Recently updated open-source projects

Mirador lets several people refine the same Markdown-based artifact while each collaborator’s Claude Code, Codex, or Gemini session interprets changes through its own private context. A deterministic CLI handles rendering, section-level diffs, intent notes, handoffs, status, and Git-based sharing, while thin agent shims translate conversation into commands. It is asynchronous rather than real-time and expects Node 20+, GitHub repositories, and Vercel for published views.

blast-radius reads a unified diff and reports its scope, severity, and high-reach signals, including committed secrets, authentication changes, deleted tests, deployment files, migrations, and dependency manifests. It can fail a pre-merge gate at a chosen threshold, but the score is a review signal rather than a correctness verdict.

import-sentinel resolves Python and JavaScript/TypeScript imports against the standard library, dependency manifests, and local modules. It flags missing packages, one-edit typosquats of popular names, and real dependencies that are not declared. The offline standard-library tool can scan a project or diff as a pre-commit gate, with heuristics clearly separated from package-index checks.

netguard-config validates a proposed Cisco IOS-style configuration without contacting a device. It checks addressing math, duplicate IPs, routes, dangerous access and management lines, then produces an ordered rollback. The tool is a heuristic linter rather than a full IOS parser or simulator, so lab validation and human approval still matter.

intune-drift-sentry compares two exported Intune snapshots by canonical setting identity, classifies changes by risk, and writes one reverse step for each change. It can emit a receipt that checks whether the rollback covers the diff, but remains offline and read-only; applying Graph-level actions is a separate human-approved step.

intune-conflict-detector reads an offline Microsoft Graph export, normalizes settings across policy types, and intersects assignments before reporting conflicts or duplicate values. It catches custom OMA-URI and cross-type issues that the Intune view can miss. The analyzer is read-only, uses only the Python standard library, and does not contact a tenant.

Handoff turns SESSION_HANDOFF.md into an append-only register for parallel Claude Code sessions. Each closing session adds a unique, status-marked section, another workstream can resume it by code, and consumed entries become tombstones before purge. It also appends a sprint report and never overwrites a still-pending handoff.

repo-summarizer analyzes a repository's structure, dependencies, modules, data flow, configuration, and tests, then writes an Obsidian-ready architecture document plus a one-page quick reference. It works across frontend, backend, monorepo, full-stack, and mobile projects, and is invoked with /summarize.

starter-skill asks whether you need the backend, web, or mobile layers, then scaffolds a Bun workspace with Convex functions and Auth, a React/Vite TanStack web app, and an Expo SDK 56 native app. It drives setup from the CLI and starts with a local Convex deployment, making it a foundation for new full-stack projects.

Generate a CODEMAP.md at the repository root and in each source directory, with concise structure summaries, functional domains, task-to-file routes, and gated dependency hints. Files over 1,000 lines receive companion analyses that map features to line ranges. Coding agents consult these indexes before opening source files, while maintenance mode uses git diff to update only the maps affected by structural or interface changes.

semantic-web-fetch treats a page's declared profile as the authority for meaning: only bound elements count, missing facts remain unfilled, and links are described by their operation and state transition. semantic-web-write creates readable HTML with an ALPS profile, stable bindings, and local checks.

Break an AI-generated draft into factual, citation, legal, and interpretive claims, then route each claim to suitable authoritative sources. The audit reports salience, source quality, scores or qualitative legal verdicts, fairness concerns, and action items. It is tailored to England and Wales practitioners, especially healthcare law, and is not legal advice.

Research shipped app patterns through Mobbin, turn them into an approval-gated Flutter design spec, and implement production widgets with Material 3, Cupertino, or custom styling. The workflow previews every state, supports adaptive layouts, generates accessibility tests, and can add golden tests before running validation.

Walk a folder tree, extract PDF text with PyMuPDF, and write one Markdown file per subfolder. The companion cleanup script previews non-Markdown files and only deletes them when explicitly applied; conversion uses no OCR, layout reconstruction, or cloud service.

Frame one concrete problem, list the objects already in its world, and apply Action, Variation, and Removal to every object. The skill clusters duplicates, ranks promising ideas, flags open-world trade-offs, and delivers a self-contained HTML report plus a Markdown transcript.

Develop copy and visual direction around a buyer's dominant motive, then score candidates for pattern interruption, emotional intensity, curiosity, and scarcity. A five-voice council selects a direction while critics can veto cringe, incongruence, or untruth; the process also produces a design specification and treats conversion lift as a testable hypothesis.

Use this Claude skill while building or auditing a WordPress plugin for the wp.org directory. It focuses on security, REST permission callbacks, remote assets, text domains, privacy, and bundled-library collisions, with references loaded as needed. It intentionally skips themes, standalone PHP, and pure frontend work.

Run Lighthouse, axe, Playwright, link and HTML checks, security-header probes, and secret or sink scans against a production build. ShipShape ties each finding to tool output and rule names, tests responsive breakpoints, and treats missing inputs as skipped instead of silently passing them.