magician vs kumi
magician is a plugin bundle for AI coding tools published by Alexander-Tyagunov. Marketplaces we read that list it: 2 (Claude Code: claude-plugins-community, Community GitHub marketplaces). kumi is a plugin bundle for AI coding tools published by annavetech. Marketplaces we read that list it: 1 (Community GitHub marketplaces). As of our marketplace pass on 2026-10-17. This pair was picked by a reader, not by our comparison rule, so search engines are asked not to index the page.
Where each is published
| Marketplace |  magician |  kumi |
|---|---|---|
| Claude Code: claude-plugins-community | Listed Version: 2.0.1 Forks (GitHub): 0 Stars (GitHub): 14 Not marked · First seen |
Not listed |
| Community marketplace davepoon/buildwithclaude | Not listed | Listed Version: 1.5.0 Forks (GitHub): 0 Stars (GitHub): 1 Not marked · First seen |
Each figure is the marketplace's own and is never added up across marketplaces. A marketplace that publishes no figure shows none: no number is not zero.
The facts side by side
| Fact | magician | kumi |
|---|---|---|
| Marketplaces that list it (differs) | 2 (better) | 1 (worse) |
| Stars (GitHub) (differs) | 14 (better) | 1 (worse) |
| Forks (GitHub) | 0 | 0 |
| Last release (differs) | Not published | 2026-10-16 |
| Where it runs | Installs as a plugin bundle | Installs as a plugin bundle |
| Transport | Not stated | Not stated |
| Tools it declares | Not published | Not published |
| Environment variables it asks for | Not published | Not published |
| Hooks it runs (differs) | 10 | 12 |
| Parts of the bundle (differs) | 38 | 50 |
| Licence | MIT | MIT |
| Repository (differs) | github.com/alexander-tyagunov/magician | github.com/annavetech/kumi |
| Changes in the last 7 days (differs) | 2 | 3 |
| First seen (differs) | 2026-10-11 | 2026-10-15 |
Lines where the two differ have a tinted background and bold values. Where the data itself says which value is better (more marketplaces, a larger figure of the same marketplace, a fresher release, a verified mark on a marketplace that lists both), the better one is also marked green with ▲ and the worse one red with ▼. Where it runs, the transport, the tools, the settings, hooks, the bundle's parts, the licence and the week's changes are marked only as different: more or fewer is not better in itself.
What each says it does
magician
Comprehensive software development lifecycle plugin that takes task from the idea to merged PR autonomously. Core capabilities: 1. Dynamic Project Inspector On every session start, it scans your project files (package.json, go.mod, Cargo.toml, pom.xml, etc.) and auto-assembles targeted knowledge ("lore") for every detected technology — no manual stack selection needed. Polyglot stacks get full coverage automatically. 2. Full Autonomous SDLC (/manifest) One command drives the entire flow with only 4 human approval gates: * /conjure — design dialogue (with an ability to strictly use design) → approved spec * /blueprint — implementation plan with parallelism map * /portal — git worktree isolation * /orchestrate + /ward — parallel agents with TDD * /certify — tests + types + linter + browser verification * /scrutinize → /absorb — multi-agent code review + integration * /seal — PR creation, loops until merged 3. Self-Learning A Stop hook (chronicle-stop.sh) captures git diffs at session end, writes entries to a ch
kumi
A coordinated engineering team of named AI specialists that remembers your project between sessions. Call yui to route work, or call any specialist directly: Go (jaan, siim, mart), Angular (liis, kadi, tiiu), iOS (ren, shu, ryo), Python (eero, anu, ivo), React (noa, rui, aki), SQL (saku), NoSQL (remo), architecture for any stack (kai), process management (enn), and DevOps/infra (sora). Each ships as both a skill and a subagent. Bundled hooks capture finished work when a session ends and say at the next session start that saved work exists, and record per-session and per-agent token and time usage. Every specialist follows the same discipline: read context first, make the smallest correct change, verify.
How to run each
magician As a Claude Code plugin
/plugin marketplace add anthropics/claude-plugins-community
/plugin install magician@claude-communitykumi As a Claude Code plugin
/plugin marketplace add davepoon/buildwithclaude
/plugin install kumi@buildwithclaudeCommands built only from what the marketplaces declare; the full set for every client is on each card.
Tools each declares
Neither one's marketplaces publish a list of tools, so there is nothing to compare.
Settings each asks for
Neither one's marketplaces publish the settings it asks for.
What it can do: magician · What it can do: kumi · Compare two other plugins
Data comes from the public marketplace file https://github.com/anthropics/claude-plugins-community/blob/HEAD/.claude-plugin/marketplace.json. Data comes from public marketplace files in GitHub repositories of their authors. Figures come from our own regular passes over the marketplaces, last read on . We publish no contact details for plugin publishers.