Shape agent capabilities into your workflow.

Local Delegate decides when a large, repetitive coding task is worth sending to a local LM Studio model. It prepares a self-contained request, calls the local OpenAI-compatible endpoint, verifies the result, and falls back to direct writing when the server is unavailable. The skill targets outputs of roughly 80 lines or more, such as fixtures, enums, and test cases; it is not intended for refactoring, security-sensitive code, or small snippets. Its examples assume Windows PowerShell.

Claude Skill Mesh turns connected MCP services into a self-organizing set of readable agent specifications. It discovers available tools, observes recurring patterns and bottlenecks, spawns specialized agents for worthwhile automations, and periodically promotes, merges, or prunes them. A registry, short observation window, digest, and per-agent logs keep the state compact. It requires at least one MCP connection and is designed for Claude workflows rather than a standalone runtime.

CleanMyMac and CleanMyPC inspect actual disk usage before proposing targeted cleanup. They separate caches from user data, ask for confirmation before each deletion, use Trash-first handling for larger or riskier items, and compare free space before and after. The skills cover macOS and Windows, but explicitly protect passwords, mail, cloud-synced files, browser profiles, and other sensitive locations. They are for negotiated cleanup, not blind sweeping.

Content Repurposing Engine turns an article, essay, podcast transcript, or report into platform-specific deliverables. It extracts the central claim, hook, proof, and takeaway, flags thin or technical material, and can produce a LinkedIn post, X thread, newsletter blurb, or short-video hook with a format-specific visual idea. A selector lets you request only the outputs you need, while tone presets keep the voice consistent.

Snapshot Reader lets an agent work from two Markdown files: one file tree and one snapshot containing the source in code blocks. It extracts only the files needed for a request, tracks multiple snapshots across turns, and returns minimal BEFORE/AFTER fragments ready to apply in an editor. The workflow is useful when you do not want to grant direct repository access, but changes are applied by the user rather than written automatically.

Agentic Contract Skill renders a branded Agentic Development Services Agreement from provider, client, project, commercial, and governance facts. It supports base INI configuration plus client overrides and can output a contract PDF, filled Markdown source, and optional addressed envelope. The local generator uses Python packages including ReportLab, Jinja2, and PyYAML. Its demo template is not legal advice, so real engagements need counsel-reviewed terms.

Issue Creator turns a bug, feature request, or infrastructure need into a repository-aware GitHub issue. It inspects the code to cite verified file and line locations, searches for duplicates, follows repository templates and triage rules, and prepares acceptance criteria without prescribing the fix. A full preview appears before `gh issue create`, and publication waits for approval. It does not implement changes, open pull requests, or triage existing issues.

CoFoundr’s handoff and pickup skills keep a multi-session coding epic in one living Markdown file. Handoff appends completed features, decisions, files, debt, and session notes, while pickup briefs the next session on current state, remaining work, open questions, and priorities. The document is archived only when the epic ships, so context accumulates instead of scattering across dated files. These skills manage continuity, not implementation.

Stem Cheat Sheet helps students turn course materials and past exams into a compact, exam-targeted reference. It asks about weak topics and page constraints before laying out source-grounded formulas, recognition rules, procedures, traps, and small examples. The generated PDF must be checked against the allowed page count, paper size, sides, color, and example policy. Without shell access, it can still produce HTML but cannot verify the printed page count.

Design Log adds a gated Stop & Think protocol to Claude Code development. It asks clarifying questions, researches several sources through an MCP tool, explores the repository in plan mode, records the design, waits for one approval, and then hands unfinished checks to testing. Projects need a design-log index and research-tool configuration, and the workflow does not auto-merge to the main branch.

Claude UXR Skills is a personal toolkit for UXR and ReOps work at Miro. Its skills draft research plans and interview guides, audit prototype repositories without changing them, summarize weekly activity, and recap action items from Slack messages. Outputs are intentionally drafts for editing, and the end-of-day recap requires a connected Slack MCP. The repository complements rather than replaces the team’s canonical research skills.

Prompt Architect treats prompt writing as an engineering exercise. It diagnoses the task, asks for constraints, recommends suitable frameworks, drafts two or three alternatives, and scores them for clarity, specificity, context, completeness, and structure. The skill includes eight prompt frameworks, agent patterns, and anti-pattern examples, with model-specific guidance for Claude, GPT, Gemini, and local models. It returns a recommended version with tradeoffs and hallucination-risk notes.

Use this Claude Code skill to turn a loose coding request into a condition with a clear end state, visible proof, scope safeguards, and a turn limit. It includes six task patterns, worked examples, and eight rules for avoiding weak completion criteria. You still need to supply project-specific commands and boundaries; the template does not verify the work itself.

Recover mistyped English messages without retyping when the Israeli Hebrew keyboard layout was active. The skill detects likely gibberish, maps Hebrew characters back to their physical QWERTY keys, preserves real Hebrew and mixed text, and includes a deterministic Python decoder plus an optional prompt hook. It also has a reverse mode for English-to-Hebrew mistakes; other layouts are not included.

Use the two skills when an AI coding task needs code from elsewhere without pulling in an entire repository or runtime dependency. One extracts the smallest useful unit with a provenance comment; the other vendors only the slice you own, records its origin, trims unused parts, and supports selective resync. They are aimed at keeping borrowed code deliberate and maintainable.

Analyze a YouTube video by combining a downloaded 720p file, scene-change frames, a NotebookLM transcript and briefing, and Claude's multimodal reading. The skill writes a structured Markdown synthesis with a TL;DR, technical stack, step-by-step workflow, annotated frames, quotations, and actionable decisions. It requires yt-dlp, ffmpeg, the NotebookLM CLI, and an authenticated NotebookLM session.

Measure tempo from MP3, WAV, FLAC, M4A, OGG, and similar audio files with a two-stage analysis. A coarse pass checks the metric level, then a fine pass refines the BPM within a constrained range; the JSON output includes the primary BPM, metric-level BPM, result status, alternative half/double or 3:2 candidates, and the first 20 beat timestamps. It uses librosa and soundfile.

This experimental Claude Code skill helps solo builders surface the production concerns that an MVP can hide. Before coding, it asks four classification questions, assigns a risk tier, and can guide assumptions, security-tool checks, critical-path walkthroughs, handoff documents, and a cool-down before launch. It does not modify code, enforce deployment controls, or provide legal, compliance, or security certification; the skill content is currently German.