My personal official collection of agent skills, react ones forked from vercel
Agent Claude Code
Read-only Claude Code subagents wired for this repository. Each keeps its behaviour in a harness-agnostic system prompt (action language, model aliases not raw slugs, minimal read-only tools) so the ADR-0004 export adapter can retarget it to Codex / Cursor / other harnesses.
Agent Claude Code
Use before promoting or releasing a skill/plugin, or on demand, to review the marketplace tier for cross-artifact drift — marketplace.json vs each plugin.json, metadata.json vs CHANGELOG vs skill-registry.json versions, and skills that silently duplicate an existing artifact. Complements the per-skill…
Agent Claude Code
Use before a commit or push, or on demand, to scan staged, unstaged, or untracked changes for private or employer work references, non-personal identities, or leaked GitHub / Slack / Linear / GCP identifiers before they enter this public repository. Complements the deterministic scope-guard hook by catching…
Agent Cursor
Checks skill folders against the repository's docs/architecture.md rules (e.g. required frontmatter, progressive disclosure, no bloated logic).
Agent Cursor
Takes clustering and architecture check outputs and scores skill overlaps, proposing explicit actions like adding anti-triggers, renaming, or archiving.
Agent Cursor
Clusters a given set of skill directories heuristically by domain, triggers, and artifact outputs. Use during skill overlap audits to reduce noise.
Agent Cursor
Use when investigating a skill system, workflow framework, subagent bundle, Claude-style plugin, or mixed agentic artifact and you need cheap inventory, classification, trigger extraction, and flow evidence before synthesis.
Agent Cursor
Use proactively when auditing a project's Cursor setup for missing or conflicting subagents, rules, hooks, or MCP configuration.
Agent Cursor
Use when reviewing Cursor hook policies, command blocking behavior, file-read protection, or subagent lifecycle guard-rails.
Agent Cursor
Use when auditing Cursor MCP configuration, server trust boundaries, env interpolation, or safe installation practices.
Agent Cursor
Use when benchmark or comparison results need a concise evidence-based analysis of what changed, what patterns matter, and which follow-up should happen next.
Agent Cursor
Use when adapting a source tree into a Cursor pack or bundled skill and you need a quick artifact matrix, source classification, or migration concerns before scaffolding.
Agent Cursor
Use when two candidate outputs should be compared blindly without knowing which skill or pack variant produced them.
Agent Cursor
Use when grading eval outputs for an authoring workflow against explicit expectations and collecting evidence into grading.json.
Agent Cursor
Use when comparing two skill source trees for package shape, trigger quality, hot-path size, supporting resources, safety, and repository fit before or alongside behavior evals.
Agent Cursor
Reads and analyzes Google Cloud application logs for Cloud Run services, Cloud Run-backed functions, Cloud Functions, and GKE or Kubernetes workloads. Use when investigating errors, latency, unexpected behavior, or trace-linked requests without flooding the parent context with raw log JSON. Applies discovery-first…
Agent Cursor
Adapts the reusable node-test-verifier pack to a specific repository by inspecting package.json, local agent guidance, and test scripts, then writing a canonical repo-local verifier contract plus any thin overlay references needed by the project.
Agent Cursor
Verifies Node and Jest changes with low-noise execution, tier-aware routing, explicit prerequisite checks, pass-fail-first summaries, and durable failure evidence.
Agent Codex
Use for production Bash scripts, CI shell automation, shell safety reviews, ShellCheck/shfmt cleanup, and Bats-backed script changes.
Agent Codex
Use for strict POSIX sh scripts, shell portability reviews, bashism removal, BusyBox/dash compatibility, and cross-platform shell automation.
At most 3 mods per repository are shown here, and a mod shipped inside a plugin is left to that plugin's page — the rest are on their repository pages: