Please report security issues privately via GitHub Security Advisories, not as a public issue.

Lychi is solo-maintained, so response times are best-effort — but security reports go to the front of the queue. Please give me a reasonable window to ship a fix before disclosing publicly.

Only the latest release is supported. Lychi is pre-1.0 and moves quickly.

Lychi runs shell commands, opens files and URLs, and stores API keys, so the areas most worth scrutiny are:

• Command execution — anything that reaches a shell without passing through the rules engine (crates/lychi-core/src/rules/). Every execution path is supposed to go through one of the deciders in rules/shell.rs, path.rs, or uri.rs; a route that bypasses them is a bug even if it looks harmless.
• Injection through untrusted input — quicklink templates, script commands, AI-generated plans, @file references. Values are escaped per destination; a case where a value escapes into the template is in scope.
• Path traversal — file operations that escape their intended directory.
• Secret handling — API keys live in the system keyring. Any path that writes one to disk, logs it, or sends it somewhere unintended is in scope.
• Privacy — Lychi is local-first. Anything leaving the machine without explicit user opt-in is a security issue, not just a bug.

• AI-suggested commands being wrong or dangerous. AI proposes; the user confirms; the rules engine gates. That's the design. A bypass of that confirmation is in scope — a bad suggestion that the user then approves is not.
• A user deliberately running a destructive command. Lychi is a launcher; running things is the point. It warns on destructive patterns and refuses a denylist, but it isn't a sandbox.
• Compromised local environment. If an attacker can already write to ~/.config/lychi/scripts/ or your PATH, they don't need Lychi.