Recently updated open-source projects

linear-cli gives humans and agents the same shell workflow for listing queues, claiming issues, posting progress, and closing work with attached proof. The tool is a single Python file with no third-party runtime dependencies and offers JSON output for automation; it requires a Linear API key with the permissions needed for the requested actions.

Provide an App Store Connect API key, app ID, and configuration file to automate build attachment, listing metadata, screenshots, pricing, age rating, review details, subscriptions, and final review submission. The commands are designed to be rerunnable and include readiness checks for missing requirements. Apple still requires two manual web steps: publishing the app's data-collection declarations and attaching a first-time subscription to the version.

goal-forge compiles a task into a /goal contract with explicit deliverables and completion conditions, storing raw command output in a hash-chained ledger. In heavy mode, a prosecutor and three method-diverse jurors re-run checks before a unanimous DONE verdict; small tasks may use one tool-equipped auditor instead while keeping the same evidence and ledger rules. The ledger is tamper-evident, not cryptographic authentication.

Chat AI Desktop wraps a web-based AI chat interface in a small menu-bar application with builds for Windows, macOS, and Linux. Users can access the hosted chat experience or switch to an API mode and supply their own OpenAI API key from the app menu. It is an unofficial client rather than an OpenAI desktop product, and the macOS package may be blocked because its developer cannot be verified. In that case, users must explicitly open it through Finder's security prompt or build the binary themselves.

skill-optimizer provides both an agent skill for authoring eval suites and a local CLI for executing them in Docker against OpenRouter models. Cases and suites can be repeated across trials to benchmark skill behavior, debug failures, and compare reliability across models. The workbench can also expose hidden services such as MCP servers to the agent during a controlled test.

Startup Kickoff puts a startup idea through twelve evidence-based validation gates and summarizes the result with a Survival Index and an Autopsy Card or, for ideas that clear the gates, a Survival Certificate. Weaker ideas can continue into the Reforge process to develop pivot candidates. The output is meant to expose assumptions and direct further evidence gathering before a team commits to building, not replace real market validation.

second-brain stores notes, relationships, and distilled long-term memories as readable Markdown, with a rebuildable SQLite cache for full-text and graph queries. Use it as a Claude Code skill for capture and proactive recall, or run its zero-dependency MCP server so other clients can read and write the same data. It requires Python 3.8 or later, supports Git-based sync, and is intended for personal-scale knowledge rather than massive document collections or team wikis.

This repository publishes several OpenCode skills that can be installed individually for different parts of a coding workflow. The current set explores multiple interface directions before implementation, uses the Grok CLI for fresh research, queries local OpenCode history in read-only mode, and helps draft X posts from documented recommendation-system behavior. It is a collection rather than one integrated application, so dependencies, data access, and operating assumptions differ by skill.

Oh My Petcare lets cat and dog owners record symptoms, routines, vaccines, medication schedules, expenses, and other observations in plain language, then turns those local Markdown records into reminders, sitter handovers, or concise vet-visit summaries. Exports can be tailored for clinical or personal use and offered in multiple languages. It does not diagnose or prescribe; red-flag cases are directed to a clinic rather than labeled with a condition.

pmstack lets product managers build an evaluation lifecycle inside Claude without starting from an ML engineering dashboard. Its commands turn observed failures into task suites, graders, pass bars, and runnable YAML, then score a real target, review failed transcripts, and watch for regressions after model changes. PRDs, customer-signal synthesis, briefs, and launch checks are included, but the differentiator is making AI evals directly usable by PMs.

This Claude skill converts Markdown or HTML into WeChat Official Account content, preserves formatting, uploads local or remote images, supports the image-focused newspic format, and can target multiple authorized accounts. It only saves drafts and never publishes automatically; setup requires authorizing the account through the documented service and configuring its API key.

This Claude Code plugin guides offer creation from an explicit framing step through audience research, Reddit and YouTube studies, angles, competitor analysis, promises, guarantees, modules, bonuses, and value presentation. It emits Markdown and HTML artifacts with specialized agents and templates, while the final handoff can use a separate zenith-audience plugin.

This skill runs a read-only audit of a repository in full, differential, or targeted mode, checking security, bugs, workflow, UX, and types before deduplicating findings into Markdown and JSON reports. It can collect lint, type-check, test, dependency, and Git baselines; audit depth depends on the specialist skills installed.

Opus Compatibility Scanner checks instruction files, subagents, skills, settings, hooks, manifests, and Anthropic SDK call sites for known differences between Opus 4.6 and 4.7. You can run a dual-version compatibility scan, a 4.7-only scan, or symptom-led diagnosis. Findings include severity and source citations, while reports and fixes remain opt-in and no files are changed by default.

wechat-assistant-for-claude drives the Windows desktop WeChat client through screenshots, visual recognition, and simulated mouse and keyboard input. It can open a group or private chat, read recent messages, compose a context-aware reply, send it, and continue following the conversation. The skill requires an already logged-in client plus computer-use permissions, and its documented workflow was tested only on WeChat 4.1.8.29 for Windows.

token-saver gives Claude Code task-based model guidance, reserving lighter models for extraction and routine work while escalating demanding debugging or multi-file execution. It also promotes one task per conversation, sharing only relevant excerpts, and storing reusable context in CLAUDE.md, memory files, or skills. The project is advisory guidance rather than a separate billing, proxy, or model-routing service.

TUUI is a desktop LLM chat and experimentation workspace built around the Model Context Protocol. It lets users configure hosted or self-managed model endpoints and work with MCP tools, prompts, resources, sampling, elicitation, registry discovery, and MCP bundles. The application is available as a direct download, but MCP use requires a model backend with tool calling and the appropriate Node.js, Python/uv, or Docker runtime for each configured server.

ignition-skills provides two Claude skill sets for Ignition 8.1 and SepaSoft Batch Procedure. They emphasize source-backed conceptual grounding, mark unsupported facts as pending, and direct users to verify relevant details in the Designer GUI. The project intentionally excludes troubleshooting content, signature and option lists, how-to material, and customer-specific frameworks; it is maintained and installed by copying the two folders to a user-level skill directory.