--- title: "CLI overview" description: "cadence is a read-only CLI built primarily for coding agents to use as evidence, not just for humans." url: "https://teamcadence.ai/docs/cli/overview/" --- ## Who it's for The `cadence` CLI is designed first for **coding agents**, not humans typing commands by hand. Claude Code, Codex, or any other agent working in your repo can shell out to `cadence` mid-task to look up your own past sessions, code reviews, and blame history as evidence before making a decision - that's also the foundation [Cadence Cue](/docs/cue/overview/) is built on. It's entirely **read-only** and **developer-scoped**: it never uploads sessions (the desktop daemon does that - see [Overview & tray](/docs/desktop/overview/)), and it only exposes your own data, not org- or team-wide insight. The one exception is `cadence cue build`, which writes a local memory artifact to your machine, and the redaction-hook commands, which install local config. ## Installing it The CLI ships as a sidecar binary bundled inside the desktop app rather than a separate download. Install it from the app's overflow menu → **Install CLI** (see [Installation & onboarding](/docs/installation/)). It shares your existing sign-in via `~/.cadence/cli/config.toml` - no separate login step. ```bash cadence auth status ``` ## `--json` everywhere Every command that returns data accepts a `--json` flag. Human output can be colorized, paged, or truncated for readability; `--json` output is stable, uncolored, and complete - that's the contract scripts and agents should rely on. ```bash cadence session list --repo your-org/your-repo --json ``` ## Giving an agent the CLI as a skill Rather than explaining the CLI to your coding agent in a prompt each time, install a ready-made instruction file that teaches it how to use `cadence` as evidence: ```bash cadence skill install --agent claude cadence skill install --agent codex --project ``` `cadence skill print` prints the same content to stdout if you'd rather wire it in yourself. ## Full command reference See the [command reference](/docs/cli/commands/) for every subcommand, its flags, and what it returns.
Docs/CLI overview
CLI overview
cadence is a read-only CLI built primarily for coding agents to use as evidence, not just for humans.
Who it’s for
The cadence CLI is designed first for coding agents, not humans typing commands by hand. Claude Code, Codex, or any other agent working in your repo can shell out to cadence mid-task to look up your own past sessions, code reviews, and blame history as evidence before making a decision - that’s also the foundation Cadence Cue is built on.
It’s entirely read-only and developer-scoped: it never uploads sessions (the desktop daemon does that - see Overview & tray), and it only exposes your own data, not org- or team-wide insight. The one exception is cadence cue build, which writes a local memory artifact to your machine, and the redaction-hook commands, which install local config.
Installing it
The CLI ships as a sidecar binary bundled inside the desktop app rather than a separate download. Install it from the app’s overflow menu → Install CLI (see Installation & onboarding). It shares your existing sign-in via ~/.cadence/cli/config.toml - no separate login step.
cadence auth status--json everywhere
Every command that returns data accepts a --json flag. Human output can be colorized, paged, or truncated for readability; --json output is stable, uncolored, and complete - that’s the contract scripts and agents should rely on.
cadence session list --repo your-org/your-repo --jsonGiving an agent the CLI as a skill
Rather than explaining the CLI to your coding agent in a prompt each time, install a ready-made instruction file that teaches it how to use cadence as evidence:
cadence skill install --agent claude
cadence skill install --agent codex --projectcadence skill print prints the same content to stdout if you’d rather wire it in yourself.
Full command reference
See the command reference for every subcommand, its flags, and what it returns.