Mp

Interviews the user about a plan, design, or requirements until the design is settled, then records the decisions in project docs. Use when asked to be grilled, or to pin down requirements or a design.

Generates self-contained HTML variant mockups from a design brief, one by default or N in parallel, and opens them for comparison. Use when asked to create a mockup, mock up a component, visualize a design, or produce design variants.

Reviews a branch, working diff, or PR across specialist axes (security, performance, tests, quality) and optionally applies the fixes. Use when asked to review code, changes, or a pull request.

Writes or updates HANDOFF.md with session progress and open threads. Use for a handoff or to save progress at the end of a session.

Creates and verifies Windows symlinks and directory junctions through PowerShell New-Item. Any symlink or junction work on Windows. Git Bash ln -s and cmd mklink silently create the wrong thing here.

Full programmatic API for Google NotebookLM: create notebooks, add sources, generate every artifact type, and download in multiple formats.

Recovers interrupted sub-agent and background work after a session-limit hit, crash, or manual interrupt, then resumes it.

Detects the project's check scripts, runs them, and fixes what fails. Use when asked to run checks, fix lint or type errors, or get the build green.

Splits a large source file or a folder of them into logical modules while preserving behavior.

Merges the target base branch into the current branch, resolving conflicts and pushing the result.

Investigates a bug's root cause, designs a TDD fix plan, and opens an issue/task labelled bug in the project's tracker.

Deduplicates code, removes repetition, and deletes dead code in a given scope.

Diagnoses and fixes fallow dead-code audit failures, suppressing or baselining findings when justified.

Creates or updates the domain language section of CONTEXT.md, confirming terms with the user first.

Bootstraps a project's visual identity — derives a palette, font pairing, density, and motion language from the project's domain, then writes designs/tokens.css and designs/DESIGN_SYSTEM.md. Use when asked to init or bootstrap a design system, set up design tokens, or establish a project's visual identity. Run once per project.

Voice-enabled variant of mp-grill: publishes each interview round as a JSON file for the companion mobile voice app, waits for the spoken answers, and continues until the design is settled and recorded in project docs.

Audits one skill or every active skill root for portable discovery, authoring conventions, hierarchy, workflow endpoints, tool integrity, and stale or ineffective instructions; safely fixes mechanical drift.

Creates or restructures a custom agent when a user needs a delegated role, distinct delegation branches, tools, workflow, or structured output.

Writes a standalone design brief for a UI component — surrounding context, exhaustive requirements and states, and a component reuse map — then gates dependent tracker issues/tasks with a Design needed label. Use for a design brief, design spec, component spec, or UI spec.

Creates or restructures a skill for Claude, Pi, and Codex when a user asks to add skill behavior, invocation branches, procedures, or reference material, then audits the result.

Applies refinement requirements to a chosen mockup variant, producing refined.html and SUMMARY.md, updating the brief, and unblocking the tracker issues/tasks that were gated on the design. Use when asked to refine a design, accept or select a variant, polish a mockup, or refine all pending designs.

Verifies UI changes with raw Playwright over a defined scope and reports a per-surface PASS/FAIL table with screenshots. Use when asked to visually verify UI changes or run a Playwright test.

Audits design-system component usage, flagging native elements, wrong variants, missed componentization opportunities, and hardcoded colors that bypass theme tokens.

Discovers runnable scripts across package.json files and identifies the frontend, backend, and database run commands.

Scans recent Claude Code sessions for design discussions, folds the decisions into CONTEXT.md and DECISIONS.md, and flags redundant memory files.

Repo-wide audit of code-quality suppressions (eslint-disable, ts-ignore, fallow-ignore) that fixes unjustified ones and opens a PR.

Reviews codebase architecture and logs a refactor issue/task in the project's tracker with the findings.

Consolidates CONTEXT.md by removing duplicates and outdated items and tightening the language.