kagura-plugins

> One entry point for the Kagura Memory Cloud ecosystem — a memory substrate, a set of > atomic review Tools, and a couple of orchestration Harnesses, unified under the > kagura-ai brand.

kagura-plugins is a curated marketplace, not a code dump. It follows

reference-don't-vendor: every plugin lives in its own kagura-ai repo and is

referenced here. Adding this one marketplace gives you the whole ecosystem, while each plugin keeps a single canonical source.

The map

┌─ Tier 2 · Harness ── gh-issue-driven · kagura-engineer
│     multi-phase · stateful/resumable · drives loops · creates PRs · HITL gates · costly
│        consumes ↓
├─ Tier 1 · Tools ──── claude-c-suite · claude-phd-panel · kagura-cli · kagura-code-reviewer · kagura-planner
│     single-purpose · cheap · safe · ~stateless: invoke → result
│        grounded ↓
└─ Substrate ───────── kagura-memory  (Memory Cloud)
      session-start · session-summary · recall · remember · guide · smoke-test
      the shared memory everything grounds in

Read it bottom-up: memory grounds everything, tools give you a single sharp answer,

harnesses stand on top of the tools and drive a whole workflow.

Install

/plugin marketplace add kagura-ai/kagura-plugins
/plugin install kagura-memory@kagura-plugins        # start with the substrate
/plugin install kagura-code-reviewer@kagura-plugins # add the tools you want…

Update the catalog later with /plugin marketplace update.

Substrate — start here

| Plugin | Invoke | What it's for | |---|---|---| | kagura-memory (Memory Cloud) | /kagura-memory:session-start · :session-summary · :recall · :remember · :guide | Persistent memory that grounds the ecosystem. Bracket every working session: session-start resumes context in, session-summary saves knowledge out; recall/remember throughout. |

This is the connective tissue — most other plugins are more useful with memory underneath.

Tier 1 — Tools (atomic · cheap · safe)

| Plugin | Invoke | What it's for | |---|---|---| | kagura-cli | /kagura-cli:setup · :auth · :doctor · :ingest · :files · :resource | Thin skills over the Kagura Memory CLI (kagura) — auth/setup/doctor, document ingestion, resource tokens, R2 file uploads. The substrate's operational companion (distinct from the kagura-memory MCP plugin). | | kagura-code-reviewer | /kagura-code-reviewer:… | Cost-free, Ollama-first code review — multi-angle finders, adversarial verify, structured verdict. | | kagura-planner | /kagura-planner:plan | Memory-grounded PLAN layer. Thin wrapper over the kagura-planner CLI (config discovery → doctor → plan). | | claude-c-suite | /claude-c-suite:ask · :ceo · :cto · … | Executive-team review lenses for any codebase. ask routes to the single best CxO; ceo synthesizes across roles. | | claude-phd-panel | /claude-phd-panel:cs · :db · :sec · … | PhD-level deep technical critique (CS, DB, Distributed Systems, Data Science, PL, Security, Statistics). |

Tier 2 — Harnesses (multi-phase · stateful · costly)

> ⚠️ Harnesses are not atomic tools. They run loops, hold resumable state, mutate the > repo, open PRs, and gate on HITL, and can consume significant tokens. Reach for them > deliberately.

| Plugin | Invoke | What it's for | |---|---|---| | kagura-engineer | /kagura-engineer:run · :review · :goal | Autonomous coding loop over Claude Code + Kagura Memory. Drives a task/issue to a PR. | | gh-issue-driven | /gh-issue-driven:start · :ship · :goal · … | Drives GitHub issues to PRs (start → implement → ship → review) with gates and HITL. |

Brand & ownership

All plugins are first-party. The kagura-* tools live under the kagura-ai org;

claude-c-suite, claude-phd-panel, and gh-issue-driven are currently JFK-maintained (github.com/JFK/...) and the marketplace references them there — reference-don't-vendor works regardless of which org hosts the source, so these install today exactly like the rest.

Migration to kagura-ai is deferred, not cancelled. It is treated as two independent steps:

1. Ownership move (JFK → kagura-ai org) — cheap, reversible. 2. Prefix rename — if/when the move happens, the claude- tools would be renamed into the kagura- namespace. A breaking namespace change plus a config-path migration — the expensive part, and the reason it stays deferred.

Both deliver brand coherence but zero user-facing functionality, so they are gated on an external trigger (public launch / external contributors), not a self-imposed deadline. Until then, JFK-maintained is the supported state.

Rollout status

| Plugin | Tier | Status | |---|---|---| | kagura-memory | Substrate | ✅ live (kagura-ai/memory-cloud) | | kagura-cli | Tool | ✅ live (kagura-ai/kagura-memory-python-sdk) | | kagura-code-reviewer | Tool | ✅ live (kagura-ai/kagura-code-reviewer) | | kagura-engineer | Harness | ✅ live (kagura-ai/kagura-engineer) | | claude-c-suite | Tool | ✅ live (JFK/claude-c-suite-plugin) — migration deferred | | claude-phd-panel | Tool | ✅ live (JFK/claude-phd-panel-plugin) — migration deferred | | gh-issue-driven | Harness | ✅ live (JFK/gh-issue-driven) — migration deferred | | kagura-planner | Tool | ✅ live (kagura-ai/kagura-planner) |

All eight plugins install today.

Related — not Claude Code plugins

Some Kagura components are runnable products, not /plugin install-able plugins, so they are intentionally not listed in the marketplace above. They ship no .claude-plugin/ manifest; you install and run them from their own repos.

| Component | What it is | |---|---| | kagura-agent | Memory-backed autonomous agent on the Claude Agent SDK / kagura-brain — Docker security membrane, per-task credential leasing, capability graduation, and a Slack/Discord cockpit. Runnable, Apache-2.0. Run from kagura-ai/kagura-agent, not via /plugin install. |