Shape agent capabilities into your workflow.

github-repo-launch-kit organizes an open-source launch into trust signals, repository differentiation, first-wave outreach, and ongoing maintenance. It inspects the repository and can safely apply selected setup changes through a dry-run-aware script, while marking human-owned work such as authentic posts and the launch narrative. It explicitly rejects star manipulation, automated account posting, and replacing the maintainer's voice.

Commitron analyzes changes and creates one atomic Conventional Commit per call, or splits a larger set into multiple commits in batch mode. It uses git commit --only to avoid broad shared-index staging, matches the language of existing history, and never uses git add . or adds co-author trailers. The tool is designed for Claude Code and Codex workflows where unrelated agents may touch different files.

jira-dev-workflow turns a Jira ticket key into a gated Claude Code development flow: read the issue, plan the work, create sub-tasks when appropriate, implement on branches, commit, merge, update Jira states, and post a completion summary. Every state-changing step asks for approval, and interrupted runs can be resumed. A Jira MCP server registered under the `atlassian` name is required; without it, the workflow stops before making changes.

html-to-svg converts an HTML preview into an SVG whose glyphs are embedded as paths, preserving browser line breaks and Variable Font weight and optical-size settings in environments that cannot load the original fonts. It targets static assets such as brand banners; the first run installs Node packages including fontkit, Playwright, and SVGO.

spectastic-debugger classifies defects in spec-driven projects as constitutional, specification, planning, cross-spec, or implementation gaps, then proposes a fix at the highest layer that actually needs to change. Its evaluation harness covers under- and over-escalation, scope traps, hotfix pressure, cross-spec drift, and platform bugs. The skill produces a structured triage report and is designed for disciplined diagnosis, not automatic code edits.

Use this skill when Claude Code needs reliable, Mac-specific guidance for SwiftUI and AppKit work instead of defaulting to iOS patterns or stale APIs. Its references cover windows, menus, sandboxing, signing, notarization, testing, and macOS 15 and 26 features, with triggers in English, Vietnamese, and Simplified Chinese; iOS and other Apple platforms are outside its scope.

Use this skill to test a crypto quant idea against current repository evidence before investing in optimization or promotion. It separates provider coverage from data trust, designs Stage 0 and falsification checks, and returns explicit decisions such as go, no-go, blocked, or research-only. It is a research discipline for Codex and Claude Code, not an alpha generator or trading bot.

MockHunter opens a live page through Claude Code and Playwright MCP, exercises its interface, and traces each visible value through network calls, APIs, and an optional database. The report classifies data as real, mock, LLM-generated, hardcoded, broken, or unknown, with severity and recommended actions. It avoids destructive controls and payment submission, but currently audits one page per run and may require manual pre-login for OAuth, magic links, or two-factor authentication.

Crawl a page locally and run a broad set of SEO, AEO, and AI-visibility checks, then score the result and rank the fixes that matter most. Apex works as a Claude skill or standalone CLI without an account or hosted backend. Live citation checks for services such as ChatGPT, Claude, and Perplexity require your own provider keys; without them, those checks are skipped and the score is capped.

econometrics-phd-level-skill organizes a Korean lecture series into topic-specific references and runnable Stata scripts covering OLS, inference, specification, IV and GMM, difference-in-differences, regression discontinuity, panel fixed effects, MLE, Logit, and Probit. A keyword router loads only the relevant references for each question, keeping the active context smaller. The material is derived from a third-party course rather than redistributing its original recordings.

ios-marketing-att-skill guides an iOS attribution audit from context and routing through diagnosis, action, and verification. It covers common systems such as AppsFlyer, Adjust, Branch, SKAN, ATT, Firebase, Amplitude, and Singular, helping teams inspect collection, transport, consent, and attribution results. It is a troubleshooting and review workflow, not an automatic guarantee of correct attribution.

mailoutofspam guides a ten-step deliverability repair: audit SPF, DKIM, DMARC, and MX, add suppression handling, implement RFC 8058 unsubscribe, parse bounces, throttle by provider, and verify raw headers through IMAP before resuming volume. It focuses on compliant recovery rather than spam evasion and includes standalone Python helpers. DNS, SMTP, and IMAP access are needed for a real diagnosis, and sending remains an operator-controlled action.

The Craaft skill helps Claude Code create and manage Kanban cards through Craaft's JSON API. It documents bearer authentication, error recovery, rate limits, board-level access, and the full v1 endpoint surface, while calling out traps such as float positions, column keys, PATCH-only metadata, and multipart attachments. It also includes curl examples and guidance for the official Python SDK; an API token is required.

agent-saga turns long-running work with AI agents into serialized documentary fiction grounded in Git history, logs, journals, transcripts, and other configured artifacts. It can create a 20–30 chapter arc, daily entries, a consistent narrator voice, and prose in a chosen language. The skill requires a configured source set and a continuously present agent, and refuses to invent events, agents, or public surfaces that are not supported.

Advertorial Agent takes a conversational brief through intake, copy, design, images, assembly, and a Framer CMS push, pausing for approval at four checkpoints. It retrieves niche-matched examples for headlines, voice, structure, calls to action, and image prompts, then produces a parent row plus ordered section rows. The project is a single-user Claude Code setup that requires configured language, image, vector, and Framer services.

Use this agent-agnostic skill for plans, specifications, code reviews, architecture explainers, reports, prototypes, slide decks, or small editing tools that benefit from layout and interaction. It selects among 20 patterns and produces one self-contained HTML file with inline CSS and JavaScript, responsive behavior, keyboard access, contrast checks, and export controls. Simple prose, tiny tables, and code-only answers are intentionally left in Markdown.

Paste Korean content to receive a ten-module GEO diagnosis covering evidence, quotations, citations, clarity, terminology, keyword density, and FAQ structure, followed by a domain-specific rewrite and comparison report. KoGEO instructs Claude not to invent statistics or quotes. Its method adapts research conducted mainly in English, so Korean results and changing generative-search rankings still need ongoing validation.

Metric Brand Auditor evaluates a brand through seven research dimensions and five scoring lenses. Parallel researchers gather public material, independent judge panels score the same synthesis, and a lead merges consensus, disagreement, leverage, and a 90-day action plan into Markdown and HTML reports. Re-running a brand can use an evolution mode that focuses on changed dimensions and preserves dated report versions.