メニュー
Security auditor for AI agent code. Scans agent prompts, skills, and code for security vulnerabilities, prompt injection risks, and unsafe patterns before deployment.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
| name | agentshield |
| description | Security auditor for AI agent code. Scans agent prompts, skills, and code for security vulnerabilities, prompt injection risks, and unsafe patterns before deployment. |
Cherry-picked from everything-claude-code (affaan-m/everything-claude-code v1.9.0)
Security-first auditing for AI agent code, skills, and prompts.
□ No prompt injection vulnerabilities (user input directly in system prompt)
□ No role-confusion patterns ("ignore previous instructions")
□ No data exfiltration paths (agent sending data to external URLs without auth)
□ No hardcoded credentials or API keys in prompts
□ Scope is clearly bounded (agent can't exceed its defined role)
□ No exec() or eval() on user input
□ No subprocess with shell=True on untrusted input
□ File access scoped to AI OS root only
□ No unrestricted network calls
□ Dependencies locked (pinned versions)
□ No PII logging
□ No sensitive data in plain-text files
□ Memory (Mem0) stores only non-sensitive summaries
□ Outputs sanitized before display
# Run AgentShield on a skill or agent file
agentshield audit ./skills/my-skill/SKILL.md
agentshield audit ./corp/agents/my-agent.md
agentshield scan-dir ./skills/ # Scan all skills
AgentShield Score: 87/100
✅ No prompt injection patterns
✅ No hardcoded secrets
⚠️ WARN: Subprocess call without validation (line 42)
✅ Data handling: SAFE
Decision: CONDITIONAL — fix warning before deploy
security-guide.md within this folder.Multi-domain skill reference collection covering ai-integration, databases, finance, frontend, google-workspace, and POS systems. Each domain contains reference docs, best practices, and agent skill blueprints.
Multi-source deep research using firecrawl and exa MCPs. Searches the web, synthesizes findings, and delivers cited reports with source attribution. Use when the user wants thorough research on any topic with evidence and citations.
Localized RAG and Notebook injection via MCP
Build applications using Azure App Configuration SDK for JavaScript (@azure/app-configuration). Use when working with configuration settings, feature flags, Key Vault references, dynamic refresh, or centralized configuration management.
Assess and migrate cross-cloud workloads to Azure with migration reports and code conversion guidance. Supports AWS, GCP, and other providers. WHEN: migrate Lambda to Azure Functions, migrate AWS to Azure, Lambda migration assessment, convert AWS serverless to Azure, migration readiness report, migrate from AWS, migrate from GCP, cross-cloud migration.
Master Go concurrency with goroutines, channels, sync primitives, and context. Use when building concurrent Go applications, implementing worker pools, or debugging race conditions.