Recently added
Plugins grouped by the day they were added to the directory, newest first.
Sep 5, 2026
- prothesis160stars
Multi-perspective framing — /frame (πρόθεσις: a placing before)
CodexClaude Code1 skill - proplasma171stars
Compare throwaway probes of candidate directions before committing to one — /preview (πρόπλασμα: a preliminary clay model)
CodexClaude Code1 skill - periagoge171stars
Align concrete cases first, then name what they share — /induce (περιαγωγή: turning-around)
CodexClaude Code1 skill - merismos171stars
Apportion an autonomous goal into coarse execution units and derive each unit's completion conditions — /apportion (μερισμός: an apportionment)
CodexClaude Code1 skill - katalepsis171stars
Check that something in play — code, a document, a result — is actually understood, step by step — /grasp (κατάληψις: a grasping firmly)
CodexClaude Code1 skill - hyphegesis171stars
Settle how work with several lines of thinking runs before starting it — /conduct (ὑφήγησις: a leading-the-way)
CodexClaude Code1 skill - horismos171stars
See what a task needs decided, then choose what to keep or entrust — /bound (ὁρισμός: a bounding)
CodexClaude Code1 skill - heuresis171stars
Widen an empty or prematurely narrowed field of ideas before choosing — /ideate (εὕρεσις: finding/invention)
CodexClaude Code1 skill - euporia171stars
Find the decisions a rough intent turns on, traced from your own material and the domain's usual decisions — /elicit (εὐπορία: way through)
CodexClaude Code1 skill - epistemic-cooperative171stars
Utility skills layered on top of the core protocols; each utility skill's own SKILL.md carries its contract.
CodexClaude Code9 skills - epharmoge171stars
Check that a result fits everywhere it lands — misfits and omissions on one sheet — /contextualize (ἐφαρμογή: an application)
CodexClaude Code1 skill - elenchus171stars
Challenge the context you are about to act on, before acting — /sublate (ἔλεγχος: cross-examination)
CodexClaude Code1 skill - anamnesis171stars
Resolve vague recall into recognized context — one session, or a line of work, topic, or settled concept across several — /recollect (ἀνάμνησις: a recollecting)
CodexClaude Code1 skill - analogia171stars
Check what an analogy, framework, or earlier design carried to a case actually supports — /ground (ἀναλογία: a proportion)
CodexClaude Code1 skill - aitesis171stars
Collect what the AI can reach on its own, hand back the rest as the user's unknown — /inquire (αἴτησις: a requesting)
CodexClaude Code1 skill - webgpt-orchestrator1stars
Orchestrate WebGPT, Codex, and Claude into an autonomous delivery loop.
Codex1 skill1 MCP server - boxlite-agent-tooling3stars
Shared BoxLite coding-agent skills, auditors, hooks, and repository tooling.
CodexClaude Code - sloth0stars
Headless board worker: pick a GitHub issue off a project board, implement it in an isolated worktree, review it, and hand back a PR — asking on the issue when blocked.
Claude Code2 skills - harness0stars
Shared Agent Skills, packaged for Claude Code and Codex from one canonical implementation.
CodexClaude Code11 skills - atlas0stars
Local engineering memory: keeps a git repository indexed, records why files changed, and answers repository questions without guessing.
Claude Code1 skill1 MCP server - agent-signage0stars
Road signs for coding agents: a PreToolUse hook that reports when the file you're about to touch is in a git checkout behind its upstream, and stays silent otherwise.
Claude Code - human-design-ai1stars
Connect ChatGPT and Codex to validated HumanDesign.ai charts and saved composite charts, deterministic bodygraph graphics, and authorized account workflows.
CodexClaude Code1 skill1 MCP server - web-dev1stars
Web development in one plugin: a web-developer worker (routing, REST/API integration, forms, state, SSR/CSR), an opus-floored frontend-reviewer for component and view logic, and three version-pinned stack skills — Next.js (App Router boundaries, opt-in caching, server actions, 14→16), React Native (lists, navigation, native-driver animation, Expo inversions), Vite (VITE_ env security, chunking, base, proxy, 5→8) — loaded by the code-review fan-in from the lockfile. Inertia and Laravel stay in their own plugins; plain JS/TS needs no skill.
Claude Code3 skills - vercel-skills-scout1stars
Scout skills.sh — Vercel's open agent-skills directory — for third-party skills matching the project's stack: scans manifests, queries the skills.sh search API, reports each hit with provenance (source repo, install count, skills.sh URL), previews the picked skills' SKILL.md before anything lands, then installs only explicit picks via npx skills add — project-level, never global, never auto-installed. Includes /vercel-skills-scout:suggest.
Claude Code1 skill - ultra-deep-research1stars
Deep-research harness that fans out parallel web searches, tiers sources by provenance, extracts date-stamped claims, and adversarially refutes them before synthesizing a cited report with a contradiction ledger and per-section confidence. Trigger 'ultra-deep-research' escalates to a Workflow-driven loop-until-dry fan-out with multi-vote refutation panels; standard depth runs a portable parallel-Agent fan-out. Ships /ultra-deep-research:research plus researcher and verifier subagents. Writes the report inline and to research/<topic>-<date>.md.
Claude Code1 skill - ui-ux1stars
UI/UX best practices with per-stack skills: shadcn/ui, ReUI, Aceternity UI, Astryx, Material UI, Tailwind, plus a library-agnostic component-libraries skill (headless and styled React libraries, owned-code vs dependency rules, a per-library map) — plus a design-tokens skill (spacing/type/radius/elevation/motion scales + semantic color tiering), a theming-system skill, theme creation with a live colour preview, a /ui-ux:build entry point, a WCAG 2.2 AA accessibility audit (/ui-ux:audit, the a11y-audit skill, and an a11y-engineer worker that applies the fix list with native semantics over ARIA patches), and ui-ux-reviewer + ui-ux-engineer agents.
Claude Code9 skills - testing1stars
Testing best practices: test pyramid and what to actually test, Pest/PHPUnit and Vitest/Jest idioms, Playwright/Dusk e2e discipline, factories and fixtures, mocking boundaries, flaky-test causes, coverage traps — plus the TDD workflow (red-green-refactor, red-green regression proof for bug fixes). Includes /testing:flake-hunt and a test-engineer worker agent; the test rubric reaches any diff through /code-review:review.
Claude Code2 skills - terse1stars
Chat-message brevity as a shape contract, not a word filter: prose-line budgets per turn kind, a fixed work-done skeleton (verdict → artifacts → max 5 findings → blocker → next), and a named cut list. Levels lite/full/ultra plus wenyan (classical-Chinese novelty) variants, hook-reinforced each turn. Out of scope: code, files written, subagent prompts, and how much work a turn does. Ships /terse:commit, /terse:compress, /terse:check, three compressed-return agents, and an opt-in MCP catalog shrinker.
Claude Code3 skills - taskmaster1stars
Idea-to-execution clarification: brainstorm fuzzy ideas into approved designs, grill requirements until zero ambiguity, deliberate the implementation approach with a blind four-persona round (via the approaches plugin when installed; inline 2-3 approaches otherwise), decide visual choices with mockups, validate whole experiences with an interactive clickable demo, then emit milestone-grouped single-prompt task cards. Outputs live under taskmaster-docs/ (no collision with a project's own docs/).
Claude Code4 skills - taskmaster-suite1stars
Meta-bundle (10 plugins): the clarification-to-execution pipeline and only what it dispatches into — taskmaster planning, task-runner execution, orchestration, code-architecture, approaches, stack-scan, skill-router, plus the ui-ux, testing and security lanes its cards route to. Trimmed from 32 on 2026-08-31 to fit the host's skill-listing budget: over that cap the host drops descriptions name-only and the surviving set varies per reload, so the extra members cost reachability for every member. Install the rest individually. Uninstall cleanly with /taskmaster-suite:uninstall. 200k-window installs: see the README's context-window requirement first.
Claude Code - task-runner1stars
Disciplined task execution: one task at a time, scope locked to the task, bounded verify-fix inner loop per task, no drift, no unbounded outer loop. Completion means every check passed with evidence — held at Stop by candor's gate (clause 4) when candor is installed; without it a run can end by narration. Ships the shared task-executor agent — the delegatable 'apply the fixes' sink every review routes to — a PostToolUse scope-lock tripwire, parallel-planning (/task-runner:plan), and the subagent discipline (the orchestration plugin was merged in on 2026-09-14): delegation contracts, verification panels, scripts/dispatch-lint.sh, and the ultra-assess boost hook.
Claude Code5 skills - system-design1stars
System-level design: service boundaries on data ownership, scaling paths, cache placement, sync vs async integration and its failure modes, single points of failure, plus domain modeling (bounded contexts, aggregates, ubiquitous language) and an event-driven skill (delivery semantics, outbox, sagas, DLQ) — firing system-design, domain-modeling, and event-driven skills, a /system-design:review command, and a system-architect worker + system-design-reviewer pair. Complements code-architecture (code-level structure) without overlapping it.
Claude Code3 skills - stack-scan1stars
Know what is installed, keep it healthy, find what to add. /stack-scan:report inventories manifests, lockfiles (composer, npm/yarn/pnpm/bun), runtimes and docker images into a required-vs-installed table with drift, missing locks and EOL majors flagged. /stack-scan:audit runs the composer/npm vulnerability, outdated and licence checks (package-hygiene) — report-only. /stack-scan:suggest reads the same manifests and suggests every marketplace plugin in three tiers with cited evidence (--yes auto-installs tiers 1-2, --full the whole stack-relevant set after one confirm, --persist/--global set scope); --skills searches skills.sh for third-party skills instead — provenance, preview, explicit picks only. Add-at-all is approaches' build-vs-buy.
Claude Code3 skills - skill-router1stars
File-aware skill auto-routing: a PostToolUse hook injects one additionalContext directive to load the relevant best-practice skill(s) when you edit a matching file (PHP/Laravel/Blade, React/Vue/Inertia/Livewire via .tsx/.jsx/.vue, plain .ts/.js/.py/.go/.rb/.rs, SQL and migrations (one engine-agnostic rubric, plus a MariaDB row where that engine is detected), components, tests, Dockerfiles, OpenAPI), a SessionStart hook primes a repo-tailored skill index, and low-confidence content signals flush as one digest on your next prompt (SessionEnd keeps the ledger). Every relevant skill co-fires per edit; fail-open, once per skill per session.
Claude Code - security1stars
Security for web apps: OWASP-aligned code review (injection, XSS, CSRF, authz vs authn, mass assignment, uploads, secrets, dependency audit) mapped to PHP/Laravel and JS/Vue, PLUS design-phase threat-modeling (STRIDE, trust boundaries), a data-privacy (GDPR/CCPA) regulatory lens, and an api-auth lens (token model, OAuth2+PKCE, scopes). Includes /security:review and a security-engineer worker agent.
Claude Code3 skills - secret-scanning1stars
Secret-leak prevention: a PreToolUse hook that blocks a Write/Edit introducing a high-confidence secret (cloud keys, private-key blocks, provider tokens, assigned secret literals) before it reaches disk, plus a /secret-scanning:scan command to sweep a path or diff on demand. Fail-open, high-confidence-only, with a fixture escape.
Claude Code1 skill - resilience1stars
Runtime quality in one plugin — does it stay up, can you see it, is it fast. Failure-mode design for integration points (timeouts, retries with backoff and idempotency, circuit breaking, graceful degradation, delivery semantics) with the merged error-handling and concurrency disciplines and the event-driven skill (brokers, outbox, sagas, DLQ, consumer idempotency); application observability (structured logs with correlation IDs, RED/USE metrics without cardinality bombs, trace propagation, honest health checks); and measure-first performance (N+1, payload, bundle, Core Web Vitals, cache correctness, percentile load tests). One audit, /resilience:review, over all six rubrics or one of them via --concern, and two workers, observability-engineer and performance-engineer, that apply only what a measurement or a finding proves.
Claude Code6 skills - quality-suite1stars
Meta-bundle: the code-quality plugins that carry an enforcement mechanism — review and code smells with the single-pass fan-in, architecture's evidence-at-claim Stop gate, candor's citation-and-reversal Stop gate, comment discipline (warn-on-write plus a narrow pre-write deny), destructive-command blocking, secret-leak prevention, a token-cost warn hook (lean), and file-aware skill auto-routing. The advisory nine live in quality-principles-suite; the membership rule is prose, not script-checked. Uninstall cleanly with /quality-suite:uninstall.
Claude Code - quality-principles-suite1stars
Meta-bundle: the code-quality disciplines that advise rather than enforce — approach deliberation and design patterns, OWASP security review and threat modeling, systematic debugging, resilience (failure modes, error handling, concurrency, observability, measure-first performance), dependency hygiene and stack inventory (stack-scan), and test design and coverage review. Accessibility auditing lives in ui-ux (frontend-suite, craft-suite). Split out of quality-suite so a project can take the enforcing half without the advisory half's always-on context; the membership rule is prose, not script-checked. Uninstall cleanly with /quality-principles-suite:uninstall. 200k-window installs: see the README's context-window requirement first.
Claude Code - process-suite1stars
Meta-bundle: the engineering-process category in one install — git workflow, approach deliberation (build-vs-buy, estimation, rollout, pattern selection folded in), hindsight mining, docs discipline (api-design), subagent orchestration, task execution, stack scanning, plugin scouting, a token-cost warn hook (lean), file-aware skill auto-routing (skill-router), a stronger-model second opinion when stuck (fresh-take), a refuting research harness (ultra-deep-research), and a committed codebase map (brain). Uninstall cleanly with /process-suite:uninstall — removes the bundle and prunes its auto-installed plugins. 200k-window installs: see the README's context-window requirement first.
Claude Code - plugin-scout1stars
Scan the current project's manifests and suggest every cc-plugins-marketplace plugin in three tiers — stack-matched, the curated any-project core, and the universal remainder — then install the picked ones after confirmation; --yes auto-installs the first two tiers, --full installs everything relevant to the detected stack after one plan confirm (--stack a,b,c fills in a stack the manifests do not show yet), --all offers every row explicitly, --persist/--global choose scope. Includes /plugin-scout:suggest.
Claude Code - php-suite1stars
Meta-bundle: the PHP category in one install — laravel (Laravel best practices and the Inertia.js skill behind one /laravel:review, plus the backend-engineer worker), web-dev (the generalist worker, the frontend-reviewer, and the Vite skill for Laravel's default asset bundler via /web-dev:review) and code-review (the review fan-in plus comment discipline: no-comment default, write-time denies for restatement, commented-out code, signature-repeating docblocks and files over the comment ceiling). Uninstall cleanly with /php-suite:uninstall — removes the bundle and prunes its auto-installed plugins.
Claude Code - payments1stars
Payments and billing discipline (Stripe/Paddle/Braintree): PCI scope minimization via tokenization, integer-minor-unit money, signature-verified idempotent webhooks, subscription state machines and activation races, dunning and proration, and an append-only ledger with reconciliation — a firing payments skill and a /payments:review command. The failure class is double-charges and revenue leaks.
Claude Code1 skill - orchestration1stars
Subagent orchestration discipline: delegation contracts (self-contained prompts, compressed evidence-backed returns, model/effort tiering, scout-then-fanout, writer isolation), verification panels (refuter voting, judge panels, loop-until-dry, completeness critic), plus the shared contract review commands route their fix lists through. Includes /orchestration:review.
Claude Code3 skills - llm-app1stars
LLM application engineering: eval harnesses and regression gates (evaluate, don't eyeball), RAG pipelines (chunking, embeddings, retrieval quality + re-ranking, grounding and citation), prompt versioning, prompt-injection defense, token-cost control, and context management — a firing llm-app skill and a /llm-app:review command. Defers provider API specifics to Claude Code's built-in claude-api skill (harness-provided, not part of this marketplace).
Claude Code1 skill - lean1stars
Prices every output as a debit and asks how much. Five surfaces, one bar each: code, tests, comments, files, and actions — a tool call, subagent spawn or verify pass whose result cannot change the next step is a paid no-op, and nothing else here prices that. Only four things buy more than the minimum: blast radius, an observed defect, a stated criterion, the user asking; the excess must name which, where it happens. Minimum means risk coverage, not count — never delete a test to hit a ratio.
Claude Code1 skill - laravel1stars
Laravel best practices: Eloquent N+1 prevention, form requests, service layer, queues, policies, a per-version leverage map (Laravel 11/12/13), mass-assignment security — and the Inertia.js skill (v1/v2/v3, Vue/React/Svelte adapters: partial reloads, deferred props, useForm, shared data, SSR) it pairs with, loaded by the code-review fan-in when the manifests show it — plus the backend-engineer worker (PHP/Laravel) that applies the review's fixes.
Claude Code2 skills - hindsight1stars
Cross-session self-improvement loop: mines session and subagent transcripts for recurring friction and proposes CLAUDE.md rules, skill/plugin ideas, and defects against the plugin artifact that was running — applied only on approval.
Claude Code1 skill - git-workflow1stars
Git workflow discipline: isolated worktrees for feature work, a structured finish protocol for development branches (verify, present options, merge/PR/park, clean up), and review-exchange rigor — what to include when requesting review, and technical verification before implementing received feedback. Ships a PreToolUse guard that refuses any commit, merge, tag or PR carrying a Co-Authored-By: Claude trailer or a Generated-with-Claude-Code line. Includes /git-workflow:finish.
Claude Code2 skills - frontend-suite1stars
Meta-bundle: the frontend stack category in one install — UI/UX stacks, web-dev (Next.js, React Native, Vite skills, the generalist worker and the opus-floored frontend-reviewer), the WCAG audit inside ui-ux, skill auto-routing (skill-router), and code-review (the review fan-in that loads those skills for whatever the diff touches, plus comment discipline: no-comment default, write-time denies for restatement, commented-out code, signature-repeating docblocks and files over the comment ceiling). Inertia lives in the laravel plugin (install by name). The creative-build studio — craft-layer, with its Three.js skill — moved to craft-suite, so ordinary frontend app work does not pay the studio's always-on context. Uninstall cleanly with /frontend-suite:uninstall — removes the bundle and prunes its auto-installed plugins.
Claude Code