By minjun0219
OpenAPI / MySQL cache-first toolkit + agent journal + spec-pact lifecycle (Claude Code plugin, Bun runtime). 15 tools: openapi_* / mysql_* / journal_* / spec_pact_fragment.
Spec-lifecycle sub-agent. Owns the project-local SPEC layer that lives between a Notion 기획문서 and the code. Conducts the `spec-pact` skill end-to-end: DRAFT (Notion → 합의 → SPEC write + INDEX 갱신), VERIFY (SPEC 의 `합의 TODO` / `API 의존성` 체크리스트화 후 caller 응답 수집), DRIFT-CHECK (SPEC frontmatter `source_content_hash` vs `notion_get(pageId).entry.contentHash` 비교), AMEND (drift 항목별 keep/update/reject → SPEC patch + version bump + INDEX 갱신). An LLM-wiki-inspired entry point lives at `<spec.dir>/<spec.indexFile>` (resolved from `agent-toolkit.json`, default `.agent/specs/INDEX.md` — concept borrowed from [Karpathy's LLM wiki](https://gist.github.com/karpathy/442a6bf555914893e9891c11519de94f) — wiki TOC + per-page wiki bodies + dedup by source — not a 1:1 implementation). SPEC bodies live at `<spec.dir>/<slug>.md` (slug mode, default `.agent/specs/<slug>.md`) or `**/SPEC.md` (directory-scoped, AGENTS.md style). Auto-trigger when a Notion URL / page id appears together with phrases like "스펙 합의" / "SPEC 작성" / "SPEC 검증" / "SPEC drift" / "기획문서 변경 반영". Single finalize/lock authority — even when negotiation is delegated to an external sub-agent / skill, only grace writes SPEC frontmatter and INDEX.
PR review watch sub-agent. Watches an existing GitHub PR (already created externally by the user / Claude Code / `gh` CLI) for new review comments, reviews, check-run signals, and merge events through polling. For each new comment, validates it against the codebase with `read` / `glob` / `grep` only, drafts a Korean reply or counter-argument, and posts the reply through the external GitHub MCP server. On merge / close, unsubscribes the watch automatically. Conducts the `pr-review-watch` skill end-to-end (WATCH-START / PULL / VALIDATE / WATCH-STOP). Single finalize authority over `pr_event_resolved` journal entries — even when validation reasoning was delegated to another sub-agent / skill, only mindy writes the resolved entry. Auto-trigger only when a PR URL / `owner/repo#123` handle appears together with explicit review-watch phrases like "PR review" / "리뷰 봐줘" / "리뷰 확인" / "코멘트 확인" / "머지까지 watch" / "리뷰 답글" / "PR drift"; a bare PR link must not start watch. The toolkit never calls the GitHub API itself — credentials live with the external GitHub MCP. mindy never edits code, never runs tests / typecheck / lint, and never merges the PR — those return to the caller.
Work partner with frontend specialty and fullstack range. Primary conductor of the agent-toolkit — wraps the `notion-context`, `openapi-client`, and `mysql-query` skills and the `notion_*` / `openapi_*` / `mysql_*` tools, owns the `journal_*` tools directly (no separate skill), routes the SPEC-합의 lifecycle to `@grace` (the `spec-pact` skill) without running the four modes itself, and routes only explicit PR review watch lifecycle requests to `@mindy` (the `pr-review-watch` skill) without running the four modes itself. When the work exceeds the toolkit, Rocky may delegate to external sub-agents / skills and pass their output through. Any input that mentions a Notion URL / page id, an OpenAPI / Swagger spec URL, a 16-hex spec cache key, an `agent-toolkit.json` `host:env:spec` handle, an `agent-toolkit.json` `host:env:db` MySQL handle, a GitHub PR URL / `owner/repo#NUMBER` handle (generic GitHub surface only — not enough to start PR watch by itself), or phrases like "스펙 정리해줘" / "요구사항 뽑아줘" / "긴 문서에서 작업만 뽑아줘" / "기능 단위로 쪼개줘" / "이 페이지 뭐라고 했지" / "이 endpoint 호출 코드" / "POST /pets axios 로" / "`acme:dev:users` 의 …" / "users 테이블 조회" / "schema 보여줘" / "SELECT … FROM …" / "스펙 합의" / "SPEC 작성" / "SPEC 검증" / "SPEC drift" / "기획문서 변경 반영" / "PR review" / "리뷰 봐줘" / "코멘트 확인" / "머지까지 watch" / "리뷰 답글" / "PR drift" must route here. Output is one of: cached markdown (context mode), Notion chunk/action extraction, Korean-language spec (notion-context spec mode), TypeScript `fetch` / `axios` snippet (openapi-client mode), markdown table of MySQL rows / schema (mysql-query mode), `@grace` sub-agent result (spec-pact 4-mode lifecycle) passed through, `@mindy` sub-agent result (pr-review-watch 4-mode lifecycle) passed through, or another sub-agent / skill result passed through. When a generic primary agent (e.g. OmO Sisyphus, Superpowers — synergy when present) shares the environment, that agent brings OSS / patterns / libraries while Rocky brings the toolkit and the user's working context — toolkit-shaped or context-shaped lookups should route to `@rocky`.
Inspect a MySQL database under a strict read-only policy — list tables, dump a table's CREATE / index detail, or run a single SELECT / SHOW / DESCRIBE / EXPLAIN with an enforced LIMIT. Auto-trigger when the user supplies a `host:env:db` handle (registered in `agent-toolkit.json` 의 `mysql.connections`) or phrases like "users 테이블 조회" / "schema 보여줘" / "users 컬럼 뭐 있더라" / "SELECT id FROM users where status='active'". Writes are always rejected — INSERT / UPDATE / DELETE / DDL / SET / CALL / LOAD / multi-statement / INTO OUTFILE 모두 거부한다.
Read a Notion page under a cache-first policy and either feed the markdown straight to the LLM as grounding, extract long-page chunks/action candidates, or rewrite it into a Korean-language spec ("문서 요약 / 요구사항 / 화면 단위 / API 의존성 / TODO / 확인 필요 사항"). Auto-trigger when the user supplies a Notion URL or page id together with phrases like "스펙 정리해줘" / "요구사항 뽑아줘" / "Notion 페이지 X 가 Y 에 대해 뭐라고 하는지" / "긴 문서에서 작업만 뽑아줘".
Read a cached OpenAPI / Swagger spec under a cache-first policy and emit a `fetch` or `axios` call snippet (TypeScript) for one endpoint. Auto-trigger when the user supplies an OpenAPI spec URL, a 16-hex cache key, or a `host:env:spec` registry handle (configured in `agent-toolkit.json`) together with phrases like "이 endpoint 호출 코드 만들어줘" / "axios 로 작성해줘" / "fetch snippet 줘" / "POST /pets 호출 코드 / acme:dev:users 의 …".
Watch an existing GitHub PR for new review comments / reviews / check-runs / merge signals through polling, validate each comment against the codebase, draft a Korean reply or counter-argument, post the reply through the external GitHub MCP server, and unsubscribe on merge / close. PR creation and the actual merge stay outside the toolkit (Claude Code / `gh` CLI / external GitHub MCP own those). Conducted by the `mindy` sub-agent. The toolkit never calls the GitHub API itself — credentials live with the external MCP. Auto-trigger only when a PR URL / `owner/repo#123` handle appears together with explicit review-watch phrases like "PR review" / "리뷰 봐줘" / "리뷰 확인" / "코멘트 확인" / "머지까지 watch" / "리뷰 답글" / "PR drift"; a bare PR link must not start watch.
Negotiate, anchor, verify, and amend a project-local SPEC against a Notion 기획문서. Four-mode lifecycle (DRAFT / VERIFY / DRIFT-CHECK / AMEND) on top of an LLM-wiki-inspired INDEX (`<spec.dir>/<spec.indexFile>`, default `.agent/specs/INDEX.md`) and per-page SPEC files (`<spec.dir>/<slug>.md` slug mode, default `.agent/specs/<slug>.md`, or `**/SPEC.md` directory mode). Conducted by the `grace` sub-agent. Auto-trigger when a Notion URL / page id appears together with phrases like "스펙 합의" / "SPEC 작성" / "SPEC 검증" / "SPEC drift" / "기획문서 변경 반영".
Admin access level
Server config contains admin-level keywords
External network access
Connects to servers outside your machine
Own this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimOwn this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimBased on adoption, maintenance, documentation, and repository signals. Not a security audit or endorsement.
Uses power tools
Uses Bash, Write, or Edit tools
Uses power tools
Uses Bash, Write, or Edit tools
Claude Code (1차) + opencode (2차) 듀얼 host MCP / 플러그인 toolkit. OpenAPI 캐시 / MySQL read-only 검사 / 저널 / SPEC 합의 lifecycle / PR 리뷰 watch / SPEC → GitHub Issue 동기화 를 turn 단위 컨텍스트로 묶어 준다. 일하면서 자주 쓰는 "기획문서 → 합의된 SPEC → 이슈 → PR 리뷰 → 머지" 흐름을 한국어 / 한 사용자 환경에서 매끄럽게 굴리는 게 목적.
개인 프로젝트라 유지보수가 꾸준하지 않을 수 있다.
FEATURES.md (한국어) — 도구 / 스킬 / 에이전트 / 설정 / Quick start / 검증 한 페이지.AGENTS.md (영문) — Layout / MVP scope / coding rules / change checklist.bun install..mcp.json 의 agent-toolkit (스토디오 MCP, bun run server/index.ts) + context7 두 서버를 처음 로드할 때 trust prompt 가 한 번씩 뜬다 — 둘 다 승인.CLAUDE.md 가 자동 로드되어 AGENTS.md 로 위임된다.openapi_envs, mysql_envs, journal_status, spec_pact_fragment 중 아무거나.Claude Code 진입점은 15 tool 만 노출 — openapi_* ×5 + journal_* ×4 + mysql_* ×5 + spec_pact_fragment ×1. 빠진 10 tool (notion_* ×4, pr_* ×6) 은 AGENTS.md 의 MVP scope → Removal candidates 절에 추적된다.
{ "plugin": ["agent-toolkit@git+https://github.com/minjun0219/agent-toolkit.git"] }
opencode 진입점은 25 tool 모두 노출. 자세한 절차는 .opencode/INSTALL.md.
도구 한 개씩의 입출력, 스킬 / 에이전트 권한, 환경변수, agent-toolkit.json 스펙은 모두 FEATURES.md 에 있다. 장기 비전은 ROADMAP.md.
bun install
bun run check # Biome 검증
bun run typecheck # tsc --noEmit
bun test # lib/ + .opencode/plugins/ + server/ 단위 테스트
npx claudepluginhub minjun0219/agent-toolkitA growing collection of Claude-compatible academic workflow bundles. Covers scientific figures, manuscript writing and polishing, reviewer assessment, citation retrieval, data availability, paper reading, literature search, response letters, paper-to-PPTX conversion, and evidence-grounded Chinese invention patent drafting. Rules are organized as reusable skill folders with explicit workflows and quality checks.
Intelligent draw.io diagramming plugin with AI-powered diagram generation, multi-platform embedding (GitHub, Confluence, Azure DevOps, Notion, Teams, Harness), conditional formatting, live data binding, and MCP server integration for programmatic diagram creation and management.
Persistent file-based planning for AI coding agents. Crash-proof markdown plans (task_plan.md, findings.md, progress.md) that survive context loss and /clear, with an opt-in completion gate and multi-agent shared state. Manus-style. Works with Claude Code, Codex CLI, Cursor, Kiro, OpenCode and 60+ agents via the SKILL.md standard. Includes Arabic, German, Spanish, and Chinese (Simplified and Traditional).
Harness-native ECC plugin for engineering teams - 67 agents, 271 skills, 92 legacy command shims, reusable hooks, rules, MCP conventions, and operator workflows for Claude Code plus adjacent agent harnesses
Reliable automation, in-depth debugging, and performance analysis in Chrome using Chrome DevTools and Puppeteer
v9.44.1 — Patch release for Gemini environment/version detection and qwen auth gating. Run /octo:setup.