AVA-2568

decide-invest

Investment decision advisor — evaluate a potential investment, produce buy/hold/sell direction with confidence level and key uncertainties. Use when the user asks about investing decisions, stock picks, or portfolio allocation.

execute-git

Git operations — commit, push, branch, merge, and git workflow management. Use when the user needs git actions or version control operations. Related user-level skill: git-workflow (workflow conventions).

API & Backend Solid

api-design

Design a REST or GraphQL API contract before any code is written. Use when the user wants to define resources, endpoints, request/response schemas, auth, pagination, errors, idempotency, rate limits, versioning, or emit an OpenAPI/GraphQL SDL artifact. Triggers: "design an API for X", "draft an OpenAPI spec for Y", "define the resource model and endpoints", "spec the REST contract for Z", "model the API surface for W", "write the GraphQL schema".

API & Backend Solid

generate-api

Generate API implementations from contracts — route definitions, handler skeletons, request/response schemas, and error types. Use when the user asks to implement an API endpoint or service with a framework, after an API contract already exists from api-design.

handoff

Compact the current conversation into a handoff document for another agent to pick up and continue the work. Suggested skills, task summary, artifacts reference, and redacted sensitive info.

teach

Teach the user a new skill or concept, within this workspace.

Web & Frontend Solid

ui-design

Generate the visual and structural UI specification for a product, feature, page, or component from product goals and brand constraints. Use whenever the user asks for a UI design, mockup spec, screen layout, design tokens, component breakdown, state catalog, responsive rules, or accessibility treatment—even if the brief is just "design the screen for X," "build a UI for Y," "mock up the dashboard," "define the visual language," or "style the component." Covers information hierarchy, page layout, components, states, design tokens, color, typography, responsive behavior, accessibility, interaction notes, and handoff specs. Does not replace UX research, task flow, or information architecture—assume those are already decided or produced by `ux-design`.

writing-great-skills

Reference for writing and editing skills well — the vocabulary and principles that make a skill predictable.

retrieve-sql

Retrieve data from SQL databases — query construction, EXPLAIN verification, and safe execution. Use when the user needs data from relational databases or asks database-backed analytical questions.

comprehend-code

Understand and explain code — language detection, entry points, data flow, dependencies, and intent. Use when the user wants to understand code, review code logic, or get an architectural overview.

builder

Create a new skill on demand when Distributor detects a skill gap, or when the user wants to add a new capability to the library. Three internal sub-modules — confirm (clarify user intent) → plan (design skill structure) → generate (write SKILL.md + bundled resources). Loaded on demand by Distributor; manually invocable via `/builder`.

distributor

Route user tasks to the most appropriate skill via description matching plus risk scoring; detect skill gaps and trigger Builder. Always-active; the entry point of every task. Use when the user submits any task — even if they don't explicitly ask for routing.

domain-entry

Overview of the 5 LLM-capability categories in the domain layer (Understanding / Generation / Retrieval / Execution / Decision). Use when the user wants to browse the skill library, asks "what skills do you have", or is uncertain which category fits their need.

installer

Translate MY_SKILL's generic SKILL.md into platform-specific formats (workbuddy / codex / hermes) and bootstrap session activation. Two internal sub-modules — adapters/ (per-platform translation) + bootstrap/ (session activation). Manually invocable via `installer sync` and `installer install`.

review

Horizontal review layer intercepting every task. Two scopes — per-task (Code Review / Task Recheck / Security Review with Reflection Gate) and per-skill (Lifecycle Governance: review / merge / split / retire). Always-active; runs on every task and on every library lifecycle event. Cross-cutting concern, not a peer of on-demand skills.

decide-product

Product decision advisor — weigh tradeoffs between features, priorities, or product strategies. Use when the user faces product decisions, feature prioritization, or strategic product choices.

execute-bash

Execute shell commands, scripts, and automation tasks in the Bash environment. Use when the user needs to run commands, scripts, file operations, or system automation.

database-design

Design a safe, query-aware database schema from business entities, access patterns, consistency needs, and operational constraints. Use whenever the user asks for database design, schema design, table modeling, ERD, indexing, query optimization, data partitioning, multi-tenancy, migration planning, backups, or relational-versus-document tradeoffs—even if they only say "design the data model." Produce a reviewable schema specification before SQL execution; do not silently run migrations or modify a database.

generate-doc

Generate documentation — README, API docs, technical docs, changelogs, and usage guides. Use when the user needs new documentation created or template-based document generation.

system-design

Design a production-ready system architecture from product goals, workload, reliability, security, and operational constraints. Use whenever the user asks for system design, architecture, distributed systems, service decomposition, capacity planning, scalability, high availability, SLOs, data flow, migration strategy, or tradeoff analysis—even if they only say "design the backend" or "how should this system be structured." Produce an explicit architecture specification before implementation; do not silently jump to code.

Web & Frontend Listed

ux-design

Design user experience flows, information architecture, task journeys, and interaction behavior from user goals and business constraints. Use whenever the user asks for UX design, user flow, journey map, information architecture, onboarding, task flow, interaction design, edge cases, error recovery, usability assumptions, or usability validation—even if they only say "make this easier to use" or "design the experience." Produce a testable UX specification before visual styling; do not silently replace research with assumptions.

retrieve-rag

Retrieve information from personal knowledge sources (notes, docs, wikis, knowledge bases) using RAG-style retrieval. Use when the user asks questions about their personal notes, saved documents, or knowledge base.

comprehend-doc

Understand and explain documents — technical docs, specifications, manuals, reports. Use when the user wants to understand a document, extract key information, or get a structured summary.

caveman

思考 + 输出双层精简——让 AI 在思考阶段就只列核心要点（不展开不写过渡），输出阶段再兜底压一遍。Distributor 在 pre-think（thinking-first 之后）和 post-output 两个时点强制触发。Triggers: "简短点", "caveman mode", "少废话", "token 压缩", "少说废话"。即使没显式要求，对长思考链 + 长输出也默认应用。

grill-with-docs

设计拷问 + ADR/glossary 落地——Distributor 检测到模糊任务或设计请求时强制触发。Relentless 拷问逐个收窄设计分支，同时落地 ADR 和术语表。Triggers: "帮我设计 X", "规划 Y", "做项目 Z", 或任何 description match_score 低于阈值的模糊任务。user-invocable: false。