zirvDocsProject
Disclaimer
zirv is provided "AS IS", without warranty of any kind, express or implied, as stated in full in the MIT License. Use it at your own risk.
Autonomous supervision#
zirv ctx wrap, exec, loop, and the dashboard supervise a live coding
agent on your machine: at a verified-idle turn boundary they can type a
/compact or a restart into the session, and they can inject one labelled
[zirv ▸ mail] ... advisory line when mail arrives. exec/loop can also
kill and relaunch the underlying process with a distilled handoff. None of
this happens unless you run one of these commands (or the dashboard, which
uses them) -- running your harness directly bypasses it entirely. Individual
pieces can be turned off: mail's advisory injection, for example, stops with
[mail] enabled = false in ctx.toml. See Configuration
and Context Management for the rest.
Binary integrity#
Every release asset ships with a .sha256 checksum file; zirv update
downloads and verifies it before installing, and refuses to install an
unverified binary. Verify a manual download yourself before running it.
Third-party harnesses and repo-owned configuration#
zirv wraps third-party coding harnesses (Claude Code, Codex, and others) that
it does not control -- their own behavior, billing, and terms are between you
and their vendors. A checked-out repository's own zirv configuration,
system prompt, context files, memory, skills, and agents are untrusted input
by design: they can narrow what a session does but never widen a security
setting or grant themselves capability zirv itself did not already allow.
See the Trust boundary table for exactly what
each surface can and cannot do.