skills

Performance Improvement Plan for Claude. Five triggers covering a rule's full lifecycle. CLAUDE PIP locks a behavioral rule into the project's CLAUDE-PIP.md. LOCAL PIP scopes a rule to the current working directory. OFF THE PIP removes a rule by id or keyword. PRUNE THE PIP audits the rules for staleness and reports the context tax (per-rule token cost, hits, caps); default verdict is keep, drops require concrete defects. PROMOTE THE PIP graduates stable rules to their canonical homes (project CLAUDE.md, DESIGN.md, user-global CLAUDE.md, or ~/.claude/rules/lessons.md) — the user calls out which to promote. PIP is the single lifecycle engine for both the project PIP file and the global lessons.md store. Every PIP-added rule is wrapped in HTML-comment markers so it can be cleanly removed later. Activates on any all-caps occurrence of `CLAUDE PIP`, `LOCAL PIP`, `OFF THE PIP`, `PRUNE THE PIP`, or `PROMOTE THE PIP` in a message. Also activates on phrases like "before compact", "audit PIP", "graduate PIP rule".

Use this skill when designing, reviewing, or refactoring data visualizations, dashboards, analytical graphics, reports, charts, maps, product metrics views, or quantitative UI surfaces. Applies Tufte-style graphical excellence: show the data, maximize meaningful comparison, reduce non-data ink, preserve statistical truth, increase useful density, and make complex ideas clear. Activate with 'tufte audit', 'redesign this chart', 'data viz review', or '/tufte-love'.

Governs how every skill in this marketplace is structured. Codifies the Anthropic Agent Skills directory convention (skills/<name>/SKILL.md at the repo root — no plugin wrapper), folder roles, naming rules, and migration triggers. Scaffolds new skills against the canonical layout, audits existing skills, and recommends when a folder should become its own sibling skill. Applies Unix philosophy: one skill = one purpose, compose via plain text, prefer small over large. Activates on "scaffold a new skill", "/new-skill", "audit this skill", "/semantic-audit", "should this be a folder or its own skill", or any time a skill structure is being designed or evaluated.

Audits web projects against atomic design discipline and brand-token coherence. Detects token violations, atomic hierarchy drift, naming by appearance, scattered components, brand drift, and scale violations. Three verdicts: system-healthy (ship), drifting (refactor against plan), broken (stop; build missing tokens and patterns first). Includes failover chain for brand detection: explicit tokens → parse image → scrape live URL. Activates on "atomic audit", "brand audit", "design system audit", "audit my components", "is this on-brand", "find duplicate components", "consolidate components", or "/atomic-brand". Auto-suggests when the user pastes a URL with a visible `tokens.css` or `tailwind.config` nearby.

Sizing rubric for effort AND complexity. Use whenever a proposal, plan, scoping table, or task list contains an "effort", "duration", or "complexity" column, or whenever the user or model is about to give a time-based estimate. Two axes: Size as T-shirt (XS / S / M / L / XL / XXL) for effort; Complexity as low / medium / high for cognitive load. Applies Karpathy's four code rules before sizing: state assumptions, ask if ambiguous, prefer simpler scope, require a success criterion. Activates on "estimate", "scope", "how long", "effort", "complexity", "size this", "/estimatrix", or before any estimation table is produced.

Agent orchestration substrate built on GitHub Issues + Projects V2. Treats each issue as a task with a claim lock (access exclusion + TTL), a dependency-chain DAG, and a comment-based event bus. Multiple agents can claim and work tasks in parallel without conflict. Includes audit, triage, generate-board, list-boards, and dispatch modes. Confirm-first by default; YOLO bypass available. Four archetype templates: personal-todo, writing-pipeline, oss-triage, sprint-planning. Protocol is agent-agnostic. Activates on "kanban", "board", "dispatch", "claim issue", "triage", "audit board", "list boards", "/github-issues-kanban", or when the user mentions GitHub Projects.

Validate customer need before any build. Use when about to design, code, ship, or scope a feature, product, page, or skill. Gathers evidence from interviews, tickets, sales calls, surveys, and artifacts; clusters jobs; scores confidence; emits a Job Article that frames the upcoming build as a Jobs-to-be-Done PRD. Activates on "JTBD", "jobs to be done", "validate this", "should I build this", "is there a real need", "what job does this serve", or "/jtbd-prd". Also auto-suggests when the user pastes a build proposal, feature brief, or hypothesis without supporting evidence.

Typography skill. Generates modular type scales from a base size and named ratio, emits BODY and HEADINGS token families separately, with line-height bands and letter-spacing defaults per band. Outputs CSS custom properties and Tailwind theme.fontSize. Color slots reference external `--color-*` custom properties via var() with sensible fallbacks. Runs conversational intake when inputs are missing. Activates on "type scale", "typography", "font sizes", "modular scale", "kerning", "line height", "typographic hierarchy", or "/learn2kern".