Recently added
Plugins grouped by the day they were added to the directory, newest first.
Sep 18, 2026
- team81stars
A live browser console for Claude Code agent teams — every teammate's transcript, tasks, mail and status in one window. Starts itself the first time a team is spawned.
Claude Code7 skills - workflow-zozeei0stars
ชุดสกิลภาษาไทยสำหรับวางแผน พัฒนาซอฟต์แวร์ ตรวจ Query Performance และตรวจสอบความปลอดภัย (N+1, Index, Execution Plan, ValidatesSafeInput, OWASP Web 2025 & Mobile 2024)
Claude Code2 skills - dev-skills0stars
面向中文开发者的软件工程 Agent Skills，覆盖方案澄清、规格与任务拆分、TDD、系统化调试、代码评审、并行实现和完成验证。
CodexClaude CodeAgent Plugins38 skills - nestor0stars
Use the canonical Nestor journal for tasks, notes, reminders, projects, history, and activity time tracking.
CodexClaude Code4 skills1 MCP server - nestor-beta0stars
Nestor skills under test, before they are promoted into the nestor plugin.
CodexClaude Code4 skills - massdo-skills0stars
Response-style skills: short ASD-STE100 answers, articulate replies with an optional word cap, and a chief-of-staff stance.
CodexClaude Code3 skills - mattpocock-skills-full0stars
Matt Pocock's agent skills, all 38 of them (engineering 18, productivity 7, in-progress 9, misc 4). Vendored copy of mattpocock/skills, MIT (c) 2026 Matt Pocock.
Claude Code - compro0stars
Layer 3 Company Profile Multi-Agent Plugin with Canva Editorial Slide Deck Generator & Vercel Deployment
Claude Code - engram1stars
Cognitive memory system — episodic storage, semantic extraction, knowledge graph, and dream-state consolidation for Claude Code
Claude Code1 MCP server - dkj-subagents-shopify1stars
Add-on team for Shopify store repos (e.g. smartwatchbanden): the subagents Liam, Sandra, and Steven, the domain skill start-task (the branch that opens a task), push-preview (the branch's own unpublished preview theme, created on the first push rather than at branch creation), and an operational floor -- a PreToolUse guard that refuses a theme publish, a theme delete, or an unauthorised push to the live theme, whatever shell wraps the command, plus adopt-shopify-floor to place that floor in a repo (the guard's seams, a starter theme-check config, and the CI workflow over it) and sync-main, the pre-task sync that mirrors the live theme into the trunk without letting live overwrite what the trunk has done since. Deliberately NOT repo-neutral — only a Shopify repo enables this team. Teams stack: this one adds to 'dkj-subagents-alpha', and a commercial Shopify repo typically enables 'dkj-subagents-ecomm' alongside it. The governance layer (CLAUDE.md + the repo lenses) stays separate.
Claude Code1 skill - dkj-subagents-lifehub1stars
Add-on team for life-hub-style repos (a personal information hub / brain-based knowledge repo): the subagents Astrid, Fiona, Hugo, Ian, and Onyx. Deliberately NOT repo-neutral — only a life-hub-style repo enables this team. Teams stack: this one adds to 'dkj-subagents-alpha' rather than replacing it. The governance layer (CLAUDE.md + the repo lenses) stays separate.
Claude Code - dkj-subagents-ecomm1stars
Add-on team for commercial webshop repos (any platform): the SEO specialist Sergio, the CRO specialist Craig, and the performance/SEA specialist Sean, with room to grow into the wider e-commerce craft (lifecycle/email, analytics). Deliberately NOT repo-neutral — only a commercial webshop repo enables this team. Teams stack: platform-agnostic, so it sits alongside a platform team like 'dkj-subagents-shopify' rather than replacing it. The governance layer (CLAUDE.md + the repo lenses) and the core team ('dkj-subagents-alpha') stay separate.
Claude Code - dkj-subagents-alpha1stars
Portable executable core of the Claude Specialists system: the shared subagents plus the portable persona templates (personas/) of the main-loop specialists and the generic bootstrap skill specialists-init. This is the team every consuming repo enables; the other teams add to it, they do not replace it. Deliberately contains NO safety/guardrail hooks and NO repo-SPECIFIC skills — the governance layer (CLAUDE.md + repo lenses), the safety hooks, and the repo's own specialists stay at repo level. Two deliberate, repo-neutral exceptions: the skill specialists-init (the adoption path itself) and the informational SessionStart hook roster-sessioncheck (hooks/, read-only, never blocks).
Claude Code4 skills - dkj-policy-bwj1stars
The shared law layer for BWJ's two Shopify stores (smartwatchbanden and xoxowildhearts), packaged so both repos run one identical procedure and cannot drift on it. Four chapters. TICKET HANDLING: a discovered issue is created on GitHub first (GitHub is the source of truth), then mirrored to Asana as a colleague-friendly variant, cross-linked both ways; closing the GitHub issue makes a CI workflow post an update saying the work is ready to test and move the card, never ticking the task off - only the colleague who asked for it can do that. THE SYNC LOG: a 'sync/' branch mirrors what a third party changed on the live theme, is exempt from the changelog by design, and therefore owes a sync-log entry instead - dkj-policy-bwj/SYNC-LOG.md, newest at the top, written and committed by dkj-subagents-shopify's sync-main.ps1 in the same breath as the branch, never folded and never released. THE PREVIEW HANDOVER: where a preview is owed, it is a PAIR per market - the preview beside the live control, whose URL pins the live theme id, because preview_theme_id sets a cookie and the bare URL then keeps serving the preview while the reviewer believes they are looking at live. That pair travels as ONE link to a published page carrying a QR code per market, plus what the gates already proved and the one question being asked - never as a table of URLs, which a terminal wraps past selecting and a phone cannot scan. THE THEME LIFECYCLE: the estate owes two things at the two moments it grows - after the live push, the spent preview themes THIS repo created are swept away; at the release cut that closes that push, a fresh backup of live is taken, polled until it is provably complete (the CLI returns long before the copy is), and only then is the previous backup rotated out, so exactly one is retained and there is never a window with none. Every delete is bounded by a reserved name prefix this repo wrote, never by a theme's role: a store carries far more themes belonging to agencies, apps and colleagues than to the repo, and a role-keyed sweep would destroy them all while looking correct. It relaxes nothing about which changes owe a preview, or about no PR opening before one is approved - it tightens both, by adding exactly TWO steps to that workflow's cycle and changing nothing before the step list. The last step under '### CREATE' is always 'Is the change visible in the frontend / storefront?': dropped with its reason where nothing renders, and otherwise a preview theme, a comment on the GitHub issue carrying the steps and every URL, and ticked only once a PERSON confirms they looked - being last is what makes dkj-policy's own step-list gate hold the PR until then, since an agent can prove a theme exists and never that somebody looked at it. And just before the issue is closed, the paste-ready block gains its go-live half: the next release day (BWJ cuts on Mondays), the version that release is on course for, and the live storefront URL per market - written by build-golive-block.ps1, worded as a plan rather than a promise, and leaving out any fact it cannot derive instead of guessing one. It is an ADDITIVE add-on to 'dkj-policy', not a replacement: it extends four seams of that workflow - the ticket-work layer before a branch, what a sync branch owes, how a preview is handed over, and what the theme estate owes at a push and a cut - and says nothing about branch naming, what an ordinary change owes before a PR, or what a release is. Every chapter is policy rather than mechanism: the Asana CI lives in each repo's .github/, and the sync and preview machinery in 'dkj-subagents-shopify'. Requires the core team 'dkj-subagents-alpha' and 'dkj-policy'; the sync and preview chapters also expect 'dkj-subagents-shopify'. Enable this in BWJ's two Shopify store repos for all four chapters, or in this plugin's own source repo for ticket handling alone (admitted September 14, 2026, commit b9b2a65a) -- the other three chapters are Shopify-store policy and stay at the two stores.
Claude Code5 skills - dkj-policy1stars
One particular way of working, packaged so a repo can choose it: the branch-and-entry model, the tier ladder, the fold, the release cut and the gates that hold them, as skills plus their shared scripts. A workflow is not a team — it changes how the specialists work, not who they are — and the name says whose it is on purpose: DKJ's policy, the top rung of the order, not a baseline every consumer inherits. Deliberately NOT part of the core: a consuming repo is unique and has its own way of working, and the specialists adapt to it. Enable this only if you want THIS way of working. Requires the core team 'dkj-subagents-alpha', and a repo-side scripts/repo-config.ps1 + scripts/lib/branch-info.ps1.
Claude Code19 skills - goalspec2stars
Make any agent write its own grounded goal-spec, verify it with an independent adversary, and close through a fail-open completion gate. Hardens the built-in /goal.
Claude Code1 skill - video-intel6stars
YouTube video intelligence via Gemini: channel scanning, rich multimodal transcripts with on-screen content, concept taxonomy, hybrid search, and Bosnian/Croatian/Serbian subtitle translation. Three skills: video-intel (ingest/curate, runs in the plugin repo), video-intel-search (read-only, globally installable), translate-bcs (subtitle translation).
Claude Code1 MCP server - math-physics0stars
严谨的数学/物理推导与 Overleaf 交付。含 skill `math-physics-complete` 与子代理 `deriver`。
Claude Code1 skill - universal-plugin0stars
Research and design toolkit for building universal AI coding agent plugins that work across Claude Code, Cursor, Codex, and GitHub Copilot CLI.
CodexClaude CodeAgent Plugins8 skills - research-workbench0stars
Durable research workflows and community post authoring for AI coding agents.
Agent Plugins - stripe0stars
Stripe payment integration development with best practices, API upgrade guidance, and test environment access.
Agent Plugins - figma0stars
Translates Figma designs into code by creating and maintaining Code Connect template files that map design components to code snippets.
Agent Plugins - datadog0stars
Preconfigured access to Datadog monitoring, logging, and observability tools. Query logs, metrics, traces, and dashboards through natural conversation.
Agent Plugins - terraform0stars
Infrastructure-as-code specialist agent. Generates compliant Terraform using the latest provider and module versions via registry integration and workspace management.
Agent Plugins - python-pypi0stars
End-to-end skill for building, testing, linting, versioning, and publishing production-grade Python libraries to PyPI. Covers all four build backends and semantic versioning.
Agent Plugins - dependabot0stars
Manages dependency updates across Docker images, Python packages, and other ecosystems. Automates versioning, security scanning, and CI/CD compatibility checks.
Agent Plugins - security-review0stars
Scans authorized code to detect and confirm plausible vulnerability findings. Supports codebase investigation, change set analysis, and minimal fix preparation.
Agent Plugins - google-drive0stars
Read and work across Google Drive, Docs, Sheets, and Slides. Query documents with natural language and perform read/write operations.
Agent Plugins - gmail0stars
Read and manage Gmail. Filter threads, summarize unread messages, compose drafts, and manage labels through natural language.
Agent Plugins - github0stars
Official GitHub MCP server for repository management. Create issues, manage pull requests, review code, search repositories, and interact with GitHub's full API.
Agent Plugins - feature-dev0stars
Comprehensive feature development workflow with specialized agents for codebase exploration, architecture design, and quality review.
Agent Plugins - code-review0stars
Automated code review for pull requests using multiple specialized agents with confidence-based scoring.
Agent Plugins - singularity-agent2stars
Universal blockchain client for EVM, Solana, Bitcoin and Cosmos. Read-only: resolves addresses and names, reads balances and transactions, decodes calldata, and builds unsigned transactions for you to sign in your own wallet. Holds no keys.
Claude Code1 MCP server - vibesurfer15stars
Agent-native headless browser. Includes 2 skills and 1 MCP server.
CodexClaude Code2 skills1 MCP server - winnow8stars
A calibrated context sieve for Claude Code. Every large tool result is judged by a System One model before it enters context; irrelevant blocks are replaced with a summary and a recall key.
Claude Code1 skill1 MCP server - specpowers0stars
六阶段 spec 驱动开发流水线编排插件：constitution→brainstorm→specify→plan→build→archive，叠加结构一致性门禁与实时卡转人工。新增无人值守 auto 模式（/specpowers-auto 以设计文档为权威输入全流程直通，自动裁决确认/门禁节点）。桥接 OpenSpec CLI + superpowers 插件 + codex-review 技能，需预装。
CodexClaude Code1 skill - mailbox-radar0stars
偵測 Google Drive 交換區的未讀訊息，並在 session 開場、工具呼叫後、狀態列提示；有對話開著時用 watcher 喚醒對話，全機沒對話時用桌鈴提醒人（目前僅 macOS 橫幅，Windows 暫無）；同一封訊息多對話搶答以單機 claim 鎖避免重複。team-mailbox skill 隨本 plugin 一起安裝，使用者設定、已讀帳與通訊錄（成員名單，可從 Drive 分享名單同步）存在 ~/.claude/team-mailbox/。一臺機器可同時掛多個交換區（例如團隊一個、自己兩臺電腦互傳一個），各區代稱、成員、已讀帳獨立。支援 macOS 與原生 Windows。
Claude Code2 skills - franklin-method0stars
Turns any subject into a self-teaching campaign using Benjamin Franklin's two modes: acquisition, then reconstruct-and-diff refinement. Every check is external - Claude never grades the work.
Claude Code4 skills - mycareer2stars
Turn a career into a verified record, then render resumes, LinkedIn profiles, and executive letters from it. A recognition harness reconstructs what you actually did at autobiography depth; a validator refuses to let any document claim more than the evidence supports.
Claude Code2 skills - entwurf-receive28stars
Entwurf async receive mode for Claude Code — a portable, drop-in plugin that arms a PER-SESSION mailbox watch at SessionStart (watchPaths = $CC_MAILBOX_ROOT/<session_id>/inbox.signal) so an IDLE subscription session can be woken by an addressed external file-write (agy send-message parity; pair with cc-enqueue-addressed.sh). Only the targeted session wakes — siblings stay idle. This is the plugin form that a bare .claude/skills/ skill CANNOT achieve: plugin hooks register at startup, so the SessionStart hook actually fires and registers watchPaths. A bare skill's hooks register only on invocation (mid-session), after SessionStart has already passed.
Claude Code - entwurf-skill-plugin-example28stars
Minimum plugin shape that entwurf's `skillPlugins` validator accepts. Copy this directory anywhere on disk, point `skillPlugins` at the absolute path, and add your own `skills/<name>/SKILL.md` siblings to `hello/`. The entwurf bridge does not require any other key in this manifest — additional fields are forwarded to the Claude Agent SDK as-is.
Claude Code1 skill - entwurf-meta-receive28stars
Garden-native meta-bridge receive for Claude Code (entwurf 1.0.0). On every native session open the SessionStart hook idempotently upserts a meta-record at <pi-agent-dir>/meta-sessions/<garden-id>.meta.json (making the native session a garden citizen addressable by garden id) and arms a per-garden idle-wake watch at <pi-agent-dir>/meta-mailbox/<garden-id>/inbox.signal. CwdChanged re-arms; UserPromptSubmit backfills the record only (cannot re-arm). FileChanged is the addressed doorbell that wakes an idle session when a sender pokes its signal. Installed via the meta-bridge local marketplace with --scope user; do NOT hand-edit — use ./run.sh install-meta-bridge / doctor-meta-bridge.
Claude Code - entwurf-meta-receive-copilot28stars
Garden-native meta-bridge BIRTH for GitHub Copilot CLI (entwurf, #82). Copilot's hook schema has no `args` array and its exec form is a single `exec` string, so this unit is a sibling of entwurf-meta-receive rather than a variant of it. On the first prompt of a Copilot session (measured: sessionStart is deferred to the first prompt, userPromptSubmitted fires just before it) the hook idempotently upserts a meta-record at <pi-agent-dir>/meta-sessions/<garden-id>.meta.json, making the Copilot session a garden citizen addressable by garden id. It also writes a SENDER marker keyed to the Copilot host pid, so the session's MCP children can say which citizen is calling. This birth unit does NOT arm a receiver, and that stays true now that one exists: Copilot's doorbell is a forked first-party EXTENSION (installed separately with 'run.sh install-copilot-receive'), and only the process holding that watch may claim it. Birth gives identity and who-sent; the extension gives replyability. Installed via the meta-bridge local marketplace into Copilot; do NOT hand-edit.
Claude Code - raycast-extensions0stars
Stage-keyed skills for the full Raycast-extension lifecycle: api-changelog (derive @raycast/api release notes Raycast doesn't publish), scaffold (new), develop (change code, modernize, house-style audit fix), ship (dep hygiene, house-style audit, Store compliance, PR, mirror, cleanup), greptile-loop (drive an open Store PR's automated review to 5/5), review-pr (inbound review of a third-party PR).
Claude Code6 skills - humanizer0stars
Rewrite AI-sounding text so it reads naturally without changing what it says.
Claude Code1 skill - humanizer0stars
Rewrite AI-sounding text so it reads naturally without changing what it says.
Claude Code1 skill - anban0stars
Anban Creator plugin for OpenAI Codex — WeChat, Seednote, Montage, ecommerce, and agent-assisted publishing workflows
CodexClaude Code26 skills1 MCP server - tadw1stars
Personal agentic development workbench with role-named agents (software-engineer, code-reviewer, product-manager, project-manager, research-librarian, ux-product-designer) backed by action-named skills for code, planning, research, product strategy, design, and ASO work
Claude Code38 skills - aitmpl31.9kstars
/aitmpl opens a component browser above the prompt: the aitmpl.com catalog (agents, commands, MCPs, settings, hooks, skills, loops, mods) read live from the site's public JSON through $.http.fetch. Pick a type or a trending component, filter, page, open a component and install it from there ($.process.run with the fixed CLI argv and a validated path, no shell), drop its install command into the prompt, or open its page in your browser. Hooks session.start, command.run and ui.render on AbovePrompt; browsing costs no tokens.
Claude Code - karta0stars
Stack-agnostic, ad-hoc orchestration framework. karta synthesizes a binder of work items (karta-plan) and delivers it in parallel waves onto a per-binder integration branch (karta-deliver), building each item in an isolated git worktree (karta-build) and verifying it against its own acceptance check — behavioral (karta-verify) or visual (karta-validate). Works across stacks (frontend, backend, CLI, data, IaC, …) while keeping deep frontend support: component/icon/token mapping, DTCG conformance, and a design-validation loop.
Codex10 skills