Shape agent capabilities into your workflow.

Frequent-Flyer Advisor asks three to six questions about a trip, ranks three airline-mile options, and produces a 1080×1440 shareable savings card plus a full HTML strategy report. Its bundled knowledge base covers 53 programs with dated sources, and savings are calculated against real cash fares while the cents-per-point value stays internal. The generator runs offline with no third-party dependencies, but program rules and prices should be checked against official pages before use.

git-finder searches for repositories that fit a stated project need, fetches the candidates, and checks their license, maintenance activity, and actual architecture before recommending three or four foundations. The report includes a primary choice, a fallback, and a quickstart showing installation and real API usage. It never invents names, URLs, statistics, or structure, treats permissive licensing as a hard gate, and does not write application code unless you ask after the validation.

skill-steward manages a Claude Code skill collection through three layers: a read-only loader that finds the right pack, a coordinator that audits and moves skills, and a synchronizer for cc-switch drift and repairs. Packs are discovered from the filesystem instead of hard-coded domains, and the system uses a small set of Markdown files without external services or databases. It suits installations where a large global skill directory is becoming difficult to navigate.

geo-markdown-mirror identifies important pages in a website or documentation site and creates predictable Markdown mirrors under an /llms/ folder. It flattens JavaScript widgets such as pricing sliders, tabs, and accordions into readable tables, generates an llms.txt index, and can add alternate links or ask-AI buttons. The skill supports Next.js, Docusaurus, Vite, and static HTML, with a regeneration script to keep mirrors aligned as pages change.

ui-ux-doctor scans React source and FastAPI code with only Python’s standard library, flagging dead controls, rendering problems, accessibility gaps, and integration mistakes such as unsafe CORS. A static mode reads source files, while a live mode uses an already installed browser to inspect the rendered DOM, contrast, labels, dialogs, console warnings, and failed requests. It needs no pip, npm, or network access, making it useful on locked-down machines.

kickoff interviews the user to turn a one-line idea into a PRD and a cluster-based plan with measurable success criteria. The user approves before coding, Codex signs off each cluster as it lands, and a final regression pass plus an over-engineering audit closes the work. An Obsidian folder preserves detailed state across sessions. The skill expects related planning, frontend, audit, and authenticated Codex tools to be installed.

sketch-inspector converts Sketch documents into cleaned JSON, deduplicated image assets, and a LayoutIR describing flex, grid, absolute, or leaf structure. For large designs it splits the result into section files with a manifest, allowing an agent to implement manageable pieces before replacing markers in the main layout. The toolkit is designed for Claude Code, Cursor, and other agents and pairs with the official Sketch MCP server for HTML, CSS, React, or Vue work.

introduce-ts inspects an existing JavaScript project to identify its bundler, framework, Babel use, and package manager, then creates a matching tsconfig and adjusts the build setup. It migrates one leaf utility first and runs type checking plus a build as a smoke test before explaining the next steps. The skill deliberately avoids converting every file at once, enabling strict mode immediately, or combining conflicting loaders. Projects already using TypeScript are skipped.

claude-leandata-skill gives Claude structured guidance for LeanData FlowBuilder routing, account matching, tagging, Audit Log diagnosis, and the differences between real-time and CCIO batch behavior. It can map fallback edges, explain administrator settings, and support certification preparation while grounding field names in a reference set built from official documentation. The skill targets Claude Cowork, and its recommendations still need checking against the organization’s actual configuration and platform version.

github-launch-agent divides a GitHub launch among parallel agents that inspect the project, research comparable repositories, score naming and search choices, generate a social preview, and prepare README and release files. It can also set up GitHub Pages and produce a seven-day distribution calendar with ready-to-use copy. The skill is designed for Claude Code, Cowork, and Claude.ai, and trades a long sequential checklist for coordinated agents and a shorter launch session.

kw-asset-skill combines a Claude Code or Antigravity skill with a manual playbook for creating interactive HTML5 learning assets for the Knowledgeworker LMS. It uses Vite and vanilla JavaScript, includes portable templates and reference material, and can scaffold an asset project from a setup command with a selected type. The result is intended to run inside Knowledgeworker, so authors need access to that LMS and its embedding interface rather than a generic web host.

3d-brain-animations turns a sentence such as “rotating rat hippocampus” into a PNG figure or MP4 animation. It selects a BrainGlobe atlas, finds the named regions, builds a translucent brain scene with glow and lighting, and renders it headlessly in Blender so the agent does not need to write 3D code. More than 200 atlases support multiple species, making it useful for papers, teaching, and talks. Blender and atlas data remain required runtime dependencies.

install-tags-10dobro creates a shared gtm-loader.js and inserts the Google Tag Manager script plus noscript fallback into every HTML page in a directory. Re-running it avoids duplicate tags, each changed page receives a backup, and dry-run plus loader-path options help control the edit. The tool installs the container; GA4 and Meta Pixel tags are configured in the GTM interface afterward. A folder of HTML files and a GTM ID are the minimum requirements.

small-cap-deepdive starts from an investment theme or ticker, enumerates the relevant SEC filing universe, removes candidates with mechanical red flags such as going-concern risk or death-spiral dilution, and then ranks survivors for falsifiable due diligence. Runs keep manifests and valuation settings so results can be compared across versions, and the process forces disconfirmation rather than a one-sided story. The output is a landmine screen, not a buy list; human financial judgment remains essential.

open-collider helps Claude move beyond obvious brainstorms by combining distant domains and reference objects into a curated, ranked idea pool. It runs as a single Markdown skill with no server or external state, and asks about token cost before an intentionally large ideation pass.

safe-push-skill helps developers catch secret leaks and repository-hygiene problems before git push. It checks staged files, sensitive content, README freshness, and environment-example parity, then returns pass, warning, or block and stops high-risk pushes.

contest-idea-forge helps contest participants turn one contest link into feasible, research-backed project ideas. It gathers and deduplicates papers, checks prior art, technical soundness, citation validity, and team or demo constraints, then has Claude generate, challenge, and rank candidate ideas while deterministic Python handles fetching, paper search, and rendering.

technical-intake helps a non-technical client clarify a vague software idea through an adaptive 25-question interview. It then produces a developer-ready specification, implementation prompt, working React or HTML mockup, tailored tool recommendations, and a project communication style suited to the domain and the client's technical level.