AI Agent Workflow Collection

Record a well-formed decision in cairn — context, decision, alternatives, consequences.

Claude Code plugin development reference for this monorepo (evonest, distill, solera, flutter-cask, pencilm3flutter). Covers plugin structure, manifest, hooks, skills, commands, agents, MCP servers, and marketplace.

Recognize breakthrough moments, blocking resolutions, and design decisions worth preserving. Detect high-value insights that save future time. Suggest distillation at valuable moments, not routine work.

Use evonest MCP tools to analyze, improve, and evolve a project — observe, plan, execute, verify. Triggers when the user asks to run evonest, analyze code, find improvements, execute a proposal, or improve a project with evonest.

Implements Riverpod (v3.3.x) compile-safe state management for Flutter apps. Use when implementing @riverpod code generation providers, handling AsyncValue loading/error/data states, using ref.watch/ref.read/ref.listen, setting up ProviderScope, writing Notifier/AsyncNotifier classes, using FutureProvider/StreamProvider, applying autoDispose and family modifiers, combining providers, testing providers in isolation, scoping rebuilds with Consumer instead of ConsumerWidget, using ref.watch with select() for field-level precision, migrating from StateNotifierProvider or Riverpod 2.x patterns, handling Notifier resource lifecycle with ref.onDispose, or using experimental Mutation and offline persistence APIs.

rag plugin initial setup (configure · start) — auto-creates the .noory/rag/ directory, the raw/ material store, settings.json, and the project .gitignore. A setup task that does not erase existing data. Example utterances "rag setup", "start rag", "rag config", "init-rag", "set up rag", "attach rag to the project".

Interview the user, list, inspect, select, reset, or override Plainly response styles for the active project. Use when the user asks to personalize how the AI writes, choose a built-in Plainly style, configure a project-owned external style file, or view the active style.

Adversarial code review for a Plot branch / commit set / PR. Reads code with red-team eyes — bad-faith input, hidden coupling, rotted comments, stale spec, principle violations against the CLAUDE.md trifecta + Plot's structural guards. Triggers on review / 리뷰 / 코드리뷰 / "review the code" / "check this branch" / "I just shipped" / "before I push" plus a Plot context. Use after code is written (not before — use plot-design-red-team for pre-implementation reviews).

Adversarial design review for a Plot proposal BEFORE implementation. Reads a SPEC change / DECISIONS entry draft / plan file / verbal proposal with red-team eyes — unstated invariants, failure modes, reversibility, VISION alignment, hidden tradeoffs, premature abstraction vs YAGNI. Triggers on design / 설계 / "review the plan" / "before I implement" / "look at this proposal" / "okay to build?" plus a Plot context. Use BEFORE code is written (use plot-code-red-team after).

Procedure for placing a new domain concept into Plot's bounded-context map before writing code. Forces a decision tree — is this a new entity (own id, own kind) or a value-object (embedded in another)? Which bounded context owns it (Discovery / Retention / Planning / Execution / AICollaboration)? Does the existing 15-kind palette already cover it? Triggers on "design", "도메인 설계", "어디다 둘까", "where does this go", "bounded context", "DDD", "entity vs value", "domain layer", "도메인 레이어". Prevents the v0.13.3-v0.13.10 cursor saga's root cause (no domain map → every fix picks an ad-hoc location → next bug surfaces elsewhere).

Procedure for adding a new domain entity (a new kind) to Plot. Walks through every file that must land in lock-step (viewer domain class, server Pydantic model, per-kind node renderer, per-kind inspector, i18n keys, registry updates, tests, schema-parity, structural guards). Triggers on phrases like "add a new kind", "엔티티 추가", "새 종류", "new entity", "kind 추가", "domain entity", "fromJson", "discriminated union extend". Prevents the v0.13 god SketchNode regression — every new kind must land as a class with fromJson boundary + per-kind UI files, never as a new field appended to the god union.

Implement a Plot feature (UI, MCP tool, schema, layout rule, etc.) end-to-end through the essence-aware TDD pipeline. Anchors every step back to VISION.md's three-phase cycle (Discovery / Retention / Execution) and DOMAIN.md's bounded contexts. Triggers on feature / 기능 추가 / 만들어줘 / add button / implement / 새 / 새로운 / build keywords plus a Plot context.

Diagnose Plot canvas UI bugs (cursor flicker, hover wrong, hit-test wrong, click not firing, button not visible) by running Playwright DOM probes BEFORE proposing any code change. Triggers on cursor / hover / pointer / hit-test / flicker / 깜빡 / button doesn't / 안 눌려 / click 안 됨 / 호버 / 손가락 / grab / crosshair keywords plus a Plot canvas context.

Static audit of viewer i18n compliance per feedbackplotglobalservice (Plot is a global service). Scans viewer/src//.{ts,tsx} for hardcoded user-facing strings, undefined t() keys, stale locale keys, and untranslated values. Triggers on i18n / 다국어 / 국제화 / "translation audit" / "locale check" / "i18n 검사" plus a Plot context.

Create a new Plot project (blueprint) in a chosen directory and open it. Asks WHERE to create it — built for monorepos where several app services (e.g. Banas + Banana) are designed side by side, each with its own .plot/ next to its code.

Create a new empty Plot sketch in the current project and open it in the browser.

Read the current Plot sketch and describe what is drawn — nodes, edges, and visible relationships.

Replace the app logo in the design guide .pen file. Triggers when the user asks to change, replace, or update the logo, app icon, or brand mark in the Pencil design guide.

Change the seed color of the Material Design 3 Pencil library. Triggers when the user asks to change the seed color, primary color, brand color, or update the color palette/scheme in material-design-guide.lib.pen.

Initialize a new app design guide .lib.pen file. Triggers when the user asks to initialize, create a new design guide, or set up a new app's design system in Pencil.

Add an indexing source — add it to settings.json sources and immediately do a partial index. Utterance examples "rag add source docs/", "rag-add-source notes/", "register wiki/ in rag".

Destructive reset/wipe that clears the rag index·graph·cache. settings.json is preserved (optionally delete it if desired). Example utterances "clear rag", "reset rag", "delete rag index", "rag-clear", "start over from scratch", "wipe rag".

Batch-evaluate rag index quality with pre-registered probes — search results + the active AI session's diagnosis and tuning suggestions. Example utterances "evaluate rag", "run the probes", "check whether indexing went well", "rag-evaluate", "measure search quality".

N-hop graph exploration around a specific entity — reports what is connected to what as a visual/table. Example utterances "explore around entity X", "rag-explore Riverpod", "explore around Notifier", "show me X's relations".