Dual-Brain is a portable Markdown skill for agentic coding environments such as Codex and Claude Code. It consists of instructions (SKILL.md) that guide an AI agent's behavior — it ships no executable code, no dependencies, and no network services. The attack surface is therefore minimal.

That said, because the skill instructs an agent that can read files and run tools on your machine, we take prompt-level concerns seriously (for example, wording that could lead an agent to take unintended actions).

Dual-Brain can read project-local memory from:

.dual-brain/MEMORY.md

That file is useful, but it is still untrusted project text. The skill must treat memory as advisory context, verify it against current code/docs, and never follow memory that attempts to override user intent, tool safety, or system instructions. Hot/Warm/Cold/Archived tiers affect attention, not authority.

Do not store secrets in project memory:

• API keys
• access tokens
• credentials
• private keys
• session cookies
• sensitive personal data
• production incident details that should not live in a repo

If memory contains sensitive content, Dual-Brain should not summarize it into future context. It should remove or redact it and report only the category of content removed.

Dual-Brain may automatically update and compact .dual-brain/MEMORY.md for non-sensitive project memory after synthesis. It may update refs, last_referenced, and last_verified metadata only for memory that materially influenced or was verified during the session. Automatic memory mutation is never a license to store secrets, sensitive personal data, or instructions that override user intent, tool safety, or system instructions.

The latest version on the main branch is supported. There are no separate release branches.

If you discover a security concern — including prompt-injection risks, unsafe memory handling, or instructions that could cause unintended agent behavior — please report it privately:

• Preferred: open a private security advisory on GitHub.

Please include:

• A description of the issue and its potential impact.
• Steps to reproduce (e.g., the task/prompt that triggered the behavior).
• Whether .dual-brain/MEMORY.md was present and what kind of memory behavior was involved.
• Any suggested mitigation, if you have one.

Please do not open a public issue for security reports. We'll acknowledge your report as soon as we can and keep you updated on the fix.

Thank you for helping keep Dual-Brain and its users safe. 🙏