keys, tokens, and private key patterns before any OpenClaw skill receives file-read access. The module uses path-aware regex checks, skip rules, and sanitized reporting that lists findings with masked output instead of printing raw secrets. Metadata states the goal is reducing accidental credential exposure when untrusted or newly added skills inspect the filesystem. A developer reaches for credential-scanner immediately before enabling agent skills on repositories that may contain .env files, cloud keys, or JWT signing material.500installs6Dependency Auditordependency-auditor is a supply-chain security skill from useai-pro/openclaw-skills-security that reviews dependencies OpenClaw skills attempt to install across npm, pip, and Go ecosystems. It checks package identity, install hooks, recency, reputation, and vulnerability severity to catch typosquatting and malicious packages before they enter the environment. The skill produces dependency findings with install recommendations and explicit block conditions when risk thresholds are exceeded. Developers reach for dependency-auditor whenever an agent skill triggers npm install, pip install, or go get flows and automated vetting is required.476installs7Sandbox Guardsandbox-guard is version 1.0 agent skill in useai-pro/openclaw-skills-security that produces Docker sandbox profiles for running untrusted OpenClaw skills without manual container hardening each time. Metadata describes a sandbox configuration module covering filesystem isolation, network restrictions, privilege limits, and resource caps using pre-baked Docker patterns with explicit security flags and usage guidance. Developers reach for sandbox-guard when testing third-party agent skills, running community OpenClaw packages, or when agents need read-only versus read-write execution profiles with recommended docker run commands. The skill makes sandbox-first execution practical instead of relying on ad hoc --cap-drop flags. Catalog metadata records 402 installs. Outputs include recommended sandbox commands and profiles suited to read-only inspection or controlled read-write skill runs during local agent development. Agents can choose read-only profiles for inspection runs or controlled read-write mounts when a skill must persist artifacts, always documenting the security flags applied to each container invocation.456installs8Permission Auditorpermission-auditor is a useai-pro OpenClaw security skill at version 1.0.0 that analyzes declared skill permissions and explains exactly what each permission allows before developers grant access. The skill identifies over-privileged skills, runs permission-by-permission review, checks dangerous permission combinations, and produces least-privilege guidance with a recommended minimal access scope. Developers reach for permission-auditor when evaluating third-party or internal OpenClaw skills that request shell, network, filesystem, git, secrets, browser, or API access. Outputs include a permission fit assessment that maps declared access to actual task requirements so teams avoid blindly approving broad capability bundles. permission-auditor fits security-conscious agent builders and platform maintainers curating skill marketplaces. The skill audits permission metadata rather than executing penetration tests or scanning runtime behavior after deployment.450installs9Output Sanitizeroutput-sanitizer is an OpenClaw security skill providing a post-processing module that checks agent output for secrets, PII, and internal identifiers before operator display or logging. It uses pattern-based detection and masking rules rather than emitting raw sensitive values, producing sanitized operator-facing output with secrets and paths redacted. Developers reach for output-sanitizer when OpenClaw agents might echo credentials, personal data, or filesystem paths in otherwise useful responses. The skill prevents accidental leakage without blocking legitimate agent answers.447installs10Config Hardenerconfig-hardener is an OpenClaw configuration security auditor from useai-pro/openclaw-skills-security (version 1.0.0) that reviews four concrete surfaces: AGENTS.md behavioral rules, gateway authentication and rate limits, skill permission policies, and sandbox isolation settings. The skill applies a checklist tied to OpenClaw defaults, flags missing AGENTS.md as critical, and scores the setup out of 100 with severity-tagged findings. Developers reach for config-hardener when standing up OpenClaw on a shared machine, after following a permissive quick-start guide, or before installing untrusted skills. Output includes a prioritized hardening plan plus ready-to-review AGENTS.md and .openclaw/settings.json templates, with explicit user approval required before any writes.445installs11Setup Auditorsetup-auditor is a version 2.0.0 skill in useai-pro/openclaw-skills-security that audits an OpenClaw or coding-agent host through a five-question wizard and a four-step review protocol. Step 1 scans .env, PEM keys, and home-directory credential files for AWS, OpenAI, Anthropic, GitHub, and Slack token patterns with truncated output. Step 2 checks AGENTS.md presence, least-privilege defaults, and gateway hardening. Step 3 assesses Docker sandbox readiness and emits example run profiles with cap-drop and network-none flags. Step 4 inspects shell profiles, authorized_keys, crontab, and git hooks for persistence tampering. The skill outputs a SETUP AUDIT REPORT with READY, RISKY, or NOT_READY verdict and a fix checklist, optionally generating sandbox files under .openclaw/sandbox/ after user confirmation.444installs12Network Watchernetwork-watcher is an OpenClaw security audit skill from useai-pro/openclaw-skills-security that inspects outbound network requests made by agent skills. Developers invoke it when reviewing OpenClaw skill packages for data exfiltration, unauthorized API calls, tunneling patterns, and suspicious egress destinations hidden behind legitimate-looking network use. The skill applies endpoint scrutiny, exfiltration heuristics, and explicit safe-pattern checks to classify connections as allowed, suspicious, or blocked. Reach for network-watcher during OpenClaw skill review, pre-release security checks, or ongoing monitoring when agent tools request outbound network access.432installs13Incident Responderincident-responder is an OpenClaw security skill (version 1.0.0) that coordinates response when a malicious or over-privileged skill is installed in a coding-agent workspace. It defines four severity levels from SEV-1 active exfiltration through SEV-4 policy violations and runs a five-phase protocol: containment, investigation, credential rotation, recovery, and post-incident reporting. The playbook preserves malicious SKILL.md evidence, checks persistence in shell startup, SSH authorized_keys, crontab, git hooks, and cloud audit logs, and supplies a three-tier credential rotation checklist. Developers reach for incident-responder after typosquat installs, unexpected network requests, or .bashrc modifications triggered by agent skills. File-read and file-write permissions are enabled; network and shell access remain disabled within the skill itself.422installs