Recently updated open-source projects

Use Congruence to catch semantic drift between what an AI agent, README, UI, or configuration claims and what the code actually does. It extracts verifiable claims, searches the project for evidence, classifies each result as congruent, incongruent, partial, or unverifiable, and produces a pre-deploy decision. This complements tests and technical code review rather than replacing them, and adapters are provided for many coding tools.

claude-caffeinate adds a cross-platform /caffeinate command for keeping the system awake while a long Claude Code session runs. It supports system-only or system-and-display modes, optional time limits, stop, and status commands. The implementation uses native Windows, macOS, or Linux mechanisms, stores one temporary state file, and scopes the wake lock to a process so it clears when that process exits.

Use maxym-ai-ads to build a paid-media plan from a website or review campaigns that are already running. Its Claude Code workflows cover audiences, budgets, funnels, copy, creative briefs, landing pages, and audits across Google, Meta, LinkedIn, TikTok, Microsoft, YouTube, and Apple Ads, then guide tracking setup, launch, and biweekly optimization. Most workflows need no ad-account connection; image generation requires a separate provider.

Balzac CLI exposes the Balzac content platform to developers and AI agents. It can create domain workspaces, manage keywords and content suggestions, start or rewrite articles, export them, and schedule or publish through configured CMS and webhook integrations. JSON and quiet output modes support automation, while an API key is required and several generation actions consume Balzac credits.

material-design-md turns a brand name, style description, colors, and fonts into a `DESIGN.md` for downstream design agents. The output includes Material 3 color roles, typography, spacing, shapes, component tokens, and eight ordered sections with brand-specific guidance. It can run as an Agent Skill or through standalone Python and Node.js token generators, so teams can use the document beyond one assistant.

riverpod-internals explains the mechanics of Riverpod 3.2.1, Flutter Riverpod 3.3.1, and Hooks Riverpod 3.3.1 for Claude Code users. It records 43 source-verified findings across scheduling, lifecycles, retries, widget integration, caches, and scoping, with exact file-and-line citations and a large Dart and device-test suite. The skill is for debugging behavior that public documentation may not make obvious.

high-concept-skill helps a founder, consultant, or product team turn audience language, pain points, and desired change into a single high-concept position. It interviews for context, generates three distinct angles, evaluates them against eleven criteria and anti-slop checks, then refines the selected option into structured output for hero copy, About pages, messaging, and design direction. The result is intended as a positioning backbone, not a slogan or mission statement.

CrossGen turns a difficult, trade-off-heavy problem into candidate solutions by reframing it, decomposing its causal structure, and mapping mechanisms from distant domains such as biology, physics, ecology, and the arts. Each analogy must connect relations rather than surface similarities, identify where it breaks, and produce testable predictions, implementation steps, and research pointers. It is designed for rigorous creative exploration rather than loose brainstorming.

kickoff starts a new or existing project with seven focused intake questions, a concrete definition-of-done gate, and an Open Decisions checklist for unknowns. It can scan an existing codebase, show changes before overwriting, and generate a type-specific `CLAUDE.md`, with optional Python or Node setup. The plugin is useful when later Claude sessions need a shared starting context.

drylab-wetlab-blueprint drafts a phased construction plan for platforms that combine computation or data work with experiments or engineering. It structures modules, site needs, budgets, and measurable outcomes, then checks that module, phase, and lifecycle totals reconcile through three numeric levels. A standalone `budget_check.py` script can run the consistency checks, but a passing check is not an engineering estimate or investment advice.

claude-founder-playbook organizes a startup idea into Idea, MVP, Launch, and Scale stages, each with question sets, document templates, and exit criteria. Its CLI can initialize a project, show status, and advance only when the checklist and red-team challenge are complete; unresolved decisions stay visible instead of being silently guessed. English and Chinese templates are included.

claude-recall contains two Claude Code skills. Recall searches local session transcripts and can transplant a feature from another project with path and import adaptations shown before writing; Recall Pro adds a project registry, status dashboard, focus mode, and memory routing. Both depend on the `context-mode` MCP server, and the Pro component has a separate non-commercial license boundary.

This writing skill edits reports, proposals, emails, white papers, status updates, and other professional copy for common AI signals such as inflated significance, vague attribution, filler, formulaic transitions, and promotional language. It rewrites toward specific, direct statements while keeping the tone professional and confident. The same guidance can be installed or adapted across several AI and editor platforms.

claude-project-rounds preserves continuity by adding numbered Round entries to a per-project work log and dated entries to a shared `work/CHANGELOG.md`. A new Claude Code session reads those files to recover where work stopped and why, without repeating the full exploration. The convention is domain-agnostic and fits development, SEO, operations, research, and client projects.

precision-context places Claude in a constrained mode that anchors answers to established project context, flags uncertainty, avoids vague filler, and uses a checklist for deep file and image reading. When clarification is needed, it limits the exchange to one focused question. The repository includes six test cases and requires a Claude Pro or Team plan, making it suited to technical, research, and other complex tasks.

This skill parses the current Claude Code transcript to list packages installed, files created or edited, and skills loaded during the active session. It filters out read-only actions and leaves other projects, conversations, global settings, and skills alone; optional cooldown hooks can remove only safe cache categories after inactivity, while broader cleanup supports backup and rollback.

ASCENT scaffolds a containerized Express, SQLite-WAL, React, and Vite project, then keeps it organized with ADRs, phase-gated delivery, tests, health checks, and session-resumption files. Its parent Claude skill routes work by engineering role and operation mode, while embedded ascent-* skills audit the project as it grows. The framework is deliberately opinionated, so it suits teams that want strong conventions rather than a minimal or freely chosen stack.

Get Elected guides U.S. candidates from campaign launch through fundraising, messaging, field operations, and voter engagement, with commands that generate plans, compliance checklists, speeches, emails, and volunteer materials. It includes federal guidance and state-specific coverage for eleven states. Because election and campaign-finance rules change by jurisdiction, the project is educational and should not replace advice from an attorney or election authority.