The `we` plugin

The we plugin is the tooling layer of Agentic Product Ownership. It adds a family of /we:* skills to Claude Code that cover the full chain — the four altitudes, the Build pipeline, and the quality gates around them. One install, in your own repo.

Install

/we:setup    # detect stack + ticketing tool, set up project vision and the crew

/we:setup runs once per project. It sniffs your stack (Python, Node, Rust, Go), finds your ticketing tool, and optionally initializes a vision doc and the repo crew. After that you are working.

The skill catalogue

# Plan altitudes (Solo) + Council
/we:vision /we:saga /we:epic /we:story    # formulate one artifact per altitude
/we:council /we:meet                       # multi-lens deliberation + decomposition

# Build
/we:build        # autonomous pipeline: develop → verify → gates → PR → CI
/we:ci-review    # collect and fix CI / review findings, then push

# Quality gates (standalone)
/we:review /we:static /we:test /we:pr /we:docs

# Advisors and memory
/we:coach        # where am I, what's the next sensible move?
/we:retro /we:handoff

There is more — /we:audit, /we:smoketest, /we:find-dead-code — but those are the spine.

Jira and GitHub

The plugin never hardcodes one ticketing tool. Skills speak in generic actions — “create ticket”, “move to In Review” — and map them to whatever you have: Jira (via MCP), GitHub Issues (via the gh CLI), or plan-only mode when there is no tracker at all. PRs go through GitHub; the CI-review step pulls findings from CI and Claude review.

The companion upgrade

The plugin works fully on its own. With a weside account it gets a memory. A backing MCP server connects your Companion — persistent project memory across sessions, vision held as Companion goals, context that survives a /clear. The skills become companion-aware: they remember the decisions you made, not just the spec you wrote.

See the Claude Code plugin product page for distribution and the CLI for the command-line surface.