Security fixes are applied to the main branch and shipped in the next release. Only the latest release is supported.

Do not open a public issue for security vulnerabilities.

Report privately using GitHub's private vulnerability reporting through this repository's Security → Report a vulnerability tab.

You should receive a response within a few days. Please include reproduction steps and the impact you believe the issue has.

agent reads and writes agent configuration, memory, MCP and skill files in your home directory. Its default auto synthesis mode sends the synthesized memory document to your configured Claude Code account, then Codex if Claude is unavailable or fails. Select --synthesizer deterministic to keep memory out of model calls. A custom AGENT_SYNC_SYNTHESIZER command receives the same document on standard input.

Memory files can contain private context. Treat the synthesized file, everything gather stages, and synthesizer input or output as sensitive. MCP registry specifications can contain environment values and HTTP headers; they are stored with private permissions, but agent mcp snippet prints them for manual pasting. Do not publish MCP state or snippet output.

Memory-pack archives are untrusted input. agent-sync rejects a selected pack subdirectory unless its resolved physical path remains inside the downloaded repository, and it never follows Markdown file symlinks. Skills are copied only into configured user-scope directories; unmanaged same-name collisions are refused rather than overwritten.