Claude Code plugin marketplace by mralabs.
# Add marketplace (once)
/plugin marketplace add mralabs/claude-plugins
# Install plugins you want
/plugin install devil-review@mralabs
/plugin install radar@mralabs
/plugin install commit@mralabs
/plugin install human-voice@mralabsThe devil is in the details.
Adversarial code review that finds the subtle bugs, race conditions, and violated invariants hiding in your diff.
/devil-review # auto-detect scope
/devil-review --scope working-tree # working tree only
/devil-review --scope branch # branch diff against main
/devil-review --pr 42 # review a GitHub PR
/devil-review --base develop # diff against specific ref
/devil-review concurrency handling # focus on specific area
/devil-review --reject 2,5 # reject prior findings by index, then re-reviewRelease radar for your repo.
Track competitors, upstream tools and dependencies — release tracking for changelogs you care about. Optional weekly GitHub Actions check keeps a rolling "Radar digest" issue up to date. Lives in its own repository; this marketplace catalogs it from there.
/radar init # creates .radar/ (git-tracked), proposes what to track
/radar # check → changelogs → analysis grounded in YOUR repo
/radar add <url> # track a new tool
/radar discover # find new tools in your categories
/radar deep <x> # research one tool in depth — tracked or notFast conventional commits via Haiku delegation.
Delegates commit message generation to a dedicated Haiku subagent running in a forked context, so the main conversation stays clean and the token cost stays low. Enforces the 11 Angular-convention commit types, respects repo-specific commit formats defined in CLAUDE.md, and refuses to commit files that look like secrets. Opt-in PR flow for solo work: commit → push → PR → squash merge in one command.
/commit # long form
/cc # short alias — same behavior
/cc fix typo in readme # optional hint — subagent still decides from diff
/cc --pr # commit, push, open a PR (branch auto-created on default branch)
/cc --merge # --pr + squash merge + branch cleanup (also: --m, --prm)Ghostwriting that doesn't read as AI.
Style guard loaded before drafting any text sent in the user's name — emails, PR descriptions, issue/review comments, chat messages. A compact rule sheet distilled from Wikipedia:Signs of AI writing, plus ghostwriting-specific tone/register/no-fabrication rules. Applies in whatever language the text is written. Invoke on demand or let it load automatically — the description model-invokes it when Claude is about to write as you; pair it with a CLAUDE.md directive for guaranteed loading.
/human-voice reply to this email: <paste>
/human-voice rewrite this sentence so it doesn't sound AI-written: <sentence>MIT