Agent Plugins Marketplace

Recently added

Plugins grouped by the day they were added to the directory, newest first.

Sep 22, 2026

  1. cueloop2stars

    Review plans, changes, replies, and prototypes in cueloop Threads

    CodexClaude CodeAgent Plugins7 skills1 MCP server
  2. Помощник маршрута Jadlis: снимает состояние машины, ведёт от обязательной настройки к необязательным инструментам, объясняет простым языком.

    Claude Code1 skill
  3. Hands the small judgments in a session to TypeSafe's Jev, a System One model that returns typed answers instead of generating text: what kind of request each prompt is, which model tier a subagent needs, whether an edit breaks a rule already written in your CLAUDE.md or AGENTS.md, and which transcript blocks survive compaction. Kept blocks carry over verbatim — no LLM writes a summary.

    Claude Code2 skills
  4. Claude Code integration for portable research-council: bounded investigations, competing explanations, evidence-linked reports, and structured handoffs for people and agents. Model-neutral workflow with return-only council roles and a script-gated skill library.

    Claude Code3 skills
  5. Review local Git changes in diffle and return structured feedback to Codex.

    Codex1 skill1 MCP server
  6. Review local Git changes in diffle and return structured feedback to Claude Code.

    Claude Code1 skill1 MCP server
  7. baton0stars

    Context-aware, beads-backed git-worktree workflow orchestrator for solo developers. Pass the baton: dispatch a task to a fresh worker session, then bring it home.

    Claude Code16 skills
  8. biik0stars

    수리소 수학학원 표준 양식 스킬 — A4 인쇄물

    Claude Code
  9. Шестнадцать книжных советников по продуктовому строительству отвечают параллельно, скептики проверяют утверждения, валидатор пишет единый вердикт с SWOT, картой консенсуса и действиями.

    Claude Code
  10. Пятнадцать советников по маркетингу и влиянию (14 книг плюс тактический синтез 2026) разбирают запрос через призму своей книги, валидатор сводит вердикт с SWOT и картой консенсуса.

    Claude Code
  11. Четыре режима: разбор транскрипта звонка (--call), диагноз застрявшей сделки (--deal), оценка sales-артефакта (--verdict) и конвейер написания (--write). Шестнадцать советников-слоёв сделки по 21 книге, подмножество по типу задачи.

    Claude Code
  12. Два режима: VERDICT — совет оценивает готовый текст (SWOT, карта консенсуса, переписанные фрагменты); WRITE — конвейер бриф → три драфта → критика по типу текста → синтез финала.

    Claude Code
  13. Анализ ситуаций и решений через 140+ принципов из The 48 Laws of Power, The Art of Seduction, The 33 Strategies of War, Mastery и The Laws of Human Nature; каждый совет привязан к книге и закону тегом.

    Claude Code
  14. nupp0stars

    Советник по проектам на мета-системе NUPP (Nearly Universal Principles of Projects): шесть универсальных принципов под PRINCE2, PMBOK, P3.express, DSDM, Scrum и XP — диагноз ситуации и что менять.

    Claude Code1 skill
  15. b0stars

    Personal Claude Code skills: PR automation, X content, sludge video rendering, and design review.

    Claude Code6 skills
  16. toy-check1stars

    A toy plugin used as a teaching fixture for plugin review.

    Claude Code1 skill
  17. probe1stars

    Finds broken relative links in a docs tree. Read-only; writes nothing.

    Claude Code1 skill
  18. tally1stars

    Totals a ledger directory by record kind into analysis/tally/report.json. Writes only that directory.

    Claude Code1 skill
  19. ledgerly1stars

    Reconciles a ledger directory against its index.

    Claude Code1 skill
  20. fence1stars

    Audits code fences in Markdown for a missing language tag and fixes them automatically.

    Claude Code1 skill
  21. sentry1stars

    Freezes generated files while a review is open via a PreToolUse hook. Inert until .sentry/config.json exists; writes nothing itself.

    Claude Code
  22. gatekeep1stars

    Freezes generated files while a review is open: a PreToolUse hook denies the write. Inert until .gatekeep/ exists; writes nothing itself.

    Claude Code
  23. quill1stars

    Checks prose for passive voice and nominalisations and reports them by sentence. Read-only; writes nothing.

    Claude Code1 skill
  24. wick1stars

    Reports formatting violations in Markdown against a style file. Read-only; writes nothing.

    Claude Code1 skill
  25. drift1stars

    Renders contract drift as one static page. Writes only the page it is asked to write.

    Claude Code
  26. burn1stars

    Renders a burndown of open findings as one static page. Writes only the page it is asked to write.

    Claude Code
  27. weigh1stars

    Checks bundle weights against a budget and fails the build when one is over. Writes only analysis/weigh/.

    Claude Code1 skill
  28. spool1stars

    Queues long-running checks and reports their results as they finish. Read-only except analysis/spool/.

    Claude Code1 skill
  29. lockstep1stars

    Denies a write that runs ahead of a declared beat, at the tool-call layer. Inert until .lockstep/beats.json exists; writes nothing itself.

    Claude Code
  30. tether1stars

    Ties each test to the requirement it covers and writes analysis/tether/coverage.json. Writes only that directory.

    Claude Code1 skill
  31. gauge1stars

    Renders doctrine violations as one static page. Writes only the page it is asked to write.

    Claude Code
  32. demo1stars

    Denies writes to generated files while a review is open. Inert until .demo/ exists.

    Claude Code1 skill
  33. demo1stars

    Denies writes to generated files while a review is open. Inert until .demo/ exists.

    Claude Code1 skill
  34. demo1stars

    Denies writes to generated files while a review is open. Inert until .demo/ exists.

    Claude Code1 skill
  35. demo1stars

    Denies writes to generated files while a review is open. Inert until .demo/ exists.

    Claude Code1 skill
  36. zirkel1stars

    Staged reasoning-composition techniques from prompt-engineering theory for complex, ambiguous, or multi-faceted tasks: a Clarify -> Explore -> Decompose -> Execute -> Revise pipeline with code-enforced guarantees.

    Claude Code14 skills
  37. zeugnis1stars

    Audit AI-generated code quality across four independent domains — hallucinated dependencies, weak test assertions, contract drift, and agentic-loop reliability — with independent verification and a bounded self-optimization fix cycle.

    Claude Code8 skills
  38. takt1stars

    Enforces declared beat order at the tool-call layer: denies an edit or a dispatch that runs ahead of the step it depends on, so cross-plugin sequencing is a runtime gate rather than a sentence a model may skip. Inert until a repository declares its beats.

    Claude Code
  39. passung1stars

    Harmonize an already-modern, live codebase that grew inconsistent — divergent architecture, docs, code patterns, and style across modules — with a structured preflight / scan / map / canonize / brief / align / verify / status workflow. Derives the canonical form for undocumented, non-deprecated pattern variants (documented conventions and version-deprecated idioms are out of scope — see befund); an interactive consistency matrix; specialist agents; and optional dynamic-workflow orchestration with adversarial verification.

    Claude Code
  40. Reworks a Claude Code plugin to the official Anthropic standard the way a failed part is reworked to spec: a calibrated review of its skills, agents, commands, workflow prompts, hooks, scripts, report viewers, manifest, README and the developer docs that wire it in, then a tier-gated fix pass that applies only what a haiku or sonnet remediator can do without judgement, blind-verified, under a PreToolUse hook that denies every edit outside the fix lock. Nothing is graded until the finder passes a sealed hold-out; opus- and human-tier findings are surfaced, never auto-applied. Read-only except analysis/nacharbeit/ and, during a fix pass, the files its lock names.

    Claude Code6 skills
  41. matrize1stars

    Derives a named, platform-neutral design system from reference exemplars and emits it to any target. Measurement and interpretation stay separate artefacts rather than separate sections: decode records what each reference actually shows, graded for how far its values can be trusted and what rights its source carries, while name interprets — every recurring element gets a purpose, a rule and an anti-rule, and a rule with no stated failure case is not emitted at all. The source of truth is a DTCG token file, never CSS; CSS, a VitePress theme layer and the landscape sketchbook are formatter output run by committed scripts, so the same tokens regenerate byte-stable. References are read-only, enforced by a PreToolUse hook rather than by convention, so values and rules may be extracted while assets are never copied. Every fan-out agent is read-only; writes are confined to the configured design root's system/ and out/ directories.

    Claude Code12 skills
  42. lehre1stars

    Researches a code style, pattern and architecture doctrine, then enforces it at the tool-call layer: a blocking rule denies the write that would violate it, rather than describing the violation afterwards. Works from a blank page (greenfield project initialization, decomposed into units and seams whose build order is itself enforced) as well as over an existing codebase.

    Claude Code9 skills
  43. cupertino1stars

    A design and craft discipline rooted in Steve Jobs' documented decisions, applied through a fixed, sequenced lifecycle pipeline rather than a menu of individually-selectable design techniques.

    Claude Code15 skills
  44. Generate production-grade command-line applications in 12 languages, each following a unified five-pillar architecture doctrine and ecosystem-specific idioms, verified against that doctrine before presentation.

    Claude Code5 skills
  45. befund1stars

    Comprehensive self-assessment of live codebases: architecture, docs drift, CI/CD topology, conventions, code idioms, business rules, and UI/accessibility, synthesized into a prioritized, gated transformation plan.

    Claude Code16 skills
  46. Compiles a stated problem into an executable, budgeted agentic workflow, runs it as a redundant swarm — N candidates over the SAME scope in their own git worktrees, judged blind, exactly one landed and the rest deleted — and refuses to converge by retrying. Ordering is delegated to takt via a generated beat declaration rather than enforced twice. Writes to a scoped output directory (analysis/arbeitsplan/**), to candidate worktrees under .arbeitsplan/**, and — only behind an explicit approval gate — to .claude/takt.local.md; the single landing step is the one place it touches the working tree. A PreToolUse hook denies an identical re-dispatch, an out-of-scope write, a shared-tree write during a fan-out, and a dispatch past the declared budget. Ships a headless `claude -p` matrix runner for sweeps that need a fresh process and a real per-cell --model.

    Claude Code8 skills
  47. andon1stars

    Walks a repository's value stream stage-by-stage, proposing fixes for discovered gaps, proving each wire before advancing via seven evidence-grounded strategies, and halting rather than passing broken or unproven handoffs.

    Claude Code5 skills
  48. Self-optimizing hardening loop over a project's value stream (an ordered chain of stages connected by wires/data-contracts). Fixes one gap per stage and refuses to advance past a wire that isn't proven.

    Claude Code1 skill
  49. Turns Claude Code into a senior frontend architect for React/Next.js: verification gates on the build path, a generated inventory of your repo's own components and endpoints, house conventions enforced in CI, and independent audits — with an ablation benchmark that reports what it actually adds.

    Claude Code5 skills2 MCP servers
  50. gdocsmith0stars

    Google Docs surgical authoring and workflow engine

    Claude Code1 skill1 MCP server