Shape agent capabilities into your workflow.

Give pr-buddy a pull request number, branch, or URL and it will lead you through five Socratic stations covering the goal, architecture, decisions, risk, and ownership. You answer before the skill reveals its reading, so the review builds on a mental model you can explain. It then hands off to pr-review-toolkit and closes with a three-bullet recap; authenticated gh and the toolkit are required.

Install this collection to route Claude Code tasks through connected skills for brainstorming, planning, debugging, UI and UX, security, and review. It requires a plan before implementation, adds checks during execution, and blocks completion claims until tests or linters pass. The workflow ends with a handoff document so later sessions can continue with the important context instead of reconstructing it.

Convert manuals, datasheets, programming guides, and other structured PDFs with a digital-first extraction path and optional OCR for scanned pages. The skill reconstructs headings, recovers tables and preformatted listings, exports figures, and supports page ranges or batch folders. It writes Markdown beside the source by default, requires uv and Python, and can use pdftotext as an optional geometry source for better layout recovery.

Use this system to structure a learning topic into knowledge blocks with conditions, outcomes, examples, and fast decision patterns. It schedules practice repetitions, focus blocks, and daily checkpoints, then tracks scores, streaks, pass rates, and progress reports. Claude Code commands can create a seven-day plan or start interactive practice, while the local Python tools store the learning data without external packages.

Use this guide when Fn or Globe keys cannot be recorded, a saved shortcut does not fire, hold-to-record stops immediately, or normal typing causes false triggers. It routes symptoms to permissions, CGEventTap behavior, shared event pipelines, and ordered code entry points across Tauri, Electron, Swift, Objective-C, and Rust stacks, with manual verification procedures.

Claude Dependency Auditor runs native audit tools for Node.js, Python, Rust, Go, Ruby, JVM, and PHP projects, then separates actionable runtime issues from development-only, unreachable transitive, or disputed findings. It can also check license compatibility, generate CycloneDX or SPDX SBOMs, report stale dependencies, and propose remediations that are applied only after confirmation and verified with tests.

Boris reviews SaaS tools, APIs, automations, and AI agents as a fractional CTO, looking for duplicated capabilities, unnecessary complexity, failures, and runaway token spend. It rates proposed changes by feasibility, dependencies, security surface, and build time, while delegated agents scout platform changes, monitor scheduled jobs, and build or retire agents when a concrete operational gap is found.

Anthony reads CRM records, call transcripts, and campaign metrics to score conversations across preparation, value created, commitment gained, one-up positioning, and follow-through. It then uses pipeline math to choose the most important deal to advance and delegates prospecting, pipeline, cold-delivery, value-first, and inbound work to specialized agents. Notion, Discord, and Granola are required for the documented setup.

Tim screens holdings and watchlist candidates using return on invested capital, sustainable revenue growth, and free cash flow, with additional research around earnings dates. It can track selected 13F filings for conviction overlap and summarize portfolio quality and risk metrics. The advisor is explicitly read-only: it proposes changes but never edits the portfolio without the owner's approval.

Richard applies Rumelt's strategy kernel to separate a clear diagnosis, an excluding guiding policy, and actions that reinforce one another. It flags fluff, goals masquerading as strategy, scattered stakeholder compromises, and distant objectives, scores sprint tasks against the guiding policy, and compares actual metrics with targets. Daily resource-allocation checks cover time, capital, agents, knowledge, assets, relationships, and attention, while the weekly review realigns the next week's work.

Licensor scans twelve asset types—including methodologies, certifications, content libraries, assessments, training materials, and AI systems—across territory, format, certification, and technology licensing dimensions. It can research prospective licensees, draft SOW-style proposals with advances and royalties, and track opportunities in partner and decision databases. The documented workflow depends on web search, Google Drive, Notion, and Discord.

AI Video Tracker packages a model snapshot into a Claude Code and Claude skill for comparing leaderboard signals, public access, approximate pricing, and notable background details. Its editable JSON dataset is useful for quick recommendations and forks, but the bundled v1.0.0 snapshot was captured on April 9, 2026, so rankings, availability, and costs should be checked against live sources before making a current selection.

Describe the Notion workspace you need, such as a roadmap, team dashboard, content calendar, kanban board, or PARA system, and the skill guides Claude through a structured design. It combines visual conventions with practical database architecture, relations, rollups, formulas, and native automations, making it useful for people building a coherent workspace rather than a single isolated page.

The skill turns DOCX files into normalized Markdown by accepting inserted revisions, dropping deletions and comments, repairing heading levels, cleaning tables, replacing broken image references with readable placeholders, and collapsing excess whitespace. A small cross-platform Go binary coordinates Pandoc, so no Python, Node, or Ruby runtime is required. The converter and Pandoc are downloaded lazily on first use and then cached inside the plugin directory.

Ingest PDFs, documents, spreadsheets, and slides into an append-only vault, query the resulting knowledge with citations, and run autonomous curation waves that propose improvements to a Git-tracked Markdown wiki. Fresh-context review and mechanical citation checks reject edits that lose provenance, while a built-in graph viewer helps explore connections. The system suits sustained research over weeks or months; it is not aimed at instant retrieval from a huge corpus or native multi-user collaboration.

Using the fictional Ding Hai archetype as a memory aid, this skill organizes patterns such as self-righteous reframing, victim reversal, intimidation, and loyalty pressure into recognition cards and countermeasures. The default mode favors documentation, falsifiable questions, and escalation to a third party when appropriate. Writing, simulation, and sparring modes require explicit entry and preserve a counter-path, while the material is not a substitute for legal, mental-health, or real-world safety support.

Ask a compatible agent to turn Markdown into a formatted Google Doc, place it in a designated Drive folder, enable link-based viewing and commenting, and return the share URL. Previously published documents can be updated without maintaining a separate local state file. The workflow requires the Google Workspace CLI and Drive OAuth access, scans every create or update for common secret patterns, and demands explicit confirmation before overwriting a document outside its managed folder.

The pack separates risk in actions from risk in content. One skill governs shell commands, file changes, browser automation, uploads, personal data, and device-side effects; the other treats instructions found in webpages, PDFs, OCR, repositories, and logs as untrusted data. Both promote least privilege, read-only work first, explicit confirmation, and secret-exfiltration prevention, with shared review checklists, red-flag references, and adversarial test cases tailored to Minis on iOS.