Documentation · CLI reference

Every command, explained.

Aura is one Rust binary. The registry below is the whole surface: set up a repo, drive your agents, prove a change, and rewind a mistake. Everything the desktop app does, you can script from a shell.

$ curl -fsSL https://auravcs.com/install.sh | bashView on GitHub
Set up & inspect

Get started

$ aura enable

Graft Aura onto the current Git repository: installs the hooks, registers the MCP server, and starts the file watcher. Commit, push and branch exactly as before.

$ aura status

Show the semantic state: logic nodes tracked, the active agent session, the latest checkpoint, and whether strict mode is on.

$ aura config

Manage global settings: telemetry consent, strict mode, dev mode, API keys.

Options / args
set <key> <value>
$ aura doctor

Diagnose repo health: stuck sessions, orphaned snapshots, missing hooks, shadow-branch issues, and offer to fix them.

One engine, many brains

Drive every agent

$ aura handover

Generate a dense, token-optimized context block for cross-agent handoff: hand work from one agent to another without re-reading the repo.

Options / args
--agent <claude|gemini|codex>
$ aura loop run

Run the work-loop: agents claim unblocked tasks from the backlog, work them in dependency order, and prove each goal before it merges.

Options / args
[--jobs <n>]
$ aura crew

Lay a backlog out as a dependency graph and drive it: durable, team-visible, collision-aware tasks that survive the session.

Options / args
plan | run | status
$ aura radar

Read the live awareness plane: who is editing or just committed which symbol, so parallel agents never collide.

Options / args
show [--json]
Ship with a verdict

Prove & review

$ aura prove

Trace the logic paths in the AST graph and verify a behavioral goal: which paths are wired, partially wired, or missing.

Options / args
--goal "<description>"
$ aura pr-review

Semantic review with risk scoring: catches logical renames, layer violations, forbidden calls, and accidental deletions an agent slipped in.

Options / args
--base <branch> [--json]
$ aura review post

Post the semantic review back to the PR as AST-anchored inline comments, anchored to the exact symbol that changed.

Options / args
--base <branch>
$ aura goals list

Read the proof ledger: every goal, its runs, and the verdict tying each commit to the goal it was meant to deliver.

Options / args
[--json]
Undo & trace

Time machine & provenance

$ aura rewind

Surgically revert a single function or class to its last safe state at the AST level: zero merge conflicts, the good code untouched.

Options / args
<identifier> <file> [--amnesia]
$ aura explain

Trace any code back to its origin: git blame → checkpoint → agent session → conversation transcript. See exactly why it was written.

Options / args
<identifier> <file>
$ aura memory why

Ask the provenance-anchored memory for the reasoning behind a piece of code: code-grounded recall an agent can query before it edits.

Options / args
<query>
$ aura atlas

Browse a human-readable directory of every symbol in the repo: files, functions and classes, with the meaning of each.

Hand it to anyone

Portable proof

$ aura bundle export

Pack the signed why-and-proof for a range of work into a portable bundle a teammate, reviewer or auditor can verify without your machine.

Options / args
[--since <ref>]
$ aura bundle import

Bring a teammate's signed proof bundle into your repo and apply it at the AST level.

Options / args
<bundle>
$ aura attest

List and verify the cryptographically sealed genuine-records for the repo: provenance you can check, not take on trust.

Options / args
list | verify
The full manual · docs.auravcs.com

This page lists the commands. The manual explains them.

Everything above is the reference: what each command is and what it takes. The manual is the other half — twelve chapters on why semantic version control works the way it does, one page per topic, including the parts where Aura is the wrong tool.

What Aura is, and why it exists

The problem semantic version control solves, and what it does not try to be.

How semantic tracking works

What Aura records instead of line diffs, and how a function keeps its identity through a rename.

The commands, one page each

The long-form version of the reference above: every flag, every failure mode, worked examples.

Merging at the AST level

How two agents editing one file merge without a text conflict, and where that stops working.

Running a team server

The Mothership: what it stores, how a teammate joins it, and what happens offline.

Live sync, function by function

How a teammate’s change reaches you as a function rather than a file, and what is never sent.

Wiring Aura into your agents and tools

The MCP server, the coding agents that speak to it, and the CI and forge integrations.

Self-hosting, compliance and migration

Running Aura on your own infrastructure, including air-gapped, and moving an existing repo onto it.

Where Aura does not help

The honest limits chapter: the claims Aura does not make, and the projects it suits badly.

Security and cryptographic design

The threat model, what is signed, where data lives, and how to report a vulnerability.

It runs through your Git hooks

Most of Aura works through native Git hooks: pre-commit, commit-msg, post-commit. Capturing context, checking intent against the actual AST change, and sealing a record all happen automatically as you commit. You never run those by hand: keep using Git the way you always have, and Aura steps in only when a change doesn't match its stated intent.

Prefer a window over a prompt?

Every command here is wrapped in the Aura desktop app: the same engine, in a UI your whole team can use.