It runs inside the Claude Code process. The side panel is only what the author's README names. Read the repo before you install, or run claude plugin validate on the plugin folder.
Version
0.22.0
Repo stars
7
Last change
2026-10-18
What it can reach
Draws in the session
Reaches the network or an outside service
Sees tool calls
Runs a command when you press
Calls another program on this machine
compost is one software-engineering workflow of skills: spec, slice, implement, build, verify, review, and finish, tracked in the repo's issue tracker. Run /compost:setup in each repo. With your yes it turns off the collections compost replaces, and the README says turning those off deletes nothing, while uninstalling an install its marketplace no longer lists cannot be undone. The pane shows the candidates, top recommendation first. Explore sends the pick as a message from compost, never as your own words, because the candidate names are model-written. Explore in new session starts claude --bg in read-only plan mode running compost:deepen on that candidate. /architecture-review reopens the last set. Many steps send typed questions to TypeSafe Jev (https://docs.typesafe.ai), with the key from TYPESAFE_API_KEY or the macOS Keychain item typesafe. Without a key the skill makes the call itself. The skill-drift mod sends one Jev request per loaded skill after each turn that loaded one, and the README says a request averages about 5,000 input tokens, a fraction of a cent. /drift lists the hits saved in ~/.claude/skill-drift/. compost:review opens a diff in a local page.
The install name is compost. The marketplace name is oliphant-plugins. The marketplace also holds mochi-creator, understand, and superfan. This one installs on its own. The cards of design candidates are why this page files it under review. pi-review hands a code review to a panel of read-only Pi reviewers, and draws their progress in a live pane. Nothing else on the shelf draws a pane of module-design candidates. In the source, skill drift posts to https://api.typesafe.ai/v1/systemone.
Demo
Screenshot of the author's GitHub page.
What it hooks into
session.startWhen a session starts
tool.callAround a tool call, where it can hold one
command.runWhen you run its slash command
ui.renderDraws into the sessionPane
skill.promptA Claude Code event
turn.completeWhen a turn ends
prompt.submitWhen you send a prompt
Read from the on() calls in its hook modules. Open its source under Files to check what each one does.
Hooks 1
Stopuv run --quiet "${CLAUDE_PLUGIN_ROOT}/scripts/stop_guard.py"
Files
The author's README and source, 6 files. They load when you open them.