hoyeon

Sub-requirement spec-coverage reviewer for the execute verify pipeline. Checks whether a single sub_req's Given/When/Then contract is semantically satisfied by the submitted diff and cites the file:line that satisfies each of given, when, and then. Complements code-reviewer at gate=2 (code-reviewer asks "is the code correct?"; spec-coverage asks "does the code satisfy the spec?"). Returns a VerifyResult with verdict PASS or FAIL. Read-only — does not modify project files.

Produce a layered task DAG (L0 Foundation → L1 Feature → L2 Integration → L3 Deploy) where every sub-requirement is fulfilled by ≥1 task. Marks parallelism explicitly. Called by /blueprint Phase 2.

Extract technical requirements from Q&A log. Tech Lead perspective — ARCH/DATA/INFRA/DEPEND/COMPAT/SECURITY.

Evaluates proposed changes for risk (LOW/MED/HIGH), simpler alternatives, over-engineering, and generates structured decision_points for human approval on dangerous changes.

UX review agent that evaluates how proposed changes affect existing user experience. Focuses on simplicity, intuitiveness, and preventing UX regression. Runs early in the specify flow to ensure UX direction is solid before technical planning begins.

Value assessor for /tribunal skill. Evaluates positive impact, goal alignment, and strengths of the proposal. The constructive voice.

Explore the project's test infrastructure and establish a verification strategy for sub-requirements using the 2-axis model (machine-verifiable vs agent-verifiable)

Independent sub-requirement verifier. Executes verify_plan entries mechanically — no judgment, no bypass.

Assign verification gates (1=machine, 2=agent_semantic, 3=agent_e2e, 4=human) to every sub-requirement and every journey. Uses semantic reading of given/when/then — not keyword matching. Called by /blueprint Phase 4.

Implementation worker agent. Handles code writing, bug fixes, and test writing. Only works on tasks delegated by Orchestrator (/execute skill). Use this agent when you need to delegate implementation work during plan execution.

Behavioral guidelines to reduce common LLM coding mistakes. Use when writing, reviewing, or refactoring code to avoid overcomplication, make surgical changes, surface assumptions, and define verifiable success criteria.

Derive cross-module agreements (types, interfaces, invariants) from requirements.md. Writes a single `contracts.md` artifact (markdown, language-agnostic). Called by /blueprint Phase 1.

Root cause analysis specialist. Traces bugs backward through call stack, classifies bug type, recommends minimal fix. Read-only investigation only. Use this agent when a bug needs diagnosis before fixing.

Searches project internal docs (ADRs, READMEs, configs) for architecture decisions, team conventions, and constraints relevant to the current task.

Researches external libraries, frameworks, and best practices via web search and official docs. Use for migrations, new tech decisions, and unfamiliar APIs.

Feasibility checker for /tribunal skill. Evaluates whether the proposal is practically achievable given constraints. The pragmatic voice.

Gap analysis agent that identifies missing requirements, potential pitfalls, and explicit "must NOT do" items before plan generation. Inspired by Metis from oh-my-opencode.

Interview quality auditor that checks Q&A coverage and identifies gaps. Use after each axis interview round to determine if more questions are needed.

Git commit specialist. Enforces atomic commits, detects project style. Use this agent for ALL git commits during /execute workflow. Triggers: "commit", "git commit"

Extract interaction requirements from Q&A log. Interpreted per project-type (user/developer/operator/consumer experience) — JOURNEY/HAPPY/EDGE/STATE/FEEDBACK/ACCESS.

Socratic interviewer - questions only, no code, no implementation

Spec-driven QA verification agent. Reads sub-requirements (GWT format) from requirements.md/plan.json, determines the appropriate verification method for each (browser/CLI/desktop/shell), executes verification, and returns structured PASS/FAIL per sub-requirement. Does NOT fix code — report only. Used by verify-thorough Step 4.

Independent DoD verifier for /ralph skill. Runs in a separate context to eliminate self-verification bias. Reads the DoD checklist and independently checks each item against actual code, test results, and file state. Returns structured PASS/FAIL per item. Read-only — does not modify any project files.

Before starting any work, output a `CHARTER_CHECK` block. This is mandatory — no implementation or analysis begins without it.

Browser Explorer agent that controls the real Chrome browser via chromux (raw CDP). Parallel-safe — each agent gets its own isolated tab. Uses an isolated Chrome profile (logins persist across sessions, no bot detection). Use when you need to explore external web services (Crisp, Reddit, dashboards, etc.).

Extract business requirements from Q&A log. PM perspective — WHO/WHY/WHAT/SUCCESS/SCOPE/RISK.

Fast, read-only codebase search specialist. Finds files, code patterns, and relationships. Use for: "where is X?", "which files contain Y?", "how does X connect to Y?", "what changed recently in Z?", "find all usages of W". NOT for: documentation search (docs-researcher), external APIs (external-researcher), code implementation (worker).

Code reviewer that performs systematic cross-cutting review of complete diffs. Catches integration issues, hidden bugs, and design inconsistencies that per-task verification misses. Returns SHIP or NEEDS_FIXES verdict.

Codex-powered risk analyst for /tribunal skill. Finds vulnerabilities, failure modes, edge cases, and architectural risks. The adversarial voice.

Calls OpenAI Codex CLI to synthesize multiple analysis reports and provide big-picture strategic review. Cross-checks gap analysis, tradeoff analysis, verification planning, and external research for contradictions and blind spots.

Analyze an open-source project from What/Why perspective (not how-it's-implemented). Use when the user says "/analyze-oss", "분석해줘 이 오픈소스", "이 레포 뭐하는거야", "analyze this repo", "what does X do", "이거 왜 쓰는거야", "이 라이브러리 분석", provides a GitHub URL and wants understanding, or asks to deeply understand an OSS project's purpose, value, target users, and usage flow. Clones the repo to ~/opensource-analysis/<repo-name>/ (git pull if already exists), dispatches parallel subagents per analysis lens, then synthesizes a What/Why-focused report in chat. Supports optional user-specific follow-up questions.

"/blueprint", "blueprint", "task graph", "contract derivation", "execution plan", "plan tasks from requirements", "contract-first planning" Turn requirements.md into an executable blueprint (plan.json + contracts.md). Five phases: Contracts → Tasks → Journeys → Verify Plan → Commit. Sits between /specify and /execute. Scope-adaptive (greenfield → bugfix). Uses hoyeon-cli (plan.json only; requirements.md is read as-is via Read tool).

Recon-first browser automation. Orchestrator explores the site first via chromux, saves a guide file with insights, then delegates execution to browser-explorer agent. Use when: "/browser-work", "브라우저 작업", "사이트에서 해줘", "웹에서 해줘", "LinkedIn에서", "크롬으로", "browser task", "automate this site".

Root cause based one-shot bug fix. debugger diagnosis → requirements.md generation → /execute. /bugfix "error description" Full investigation pipeline: debugger + gap-analyzer + standard verify. QA suggestion after successful fix.

This skill should be used when the user wants to verify their changes before pushing, or update the project's rule checklists. Phase 1: validate changed files against .hoyeon/rules/ checklists and report PASS/WARN. Phase 2 (conditional): propose rule additions when unmatched patterns are detected. Essential before git push. Trigger phrases: "check", "checklist", "verify changes", "what did I miss", "pre-push check", "cascading changes", "any more changes needed", "update checklist", "update rules", "rules update", "체크", "체크리스트", "변경 확인", "빠뜨린 거 없나", "push 전 확인", "뭐 더 건드려야 해?", "연쇄 변경 확인", "규칙 갱신".

This skill should be used when the user says "/compound", "compound this", "document learnings", "save what we learned", or after completing a PR. Extracts knowledge from PR context and saves to docs/learnings/.

This skill should be used when the user says "/council", "council", "deliberate", "multi-perspective decision", "트레이드오프 분석", "위원회 소집", "여러 관점으로 검토", or wants deep multi-perspective deliberation with tradeoff mapping. Combines tribunal (structured adversarial review), agent-council (external LLM opinions), dev-scan (community sentiment), and step-back (meta-level review) into a unified decision-making committee. Uses Agent Teams for real peer-to-peer debate with iterative step-back moderation loop.

Deep web research skill using parallel subagents + chromux browser-explorer + Gemini. Spawns multiple WebSearch research agents AND browser-explorer agents (via chromux for JS-heavy/dynamic sites), plus a Gemini CLI deep research source, then synthesizes everything into a cited report. Uses WebSearch, WebFetch, chromux browser-explorer, and Gemini CLI. Invoke with /deep-research <topic>.

Collect diverse opinions on technical topics from developer communities. Use for "developer reactions", "community opinions" requests. Aggregates Reddit, HN, Dev.to, Lobsters, ProductHunt, etc.

"/discuss", "discuss this", "think with me", "is this a good idea?", "what do you think about", "problem definition", "explore this idea", "/discuss --scored", "interview me", "clarify requirements", "요구사항 정리", "인터뷰", "딥 인터뷰", "뭘 만들어야 할지 모르겠어", Korean triggers: "같이 생각해보자", "이거 어떻게 생각해?", "문제 정의", "이게 좋은 아이디어야?", "이거 맞아?", "요구사항이 불명확", "아이디어 구체화"

Plan-driven orchestrator. Reads plan.json (from /blueprint) or requirements.md, then dispatches workers to build the system. Use when: "/execute", "execute", "plan 실행", "blueprint 실행"

Google search via real Chrome browser (chromux). Use when the user asks to "search Google", "구글 검색", "구글에서 찾아줘", "find articles about", "search the web for", or needs web search results with full page content. Also trigger when: searching a specific site ("search dev.to for..."), finding recent articles/posts about a topic, extracting content from search results, or when WebSearch results are insufficient and real browser rendering is needed. Prefer this over WebSearch when the user wants site-specific search, time-filtered results, or full article body/comments extraction.

GitHub issue creation skill. Analyzes the entire codebase impact based on user request, then creates a structured issue with AI-verified/human-judgment-needed/caution sections. /issue "issue description" Trigger: "/issue", "이슈 만들어", "issue 만들자", "깃헙 이슈"

"/mirror", "mirror back", "echo back", "다시 설명해줘", "이해한 거 맞아?", "내가 뭘 원하는지 말해봐", "확인해줘", "paraphrase this", "너가 이해한 거 설명해봘", "what did I ask?"

Systematically QA test any application — web apps, native macOS apps, Electron apps, CLI tools, interactive REPLs, or anything on screen. Three modes: browser (chromux/CDP, fast, DOM-level), computer (MCP computer-use, screenshot + pixel clicks, any app), and cli (tmux, send-keys + capture-pane for interactive terminals). Auto-selects mode or accepts --browser / --computer / --cli override. Use when asked to "qa", "QA", "test this site", "test this app", "find bugs", "test and fix", "fix what's broken", "dogfood", "exploratory test", "bug hunt", "QA this app", "사이트 테스트", "앱 테스트", "브라우저 QA", "화면 보고 테스트해줘", "네이티브 앱 테스트", "screen test". Three tiers: Quick (critical/high only), Standard (+ medium), Exhaustive (+ cosmetic). Produces before/after health scores, fix evidence, and a ship-readiness summary.

Iterative task completion loop with Definition of Done verification. Combines the Ralph Wiggum technique (prompt re-injection via Stop hook) with DoD-based independent verification. Claude proposes DoD criteria, user confirms, then Claude works autonomously. Stop hook re-injects the original prompt and blocks exit until all DoD items are independently verified. "/ralph", "ralph loop", "ralph 루프", "반복 작업", "DoD 루프", "완료 검증 루프", "task loop", "keep going until done"

This skill should be used when the user asks to "find references", "참고할 만한 거", "similar implementation", "오픈소스 있나", "how others do this", "레퍼런스 찾아줘", or needs to find existing patterns (internal) and open-source examples (external) for implementing a feature.

Iterative rubric-based evaluation and self-improvement loop. Builds a scoring rubric interactively, evaluates an artifact with multiple models in parallel (Codex, Gemini, Claude), then autonomously improves the artifact one criterion at a time until a score threshold is met or circuit breaker fires. "/rulph", "rubric evaluate", "rubric score", "multi-model evaluate", "score and improve", "evaluate and iterate", "grade this", "루브릭 루프", "채점 루프", "자율 개선", "개선 루프", "루브릭 평가"

Greenfield project architecture + harness scaffolding for AI Agent productivity. Interview-driven decisions → requirements.md → execute. Produces: Code Structure (vertical slice exemplar), Test Infrastructure, Guard Rails, conditional extensions, AND Harness (CLAUDE.md with domain/team context, rules, skills, hooks). L2: architecture decisions, L3: harness setup, L4: requirements + harness decisions (tasks generated later by /execute into plan.json). Use when: "/scaffold", "scaffold", "new project", "set up project", "프로젝트 세팅", "초기 구조"

This skill should be used when the user asks to "analyze session", "evaluate skill execution", "check session logs", provides a session ID with a skill path, or wants to verify that a skill executed correctly in a past session. Post-hoc analysis of Claude Code sessions to validate skill/agent/hook behavior against SKILL.md specifications.

"/specify", "specify", "요구사항 정의", "requirements", "스펙 잡기", "뭘 만들어야 하는지", "기획 정리", "인터뷰해서 스펙" Turn a goal into structured requirements through systematic interview. Three phases: Interview → Extract → Cross-check. Writes requirements.md in the cli format (consumed by /blueprint).

One-shot perspective reset that surfaces blind spots mid-work. Scans what the user has been doing, generates one abstract reframing question, and runs 3 quick checks (scope drift, side effects, better approach) in under 10 lines. No dialogue, no code. Trigger phrases: "/stepback", "step back", "한발 물러서", "넓은 관점", "놓치는 거 없나", "방향 맞나", "zoom out", "큰 그림", "방향이 맞는 거야", "잠깐 멈춰", "지금 뭘 하고 있는 거야", "blind spot", "재확인"

---

This skill should be used when the user says "/tribunal", "tribunal", "review this", "3-way review", "risk-value-feasibility check", or wants multi-perspective adversarial review. Also triggered by: "트리뷰널", "리뷰 해줘", "3관점 리뷰", "위험성 검토". Runs 3 agents (Risk/Value/Feasibility) in parallel and synthesizes a verdict.

This skill should be used when the user says "/ultrawork", "ultrawork", or wants to run the full specify → execute pipeline automatically with a single command. Automated end-to-end workflow that chains specify and execute skills.

20 hooks across 7 events