Docs/Command reference

View as Markdown

Command reference

Every cadence CLI subcommand, its purpose, and its key flags.

All commands accept --json for stable, scriptable output. Human-readable output may be colorized or paged. See the CLI overview for installation and how agents are expected to use these.

Auth

Command Description
cadence auth status Shows current sign-in status.

Sessions

Command Description
cadence session list Lists your own sessions. Filter with --repo, --agent, --after, --before, --limit, --offset, --include-insignificant.
cadence session search <query> Full-text/semantic search across your sessions - metadata, summaries, transcripts, and touched files - and explains why each result matched. Filter with --repo, --file, --after, --before.
cadence session view <session-id|url|share-token> Views one session in detail. Accepts a private session ID, a private URL, or a shared-session token/URL. --full returns the complete transcript and normalized file list. --events-limit caps how many events come back otherwise.
cadence session export Bulk, author-free, server-redacted JSON export of session evidence for a repo - used to build Cue memory, or to feed other tooling. Requires --repo org/repo and --output <path>. --scope auto|user|organization controls whether the export covers just you or the whole org (org-wide export is currently allowlisted to select orgs). -k/--max-sessions caps how many sessions are included; --shared-session scopes the export around a specific shared session.

Reviews

Command Description
cadence review list Lists Cadence’s session-aware AI code-review feedback you have visibility into for a repo (--repo org/repo). Filter with --score, --after, --before.
cadence review view <review-id> Views one review’s detail.

Insights

Command Description
cadence insights Returns your own “My Work” maturity insight - the same coaching signal behind tune-ups.

Blame

Command Description
cadence blame <file> git blame, augmented with which Cadence session produced each line, colorized by recency. Supports the usual git-blame-style flags: --line-range, --ignore-whitespace/-w, -M/-C (move/copy detection), --ignore-rev, --first-parent.

Skills

Command Description
cadence skill install --agent <claude|opencode|codex> Installs an instruction file teaching that agent how to use cadence as evidence. --project scopes it to the current project instead of globally.
cadence skill print [--agent <agent>] Prints the same instruction content to stdout instead of installing it.

Redaction

Command Description
cadence redaction hook Runs the realtime secret-redaction hook (invoked by an agent integration, not typically run by hand).
cadence redaction install Installs the redaction hook/plugin for a given agent (--agent), optionally scoped with --project.
cadence redaction uninstall Removes a previously installed redaction hook.
cadence redaction status Shows whether redaction hooks are installed and active.

Cadence Cue

Command Description
cadence cue "<task>" Returns a short, evidence-backed starting guide for a task - relevant files, tests, and historical cautions drawn from your own session history. Abstains rather than guessing if it doesn’t have reliable evidence.
cadence cue build Builds or refreshes the local Cue memory artifact for the current repo. Options include --repo-path, --repo, --max-sessions, --dense-cap, --force, --output-dir. Downloads local embedding models on first run.
cadence cue browse Generates a local, interactive HTML graph browser of your project’s session history. --open launches it directly; --output and --memory-dir control where files are read/written.
cadence cue eval --eval-file <path> Development-only retrieval-quality evaluation against a fixture suite (--mode, --ablation, --output-dir).

See Cadence Cue for what Cue is and why it’s a separate, opt-in feature from the rest of the CLI.