Shape agent capabilities into your workflow.

xhs-card-renderer lays out Markdown as book-like vertical Xiaohongshu cards and exports multi-page PNGs at 1080×1440 with 2x DPI. HTML templates and Playwright enforce generous whitespace, strong typography, and restrained highlighting instead of poster-like decoration. It requires Node.js and Chromium.

Describe a task in plain language and Automation Level Advisor interviews you to decide whether AI should run alone, be supervised, require approval at each step, or remain an assistant to a human-led process. Built for non-developers, it explains the reasoning and creates matching Markdown and HTML project-starter reports with checkpoints, guardrails, phased rollout guidance, risks, sources, and example prompts for Claude Code or OpenAI Codex.

Citation Verification makes an AI assistant check a DOI, PMID, or arXiv identifier, read the abstract, and judge whether the paper supports the intended claim before citing it. Full text is fetched only when specific numbers or methods require it. The output includes bibliographic details, a conclusion, and a short quotation, or marks an inaccessible abstract instead of inventing one.

Map a codebase to produce a human-readable CODEDNA.md profile covering naming, comments, structure, control flow, error handling, and recurring idioms. Coding agents can then use that profile while writing new code or checking a file or diff for stylistic tells. The analysis focuses on choices formatters do not control, rather than duplicating rules already enforced by tools such as Prettier or Black.

Turn one-line LaTeX formulas into clean, printable images that can be pasted into Word, WPS, or Google Docs. The tool runs either as a Claude Code skill or a standalone local page with live preview, PNG clipboard copying, and SVG copy or download. It needs only Python 3 and a modern browser, with bundled MathJax for offline use and no LaTeX installation.

Use the corpus as a Claude Code skill, managed plugin, or MCP server in clients such as Cursor, Cline, Continue, and Zed. It guides work from elicitation and hierarchical backlogs through user stories, BDD, acceptance criteria, estimation, validation, and traceability, with Canadian English and Brazilian Portuguese editions. The project is intended for defining and checking requirements, not for implementing application code.

SEO Master gives file-based AI agents a structured workflow for technical audits, keyword and competitor research, local SEO, backlinks, and AEO/GEO readiness. It produces prioritized findings, developer briefs, and 30/60/90-day plans instead of an unranked checklist. Apify is the default data layer and requires APIFY_API_TOKEN; when data is unavailable, the skill falls back to browsing, uploaded files, or advisory analysis rather than inventing metrics.

Readme Writer separates issues that code can verify from qualities that require editorial judgment. Its Python linter checks the heading outline, image alt text, and local links, while Claude reviews the lead, value proposition, narrative flow, and reader clarity without reducing them to a score. It returns concrete findings and small proposed edits for human approval, and is intended for repository READMEs rather than llms.txt files or long-form articles.

This bundle installs an orchestrator, a section-drafting skill, and five reviewer subagents for position papers, preprints, and journal-style articles. The workflow establishes primary sources and claim-to-citation mappings before drafting, then reviews argument structure, source fidelity, vocabulary consistency, and reader clarity in parallel, with citation formatting as the final gate. The skills and reviewer agents must be installed together; a skills-only installer leaves the review loop incomplete.

Start with product screenshots or a screen recording, then answer an interactive design and voice interview. The skill builds a browser-based animated explainer, times visual cues to spoken words, and can render an MP4 up to 2K alongside separate music and sound-effect stems. Rendering requires Claude Code, ffmpeg, Node.js, Chrome, and Python; ElevenLabs is optional.

Use Rosetta to combine local conversation histories from 18 coding agents with git and project documentation, producing a cited snapshot of what is actually true now. It can also create durable architecture, product, and business decision records with provenance and status changes. The core Python CLI runs locally with no third-party dependencies and reads transcript stores and product code by default, but its normalized cache may contain secrets and should remain git-ignored.

Use Consilium for decisions with real tradeoffs, such as launching, pivoting, hiring, or accepting a commercial deal. It structures Claude's response as five contrasting advisor roles, has them cross-review the arguments anonymously, and ends with a synthesized recommendation, the main disagreement, the cost of the choice, and a concrete first step. It is not intended for factual questions, routine writing, or simple summarization.

Gemini as Tool lets Claude Code call Gemini 3.x Pro or Flash as a second opinion, builder, or judge through a Google AI Pro OAuth subscription. Its Python harness supports the Gemini CLI and Antigravity backends, including isolated parallel calls, without an API key. Users still need a signed-in CLI or agy installation, and OAuth credentials stay in the OS keyring.

This opt-in plugin has Claude Code append a short plain-language spoken summary to conversational replies, then sends only that summary to a persistent edge-tts daemon for playback. You can change voices, mute output, or interrupt speech without reading code, tables, or file dumps aloud. It requires Windows 10 or 11, Python 3.10+, and edge-tts; playback is not currently supported on macOS or Linux.

Aircraft Research OSINT Skill turns a US N-number, ADS-B screenshot, or registry excerpt into a layered ownership dossier. It combines registration history, entity records, sanctions and public-record checks, and flight-pattern analysis, then states when a principal cannot be identified. The workflow relies on web search, focuses on US registrations, and treats missing indexed data as a valid result rather than inventing an owner.

Read Arc and Zen session files to rebuild workspaces, nested folders, and pinned tabs, while converting Arc Favorites to Zen essentials; a Zen setting is required to share them across every workspace. The Python scripts run standalone or through a Claude Code skill and create backups. Zen must be fully closed first, Python 3.8+ and liblz4 are required, and bookmarks, passwords, history, and extensions need separate tools.

Give it a procedure and teaching focus to produce an anatomy-aware brief, a copy-ready image prompt, and a review checklist. It can also inspect an existing figure and flag errors or case-dependent details for human confirmation. It does not generate images and is intended for education and publication planning, not clinical decisions.

claude-gender-decoder checks job advertisements for gender-coded wording with a deterministic Python scorer based on published research, then adds contextual review to reduce false positives and catch phrasing the word list misses. It reports a coding verdict and can suggest a more balanced rewrite, with JSON output available. The project notes that its English-only, binary framing is directional: it does not cover every form of bias or replace a full inclusivity audit.