A collection of Claude Code plugins for enhanced development workflows.
Lightweight, native knowledge management for Claude Code projects. Three layers of persistent knowledge: Rules (always active), Knowledge (on demand), and Memory (automatic).
Commands: /init, /query, /prime, /curate, /reindex, /backfill-knowledge, /migrate, /statusline
Generic task and worktree workflow system. Manage tasks as markdown files, work in isolated git worktrees, and track progress through the full lifecycle. /kickoff runs the repo's default worker agent (Claude, codex, grok, kimi, or a PATH-detected cc-harness foreign agent — a single committed per-project default), or, when none is set, shows a picker and offers to save your choice; override per run with flags like --opus/--sol or --pick. Inside a herdr session it auto-opens a tab (named after the task, shortened for the sidebar and prefixed with the task's state glyph — ● active, ◇ in review, ◆ approved, ✓ merged) with the worktree as cwd, starts the chosen worker, and — for a Claude worker — runs /work-system:continue for you (plugin-qualified, since a Claude Code built-in /continue shadows the bare skill); /adopt auto-opens the same tab once it has built the worktree from an existing branch; /work-system:continue <task> from the main session reopens that tab and resumes it if a stray /exit closed it (for a non-Claude worker that resume is a new Claude session, and for a cc-harness worker it silently drops the foreign-model routing — /continue says so inline); and /close tears the tab down again when the task is merged — from inside a worktree it first offers to hand that whole close to the Manager session at the repo root, which then verifies the merge itself and closes the tab from the outside. /kickoff also records the lane's autonomy mandate (MANDATE.md: what the worker may do unasked, where it must stop, how many review rounds it gets), so /work-system:continue starts on the next unmet requirement instead of asking what to work on — and so does pr-flow when it opens the PR.
Commands: /define, /kickoff, /adopt, /continue, /status, /close, /list, /statusline
PR review feedback loop. Create PRs with readiness checks, commit + push + trigger @claude review, inspect status, work through review issues interactively, and merge safely with pre-merge documentation checks. Reads a work-system mandate when there is one (no second confirmation for an already-authorized step), and probes whether the repo has an @claude review bot at all. The probe can prove a bot but never its absence (the GitHub App needs no workflow file), so it answers yes / no / cannot-tell and routes to the local /swarm:review only where a bot is provably unreachable — otherwise it asks.
Commands: /open, /cycle, /check, /fix, /rebase, /merge
Local mixture-of-agents code review. Fans out one review across Claude lens subagents (11 lenses in 5 clusters — breakage, reach, threat, design, consistency) plus the codex and grok CLIs, and kimi on request (--kimi / SWARM_KIMI=1) — up to four independent model families (claude, openai, grok, moonshot). Every voice runs one call per gated cluster (Kimi only on the breakage and threat clusters — Moonshot meters it on 5-hour/7-day quotas, which is why it is opt-in), so the gate prunes work for all of them and each finding's lens tag is authoritative. The pipeline merges and verifies their findings, then presents one ranked report — defects and design suggestions kept apart — before anything is pushed. External voices read project files and research online under an OS secret-deny plus repository-write-deny sandbox with a hardened egress policy. Kimi uses ACP over stdio with a read-only shell command allowlist (anything else is rejected) and validates its prompt-obeyed JSON locally and fail-closed; every external voice keeps a shell for git history and grep under a jail that denies every write except to the scratch and per-user temp/cache dirs, /dev and the backend's own auth state. With --fix / --loop, only Claude applies agreed findings; external agents stay review-only. --pr [<number>] reviews a GitHub PR diff and posts the gated result after one confirmation. Complementary to PR Flow's GitHub-side loop. (0.11.x adds Kimi as an opt-in fourth family, gives every external voice a shell under an inverted write-deny jail, and retunes the effort profiles; 0.10.x added adapter-owned numeric config; 0.9.x added the reach cluster and grok discovery; 0.8.x added out-of-band prompts and telemetry.)
Commands: /swarm:review [--fix | --loop[=N]] [--max], /swarm:review --pr [<number>], /swarm:agents (planned: /swarm:adversarial, /swarm:style, /swarm:security — thin subset presets of the default lens set)
Plugin settings system: per-plugin TOML config resolved over schema defaults. Each plugin owns its config file (.work-system.toml, .knowledge-system.toml, .pr-flow.toml), defaults, and validation schema; users override only what they need. list, show, get, set, validate via one script and skill. Includes a [related_projects] sibling-project address book for cross-project orchestration. (Phase 1: config surface only — consumer wiring lands next.)
Commands: /settings (subcommands: list, show, get, set, unset, validate, defaults)
/plugin marketplace add gering/claude-plugins
/plugin install knowledge-system
/plugin install work-system
/plugin install pr-flow
/plugin install swarm
/plugin install settings
/reload-plugins
Plugins update automatically when the marketplace is refreshed. To manually update:
/plugin marketplace update gering-plugins
/reload-plugins
Structure invariants (JSON validity, version sync, skill frontmatter, internal references, shell syntax) are enforced by CI on every PR. Run the same checks locally before pushing:
python3 scripts/check-structure.py
Requires python3 and node. Node is needed by one plugin test
(plugins/swarm/scripts/test_voice_accounting.py) that executes the swarm
review workflow — the repository's only JavaScript — against synthetic inputs.
That test fails loudly rather than skipping when node is missing, because a test
guarding against silently lost review voices must not silently skip itself.
MIT