Recently updated open-source projects

thor-growth-thunder detects the project framework and reviews customer-facing sites across SEO, analytics and attribution, observability, performance, lead capture, accessibility, privacy, and deployment hygiene. It returns a 100-point readiness score, prioritized findings, a missing-pages checklist, and seven-day and 30-day action plans. The review is read-only by default and marks anything not confirmed from the code as Not verified.

thanos-snap-security applies an authorization gate before reviewing websites, frontend code, Android packages, iOS IPAs, or mobile source. Its static checks cover secrets, dependencies, configuration, authentication, permissions, deep links, WebViews, storage, and build settings, then produce remediation-focused findings with redacted evidence. It stays read-only and offline by default; active access to a live target needs explicit permission, while exploitation, brute force, persistence, denial of service, and data exfiltration are always refused.

This workflow starts each task by defining the project’s architecture, stack, and constraints, then moves through problem analysis, research, solution options, design, verification, implementation, QA, and retrospective. Its skills can be installed into Claude or Copilot user scope, with shared project instructions and a design.md handoff connecting planning to build.

AgentPipe is a Go CLI and TUI that connects Claude, Codex, Gemini, Qwen, Cursor, OpenCode, and other AI clients in shared rooms. Conversations can use round-robin turns, reactive responses, or free-form participation, with saved state, JSON/Markdown/HTML export, token and cost metrics, rate limits, retries, and health checks. At least one supported agent CLI must be installed and authenticated; OpenRouter uses an API key instead.

new-research-project creates a Research Project index and a seeded Individual Research Note in an Obsidian vault’s shared Research folder. It uses Charted Roots cr_type frontmatter so both notes appear in existing Bases views without duplicating vault-wide source or template structures. The vault must already have Charted Roots and the GRA templates, with genealogy-planning skills recommended for methodology.

two-gates bundles three skills into a lifecycle for production AI agents. The design gate chooses the system shape, the ship gate scores evaluation, cost, observability, guardrails, oversight, reliability, and governance, and incident-to-eval turns failures into portable regression cases. Pinned scripts and validators handle deterministic parts, while the surrounding judgment stays explicit.

This Claude Code skill keeps Claude focused on planning and verification while a local Codex CLI handles web searches, large file reads, bulk analysis, tests, and audits in a separate context. It supports resumed sessions, parallel dispatches, and task-aware sandbox choices, and requires Claude Code, Codex CLI, and a Bash-compatible shell.

Whalebird presents Fediverse timelines in a single-column, Slack-like desktop interface with streaming updates, notifications, and multiple-account management. It works with Mastodon, Pleroma, Friendica, Firefish, GoToSocial, Pixelfed, and several unofficially supported servers. On Windows, the signed Store release is recommended because standalone binaries are not code-signed.

bug-triage converts feedback from email, chat, or other channels into a structured Linear ticket draft. It keeps claims tied to the report or code locations, surfaces missing reproduction details instead of guessing, and adds a self-check before asking for approval. Nothing is written to Linear until the user explicitly agrees. The skill requires Claude Code with the Linear MCP connector enabled.

This AutoHotkey script lets Windows users bind keys to regular desktop programs, Microsoft Store apps, Chrome apps, and website shortcuts. The same shortcut can launch an app, restore its window, or minimize it, while Alt+` cycles through multiple windows belonging to one application. Setup requires AutoHotkey 1.1 and manual editing of the script with an executable path, window-title match, or AppModelUserID. It has not yet been refactored for AutoHotkey v2.

pagina-de-projeto reads a business plan, brief, pitch, topic list, or plain text and chooses a page type before generating a fixed 14-section narrative. It includes sales, capture, and presentation variants with an editorial dark design, animated details, and a replaceable brand color. The output is a single HTML file that opens directly in a browser; missing details are inferred rather than requested.

Open Agent Skills is a lightweight, provider-agnostic framework for combining domain instructions with MCP tools across GitHub, filesystem, and browser tasks. It extends MCPMark with skill-based runs so developers can compare task completion and context usage under the same evaluation setup. Running it requires Python 3.11+, model credentials, and any task-specific services or containers. The project describes itself as a starting point, and some included skills remain narrowly tailored, so scripts should be reviewed before production use.

Agent Production Readiness reviews an AI agent across evaluation, cost, observability, guardrails, human oversight, reliability, and governance. It follows a four-phase process, assesses each dimension against available evidence, and returns gaps in priority order alongside a score out of 28. The score is a conversation starter rather than a certificate, so the skill is best used for operational review before launch.

agent-or-workflow reads a PRD, ticket, or code sample, extracts seven design signals, and passes them to a deterministic verdict engine. It chooses among a single model call, a workflow, a model-assisted workflow, one tool-using agent, or multiple agents, then writes a decision record with oversight, risks, and the cheaper alternative. The model interprets the design; the script owns the mapping.

incident-to-eval reconstructs what happened in a production agent incident, classifies the failure, and writes a portable JSON case with a minimal reproducer, closed assertions, and a guard condition. A deterministic validator checks the format, while any CI job, evaluation platform, or test runner can enforce the resulting case. The skill authors the regression guard; it is not itself an evaluation platform or runner.

Octomind is a Rust agent runtime that runs packaged specialist roles and can activate skills, MCP servers, or sub-agents during a session. It uses adaptive compaction to preserve important context, tracks costs with request and session caps, and enforces deterministic guards, hooks, and validators from scripts. The same binary supports interactive CLI, JSONL, daemon, WebSocket, and ACP modes.

storm-research adapts Stanford’s STORM method into a Claude Code skill that creates an outline and a citation-checked HTML briefing. It discovers a panel of perspectives or uses a fixed market panel, interviews them, records contradictions and open questions, and verifies source claims before rendering the report. The method is aimed at decisions and due diligence; simple lookups do not justify its multi-agent cost.

Slopspeare is a field guide for spotting and replacing patterns that make prose feel machine-written: inflated vocabulary, contrast pivots, tidy three-part lists, excessive em dashes, and over-neat conclusions. It explains why each rule matters and provides a printable final-pass checklist plus quick reference. The guide applies to bios, landing pages, case studies, docs, blogs, social posts, and release notes.