orient is an orientation plugin for AI coding agents (Claude Code, opencode, Codex). It runs on your machine and reads your repositories, so its security properties matter.

• No network. orient makes no HTTP requests, opens no sockets, and ships no MCP server. It cannot send your code or data anywhere.
• No telemetry. Nothing is logged, phoned home, or collected.
• Read-only survey agents. repo-cartographer and flow-tracer run with write and edit denied — they can read and analyze, never modify your code.
• Secret-blind. The survey and trace agents are instructed never to open secret-bearing files (.env*, *.pem, *.key, id_rsa*, credentials) and never to copy a secret value into any generated doc.
• Injection-resistant. The agents treat repository content as data to analyze, not as instructions to follow — a malicious file in a repo cannot redirect them.
• Nothing writes silently. init, sync, decide, and commit draft their output and wait for your approval before writing.
• Minimal footprint. The Claude Code session-start hook does nothing in a repo that hasn't been set up with orient, and always exits cleanly.

orient commits none of these, and its ignore rules exclude them defensively so a fork or a co-located project can't leak one by accident. For any repo:

• Add .env and .env.* to .gitignore; commit only a .env.example with placeholder values, never real ones.
• Load secrets from the environment at runtime; never hardcode them in source.
• If a real secret was ever committed: rotate it first (deleting the file does not un-leak it), then scrub history (git filter-repo or BFG), force-push, and check the provider's audit log for abuse during the exposure window.
• A committed .env is read by any agent that opens the repo — the ignore rule protects the secret from git and from the model.

Please report security issues privately rather than opening a public issue. Open a GitHub security advisory on the repository, or contact the maintainer directly. We'll acknowledge and work a fix before any public disclosure.