Recently updated open-source projects

agentic-envelope-skill creates printable CompleteTech-branded #10 envelope PDFs from recipient and artifact facts. It can add a logo, return address, recipient block, attention line, postage box, attachment manifest, and delivery-readiness notes, with default or per-recipient overrides. Missing address or approval information leaves the package blocked rather than ready for handoff.

agentic-customer-success-skill selects and drafts account profiles, contact maps, follow-up trackers, health scorecards, QBRs, renewal reviews, expansion briefs, recovery plans, and offboarding checklists. It keeps account work tied to verified facts, clear ownership, and next steps, while warning against wrong-contact messages, invented sentiment, and unsupported renewal assumptions. Python tools can render the artifacts as branded PDFs.

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.

agentic-email-skill selects email templates by prospect or client stage and drafts individual messages or multi-email sequences for outreach, sales, follow-up, delivery, retention, referrals, and win-back. A Python renderer can produce branded Markdown and PDF outputs. The drafts require verified recipient, offer, proof, and timing details before use, and the examples explicitly avoid fabricated metrics or client names.

The repository currently bundles two installable skills for content and product teams. The GEO blog skill guides article structure, frontmatter, schema, and a pre-publication checklist intended to make content easier for AI answer engines to cite. The Laws of UX skill applies established psychology and usability principles when designing or reviewing web apps, dashboards, landing pages, forms, and navigation. Teams can install the full collection or select an individual skill through Skillshare.

Kod Fabrikası gives each project a config-driven Claude Code routine that collects context, proposes a plan, waits for approval, applies and tests changes, writes a phase report, updates project memory, and commits. It integrates NotebookLM and Graphify through setup scripts and keeps management files under `.claude/`. The documentation and skill are in Turkish, and the setup requires those tools plus a NotebookLM login.

no-slop is a constraint skill that shapes prose during drafting instead of rewriting it afterward. Its thirteen rule groups cover inflated vocabulary, promotional tone, vague attribution, formulaic structures, participle chains, synonym swapping, excessive formatting, and other recognizable AI-writing habits. It can activate automatically for prose in Claude Code and also includes setup guidance for Cursor and Codex.

critic-overkill runs three calibrated critic agents in parallel, scores the work, and implements only changes independently flagged by at least two critics. It commits each round and stops when the target score is reached or progress plateaus, with critic pools selected for UI, APIs, copy, code, or security-sensitive work. An initial scan detects the project’s stack and verification commands.

oracle-skills-ptbr bundles seven Portuguese Claude skills for an Oracle stack: PL/SQL, APEX, ORDS, DBA operations, performance tuning, Trivadis 4.4 review, and DevOps or CI/CD. The guides cover packages and triggers, REST services, data-dictionary work, AWR and ASH analysis, naming checks, versioned database changes, and Python oracledb deployment scripts. The documented baseline is Oracle 19c, APEX 24.2, and ORDS 24.4/25.x, so teams on other versions should verify compatibility.

job-application-skill deconstructs a job posting into hard requirements and ATS keywords, then re-expresses the candidate’s real experience in the employer’s language. It produces a tailored DOCX resume plus a fit analysis, change log, and missing-information flags, without fabricating experience. The input can be a CV in several formats and a job post supplied as text, HTML, or a URL.

This Claude skill drafts App Store Connect metadata, displays live character counts, and validates every requested field before presenting it. Its playbook covers title and subtitle weighting, comma-free keyword fields, the fact that descriptions are not search-indexed, screenshot-caption OCR, trademark risks, and locale-specific wording; it makes no ranking or download guarantees.

Focusgroup dynamically tests conversational systems by combining reusable personas, feature scenarios, and a system under test. It runs real multi-turn conversations, then evaluates them with state checks, safety invariants, performance metrics, persona debriefs, or an LLM judge as a last resort. Reports cluster themes, segment failures by persona axes, and compare regressions. The current scope excludes browser interfaces, batch jobs, and pure classifiers.

Use html-to-mp4 when a clip needs crisp text, exact UI placement, or deterministic data animation rather than cinematic footage. It records an HTML, CSS, and JavaScript page in headless Chromium and converts the capture to H.264 MP4 with ffmpeg, supporting both a standalone workflow and a Claude Code skill. Local use requires Node.js, Playwright with Chromium, and ffmpeg.

Manage Anki cards from a coding session through the local AnkiConnect API. The skill proposes atomic cards for approval, then supports creating, updating, deleting, and browsing notes with consistent tags and concise fronts and backs. Anki Desktop and the AnkiConnect add-on must be running.

Agent ER spec converts SQL DDL or design Markdown into schema-checked `.erd.json` bundles and renders them as draggable, zoomable HTML diagrams. Beyond table relationships, the format preserves color-coded layers, multiple views, data-flow narratives, and design-decision logs, with examples and validation and rendering scripts included. The repository defines the data shape and a minimal interactive viewer rather than a complete architecture documentation product.

Use xiao to check status and consumables, start whole-home, room, zone, or spot cleaning, return the vacuum to its dock, and operate X20+ base-station features. The Python tool offers human-readable commands, JSON output, a local web dashboard, and an optional MCP server for agents. It supports only the Xiaomi Robot Vacuum X20+ model c102gl and sends every command through Xiaomi Cloud. Initial setup requires a Xiaomi account and Playwright Chromium; token refresh relies on saved browser login state and may require a new login when that state expires.

Students can use this Claude skill to analyze a gaokao essay prompt, refine a thesis, build a paragraph-by-paragraph outline, study a model essay, receive a 60-point critique, or run a timed practice session. Its feedback identifies problems at the paragraph and sentence level and draws from curated people, quotations, and topic patterns. The material is designed mainly for Chinese high-school argumentative writing, with secondary uses for teachers and parents.

Writers first define the topic, hook, evidence, argument, and ending path, then Claude develops the article inside that structure. Editorial AI blocks unsupported numbers, marks missing facts with placeholders, checks whether advice-oriented pieces include actionable prescriptions, and enforces the requested length. It is aimed at journalism, research, PR, and B2B writing rather than casual posts, fiction, or poetry.