dundas

Argue against proposed actions to test judgment quality. Not "is the code correct" but "should we be doing this at all?" Aligned with robustness, not performance.

Orchestrate Claude Code agent teams for parallel multi-agent collaboration

Learn and integrate new APIs, creating permanent skills for external service access.

Create and update CHANGELOG.md with entries that include AI model/CLI attribution, PRD context, and task references.

Audit decisions for judgment quality, compliance bias, and manipulation vulnerability. Inspired by Anthropic's Project Vend Phase 2 finding that helpfulness training creates exploitable attack surface.

Extract visual style from reference UI screenshot and codify into reusable design system.

Build or update frontend components and pages that strictly adhere to the project design system defined in design/design-system.json (and design/design.json when present).

Orchestrate PRD creation, task generation, and gated task processing with user confirmations between phases.

Build or update frontend components and pages that strictly adhere to the project design system defined in design/design-system.json (and design/design.json when present).

Orchestrate PRD creation, task generation, and gated task processing with user confirmations between phases.

Create distinctive, production-grade frontend interfaces with high design quality. Use this skill when the user asks to build web components, pages, or applications. Generates creative, polished code that avoids generic AI aesthetics.

Create a clear, actionable PRD with clarifying questions and a junior-friendly structure.

Process a task list one sub-task at a time with pause/confirm gates, test/commit protocol, and file tracking.

Generate high-level tasks and gated sub-tasks from a PRD, with relevant files and testing guidance.

Control browsers via the ThinkBrowse CLI (the thinkbrowse / thinkrun command) — navigate pages, interact with elements, extract content, take screenshots. Use ONLY when the user explicitly names the thinkbrowse or thinkrun CLI, or asks to drive the browser from shell scripts / terminal commands. For general browse, scrape, or automation asks that don't name the CLI, prefer the web-browse skill. Do NOT use for simple URL fetching (use WebFetch tool instead).

Control browsers via ThinkBrowse MCP tools — navigate pages, interact with elements, extract content, take screenshots. Use ONLY when the user explicitly references the thinkbrowse/thinkrun MCP server or its MCP tools. For general browse, scrape, or automation asks that don't name MCP, prefer the web-browse skill. Do NOT use for simple URL fetching (use WebFetch tool instead).

Walk through a product UI as a real user — take screenshots, find broken flows, and produce a structured report with every fix listed. Use when: audit the UI or UX, do a UX review, QA a feature, check if something looks right, verify a user flow or onboarding, walk through a journey, 'is X broken?', 'check how X works', 'verify the feature we shipped'. Do NOT use for: fixing a specific known bug, reading a component's code, deploying, writing tests, or answering questions about code structure.

Browse the web programmatically with ThinkRun — drive a real or cloud browser to navigate, interact, extract, and screenshot, from the CLI or any MCP client. Use when: visit or open a URL, check a webpage, interact with a browser, verify something works live, take screenshots of a page, scrape or extract content, automate a browser task. Do NOT use for structured UX audits with scoring and fix reports — use ux-audit for that.

Orchestrate adversarial player-coach loop for high-quality code synthesis. Based on dialectical autocoding methodology.

Process tasks autonomously with automated PR reviews and gap analysis after each PR creation/commit.

Process tasks using async background subagents for massive parallelization and speedup.

Generate comprehensive project briefings from transcript history for brain onboarding