Lavra plugin
A coding agent plugin that runs the lifecycle from idea to pull request and accumulates project knowledge between tasks.
- 53 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- Commands create commits, branches and pull requests, and autonomous modes repeat runs without your involvement.
- Project memory lives in git and is visible to everyone with repository access.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add plugins/lavraDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- Commands create commits, branches and pull requests, and autonomous modes repeat runs without your involvement.
- Project memory lives in git and is visible to everyone with repository access.
This is third-party code. Review the repository files before installing.
What it does
Lavra turns an agent into a team with memory. Four main commands run the pipeline: `/lavra-design` runs a brainstorm with scope sharpening, research by domain subagents, a phased plan and an adversarial plan review; `/lavra-work` implements tasks with a commit per task; `/lavra-qa` verifies a web app in a browser with screenshots; `/lavra-ship` rebases, runs tests, scans for secrets and debug leftovers, creates the pull request and closes tasks. Knowledge is captured during work and recalled at the start of a session: memory lives in `.lavra/memory/knowledge.jsonl` and is tracked by git, so a team accumulates experience together. There are also 9 supporting and 6 power user commands, including `/lavra-review` with 15 reviewer agents and `/lavra-work-ralph` with autonomous retries.
Who it is for. Developers and teams who want planning, review and shipping from an agent on a single scheme with shared project memory.
Good fit when
- A new feature goes through brainstorm, plan and plan review before the first line of code.
- An OAuth bug returns a month later and the agent should recall the earlier fix.
- Before release a web app is verified in a browser with screenshots of changed routes.
- A teammate hits the same bug and the repository memory is available to them.
Not a fit when
- You need one short edit without a plan or review.
- The team does not want agent memory kept in git.
- The host is not Claude Code, OpenCode, Gemini CLI or Cortex Code.
Example request
/lavra-design "users upload photos to listings", then /lavra-work and /lavra-ship.Limitations
The plugin works with Claude Code, OpenCode, Gemini CLI and Cortex Code and uses tasks in the beads format by Steve Yegge. Commands create commits and pull requests, and `/lavra-work-ralph` and `/lavra-work-teams` start autonomous retries and persistent workers, so check permissions and the branch beforehand. Memory in `.lavra/memory/knowledge.jsonl` ends up in git: do not store secrets or private details in it. The README mentions v0.7.7, the project moves fast and the command set may change.
How to disable. Run `uninstall.sh` from the clone or remove the plugin from the agent directory. The `.lavra` and `.beads` folders in the project are removed by hand.
FAQ
How do I install Lavra?
Run `npx @lavralabs/lavra@latest`. Alternatively clone the repository and run `./install.sh`, with `--opencode`, `--gemini` or `--cortex` for other hosts.
Which three commands are used most?
`/lavra-design` for planning, `/lavra-work` for implementation and `/lavra-ship` for the pull request. For web apps `/lavra-qa` goes in between.
Where is memory stored?
In `.lavra/memory/knowledge.jsonl`, a file tracked by git and available to the whole team.
How many commands are there?
Four pipeline, nine supporting and six power user commands. The full catalog is on the project site.
What does `/lavra-ship` do?
Rebases on main, runs tests, scans for secrets and debug leftovers, creates the pull request, closes tasks and pushes the backup.
Discussion
No questions or reviews yet.
Related
PluginsSkills and a CLI with a product graph: roles, states and failures checked before code, the live product after the build
A local browser UI to annotate agent plans and review diffs, then send the feedback back to the agent
Twelve standalone skills for development: an agent loop with planning and review, pull request handling, TDD and email triage.
A plugin makes the agent explain its code and, before a push, ask questions so you understand what you are merging.
Three skills for Claude Code, Cursor and OpenCode: an interview-driven PRD, task slices and a design doc before any code
ADLC Harness takes a task from intent to PR with one command, and the human approves only four key gates.
README in short
The README contrasts an agent without and with the plugin, shows three main commands and describes configuration in `.lavra/config/lavra.json`. Install runs through npx or `install.sh`.