dong3
A Claude Code plugin marketplace by caasi.
Skills
compose
Describe multi-step agent workflows using an Arrow-style DSL and validate them structurally with the ocaml-compose-dsl binary. The DSL is a planning language — the agent expands it into concrete tool calls.
See plugins/compose/skills/compose/README.md for full documentation.
constraint
Natural language constraints → deterministic test artifacts. Write constraints in structured Markdown (constraints/*.md with Given/When/Then/Unless/Examples/Properties), generate unit tests / PBT / mutation tests, and enforce with a 4-layer deterministic pipeline.
dataflow
Explore an unfamiliar codebase together with the agent, one step at a time. The agent follows one value through the code, reads the types along its path, marks each finding [read] or [spike], and ends each answer with where to read next.
See plugins/dataflow/skills/dataflow/README.md for full documentation.
kami
Reflect on human-AI stewardship through Socratic dialogue. A mirror, not a checklist — rooted in Audrey Tang's Humane Intelligence framework and Civic AI 6-Pack of Care.
See plugins/kami/skills/kami/README.md for full documentation.
old-react
FP-thinking review and refactor for pre-RSC React (classes, hooks, Redux/MobX/observable, Reselect, Immer). Ships only architectural rules where neither TypeScript nor eslint-plugin-react-hooks already enforces the principle. Five active categories: state architecture (model-), effects (effect-), composition (compose-), render purity (purity-), and hook applicability (hooks-). Two categories (immutable-, message-) defer to existing tooling. One slash command: /old-react [review|refactor] [path].
See plugins/old-react/skills/old-react/README.md and SKILL.md for the canonical rule list.
review-loop
Assisted (not autonomous) multi-reviewer convergence loop for any change — code or design artifacts. Local reviewers answer blind and in parallel on the same unfixed diff (a Claude subagent always, plus any models/CLIs you enrol per host — Codex via headless codex exec when present), then GitHub Copilot for PR targets. The verdict names which reviewers actually ran; heterogeneity is a bonus, not a gate, and same-family review is never treated as a downgrade. Classifies comments into tiers, auto-fixes the mechanical ones, pauses for your judgment on the architectural ones, and never merges on its own. It publishes its stop shape before round 1 and holds to it, and a T3 that comes back after a fix makes that round's message a question about the whole change's direction as well as asking for the item's own fix. How a finding is settled follows who reads the artifact: a behaviour claim about a runtime prompt is settled by five independent simulation runs, never by a string comparison. Two slash commands: /review-loop [PR# | branch | blank] and /review-loop:init (enrol the roster for this host).
See plugins/review-loop/skills/review-loop/README.md for full documentation.
tsugu
A git-native skill for unattended work preparation and human–agent convergence. Using git's DAG as the coordination substrate, an agent prepares engineering work privately on git branches (often while you are away), records evidence in committed .tsugu/ notes, and packages a convergence packet; when you return you converge — review the prepared evidence, decide together what becomes public, and complete the disposition in the same session (cut a handoff branch / open a PR, human-gated — the merge lands when you approve it). Three routines: init (set up or migrate the repo's .tsugu/ workspace + policy.md), prepare (private git work on the configured work prefixes — defaults prepare/* investigate/* review/* — + tests + evidence while you are absent), converge (present the packet, decide with you, and complete the disposition + a read-only morning status view). State is derived from refs and the DAG, not stored in status fields, so merge commits are recommended; .tsugu/ MAY land on the mainline by default (public-branch-tsugu: include|exclude). Never auto-merges; light and script-free; invokes no user-installed skill by default (per-repo policy.md may opt-in). Three slash commands: /tsugu:init, /tsugu:prepare, /tsugu:converge.
See plugins/tsugu/skills/tsugu/README.md for full documentation.
Install
claude plugin marketplace add caasi/dong3
License
MIT
