dev-core

Plugin dev-core from marketplace ai-setup in josippapez/ai-setup. Commands read from the repo's manifest — today.

Skills (11)

• accessibility

  Use whenever accessibility or a11y work comes up, and for any WCAG question — even when WCAG is never named. Two halves: scan a running page for real violations with the bundled axe-core harness driven through the chrome-devtools MCP, and look up authoritative WCAG 2.2 / 2.1 text with the bundled wc

• agent-browser

  Fetch facts from the live web and drive a real browser. Use for ANY question whose answer is not already in this repo or in a file you read: library APIs, current versions, changelogs, config keys, CLI flags, specs, error messages, pricing, docs of any kind. Also for interacting with websites: navig

• agent-orchestration

  Route tasks to the right subagent or Claude model tier based on task type, scope, and cost. The main agent stays orchestrator and prompt-loop owner throughout.

• code-structure

  Decide where operational logic belongs: actions own domain rules (the why and when), a shared service layer owns reusable mechanics (the how).

• computer-use

  Drive the real desktop with mouse, keyboard, and screenshots: native app UI, OS dialogs, browser chrome outside the page, and seeing what is actually on screen. Use whenever the thing you need has no API and no DOM behind it.

• exhaustive-code-review

  Review that finishes instead of stopping. Freeze the review surface in a ledger, fan out one subagent per review lane over the whole surface, and require a verdict for every unit so a later re-run cannot surface issues that were sitting there the whole time.

• find-improvements

  Research a project for what makes it slow, flaky, costly or painful to work with, then fix the problems one at a time, each with evidence: where the friction shows up, a measurement before and after, a test that fails on the old code, and a check where the code really runs. Ends with the fixes made

• opensrc

  Read a library actual installed source. Resolves a local cached path with `npx opensrc path <pkg>`, then grep or read it.

• pr-standard

  The bar a pull request has to clear to merge with no review comments, and the procedure for scoring a PR against it. Covers the PR-level properties (one reason to exist, size, tests that fail on base, description, dependencies, commits, CI) that the always-on rules do not; the hunk-level half is the

• self-improve

  Turn a correction or a failure from this session into a skill: update the skill that should have caught it, or write a new one. Covers finding the owning file, writing the description so the skill actually triggers, and mirroring the change to the other adapters.

• tanstack-docs-cli

  Find current TanStack documentation with the TanStack docs CLI.

Subagents (13)

• dockerfile-specialist

  Audits and hardens Dockerfiles and container builds — minimal/secure base images, digest pinning, layer hygiene, and reproducible, scannable builds. Use for Dockerfile audits, image hardening, and compose validation.

• docs-maintainer

  Maintains the repository's owning documentation and guidance (rules and skills) so it stays aligned with code and workflow changes. Use for docs-sync triggered by behavior, workflow, or config changes.

• figma-layout-token-analyst

  Analyzes Figma designs via the Figma MCP, reports layout structure and intent, and maps design values to existing design tokens — flagging token gaps. Read-only design analysis; use before building UI from a Figma reference.

• free-tier-explorer

  Low-cost agent for exploration, research, summarization, and low-priority background context-gathering.

• frontend-specialist

  Implements and refactors frontend UI with design-system reuse, accessibility, i18n, and testability as defaults. Use for component work, UI features, and frontend refactors.

• high-tier-coder

  Opus coding agent, used only when the approach is still undecided, the change is security- or safety-critical, mid-tier-coder already failed, or the user asks for it. Size or file count alone does not qualify; use mid-tier-coder for that.

• infrastructure-specialist

  Maintains build, CI, and deployment configuration with reliability- and security-focused defaults — deterministic, auditable, least-privilege. Use for CI pipelines, build config, and deploy settings (containers → dockerfile-specialist).

• low-tier-fast

  Fast, cheap agent for small explicit edits, renames, lint and type fixes, docs and config tweaks, test runs, and one-off lookups.

• mid-tier-coder

  The default coding agent. Use for implementation work of any size once the approach is clear, including multi-file features, refactors, plugins, and conversions.

• self-improve-specialist

  Implements requested changes to repeatable agent behavior by updating rules and skills in a consistent single-source-first flow. Use for non-trivial behavior, workflow, or guidance changes that must stay consistent across adapters.

• test-specialist

  Adds and stabilizes targeted tests with deterministic scope and clear failure diagnostics. Use for test creation, refactoring, and flaky-test stabilization.

• top-tier-reasoner

  Top-tier reasoning agent for architecture, complex design, deep debugging, ambiguity resolution, and safety- or security-critical decisions. Returns a concise recommendation with trade-offs and risks.

• wcag-a11y-aa-specialist

  Audits and remediates accessibility issues to WCAG 2.2 A/AA standards for web and mobile UI.

MCP servers (4)

• interactive

  runs: npx -y @rawwee/interactive-mcp -t 1200 --disable-tools message_complete_notification,intensive_chat

• context7

  runs: npx -y @upstash/context7-mcp

• chrome-devtools

  runs: npx -y chrome-devtools-mcp@latest --autoConnect

• computer-use

  runs: npx -y computer-use-mcp

Hooks (9)

• SessionStart

  diff -q "${CLAUDE_PLUGIN_ROOT}/package.json" "${CLAUDE_PLUGIN_DATA}/.installed-package.json" >/dev/null 2>&1 || (cd "${CLAUDE_PLUGIN_DATA}" && cp "${CLAUDE_PLUGIN_ROOT}/package.json" . && npm install

• SessionStart

  node "${CLAUDE_PLUGIN_ROOT}/hooks/link-rules.cjs" 0

• SessionStart

  node "${CLAUDE_PLUGIN_ROOT}/hooks/link-rules.cjs" 1

• SessionStart

  node "${CLAUDE_PLUGIN_ROOT}/hooks/link-rules.cjs" 2

• SubagentStart

  node "${CLAUDE_PLUGIN_ROOT}/hooks/link-rules.cjs" 0 SubagentStart

• SubagentStart

  node "${CLAUDE_PLUGIN_ROOT}/hooks/link-rules.cjs" 1 SubagentStart

• SubagentStart

  node "${CLAUDE_PLUGIN_ROOT}/hooks/link-rules.cjs" 2 SubagentStart

• UserPromptSubmit

  node "${CLAUDE_PLUGIN_ROOT}/hooks/inject-rules-digest.cjs"

• Stop

  node "${CLAUDE_PLUGIN_ROOT}/hooks/recheck-stop.cjs"

Similar plugins