This repository contains a Claude Code harness (configuration, prompts, workflows) rather than executable application code. Security considerations are different from traditional software:

When adopting this harness, be aware of:

• Replace ALL placeholders ({{TICKET_PREFIX}}, {{PROJECT_NAME}}, etc.) before use
• Never commit secrets in placeholder values
• Use environment variables for sensitive configuration

• Agent prompts define tool restrictions for each role
• Review .claude/agents/ before deployment
• Customize restrictions for your security requirements

• Skills in .claude/skills/ are loaded into Claude's context
• Review skill content for any patterns that could leak sensitive info
• Custom skills should not contain credentials or internal URLs

• Hooks in .claude/hooks-config.json execute shell commands
• Review all hook commands before enabling
• Test hooks in a sandboxed environment first

Please report:

• Prompt injection vectors in skills or commands
• Agent boundary bypasses that could escalate privileges
• Sensitive data exposure patterns in templates
• Hook command injection possibilities
• Documentation that encourages insecure practices

For sensitive security issues:

1. Do NOT open a public GitHub issue
2. Email: {{SECURITY_EMAIL}} (or your preferred contact)
3. Include:
   • Description of the vulnerability
   • Steps to reproduce
   • Potential impact assessment
   • Suggested fix (if any)

For low-severity issues:

• Open a GitHub issue with the security label
• Use the bug report template

• Accepted: We'll work on a fix, credit you in the changelog (unless you prefer anonymity), and coordinate disclosure timing with you.
• Declined: We'll explain why we don't consider it a security issue and suggest alternatives (e.g., documentation update, feature request).

cat .claude/agents/*.md

cat .claude/hooks-config.json

grep -r "password|secret|key|token" .claude/skills/### During Use

• Don't paste secrets into Claude conversations
• Review agent outputs before executing suggested commands
• Use the RTE agent for release-critical operations (has additional checks)
• Enable hooks for automatic guardrails

• Audit skill changes in code review
• Restrict who can modify .claude/ directory
• Log agent invocations if required for compliance
• Train team members on prompt security basics

• All files in .claude/ directory
• Agent definitions in AGENTS.md
• Workflow templates and patterns
• Documentation that could lead to insecure implementations

• Claude Code itself (report to Anthropic)
• Your project's application code
• Third-party integrations you add
• Issues in example/case study content (marked with 📚 EXAMPLE)

We appreciate responsible disclosure and will acknowledge security researchers who help improve this project (with permission).

This security policy follows the GitHub Security Policy Guidelines.

Last updated: December 2025 | Version 2.0