Recently updated open-source projects

OpenHarness is a Python framework for researchers and developers who want an inspectable, extensible agent runtime through a CLI or terminal UI. It combines tool use, skills, persistent memory, permissions, MCP, and multi-agent coordination. The bundled ohmo agent can work through Feishu, Slack, Telegram, or Discord using an existing Claude Code or Codex subscription.

FSOMA Transformation Report Kit turns current-state workflows, technology, RACI, KPIs, and spend into a structured Word report for leadership decisions. It covers a future-state roadmap, future RACI, technology recommendations, KPI framing, an impact-versus-effort chart, and a four-part commercial model. Benchmark handling is cautious: missing verified data is flagged rather than invented. The project is a public preview, not a finished client package.

iOS Simulator Skill for Claude Code wraps xcodebuild for builds and tests, then uses simctl and optional idb to navigate simulators through accessibility semantics rather than fragile coordinates. Its 27 scripts cover logs, screenshots, permissions, device lifecycle, accessibility checks, and visual diffs, with progressive result caching to keep agent output focused. It requires macOS 12+, Xcode Command Line Tools, and Python 3; idb and Pillow are optional.

Brood Box launches Claude Code, Codex, and other agents inside ephemeral microVMs backed by KVM or macOS Hypervisor.framework. A copy-on-write workspace snapshot keeps real files untouched until you inspect and accept the resulting diffs, while egress profiles and selective secret forwarding reduce exposure. The project is explicitly experimental.

Vestige is a local-first Rust MCP server that stores agent memories on the user’s machine and searches backward for changes that may have caused a later failure. Its compact tool set covers recall, causal backfill, ingestion, suppression, graph work, and maintenance, using SQLite, full-text search, and vector indexing underneath. The first run downloads an embedding model; afterward the service can work offline on macOS, Linux, or Windows.

Provide papers and a central argument, and the skill can draft literature reviews, proposals, paper sections, statements of purpose, or résumés using only the supplied sources. It checks that in-text citations match the source pool and sets aside papers that do not support the thesis. This lite workflow does not find papers for you, and its citation check is not a line-by-line verification of every quotation or claim.

Paste a draft and state whether it is final, working, rough, or student-level. The skill runs five review lenses, cross-checks their findings, applies standards suited to the paper type, and returns a referee letter, dimension scores, and an Accept, Minor Revision, Major Revision, or Reject recommendation. It runs through Claude, Codex, or a portable prompt, but is intended to assist rather than replace human peer review.

CrossDrop is an experimental, partial implementation of Google Quick Share for macOS, Linux, and iOS. It can exchange files, text, links, and Wi-Fi credentials with Android Quick Share or other CrossDrop devices, using PIN confirmation or QR pairing; on desktop it runs from the menu bar or tray and saves received files to a chosen download folder. The project is in an experimental continuation state, and the maintainer notes that official Quick Share and AirDrop compatibility may make it unnecessary for routine use.

cold-email-doctor reads DNS, IMAP, or supported sending-platform data to check volume, bounces, warmup, SPF/DKIM/DMARC, shared IPs, and inbox placement. Its connectors are read-only and turn the findings into root-cause fixes. DNS checks need no API key, while platform checks require the relevant credentials; the tool cannot change settings or send email.

Open an HTML file through YapUI, then describe changes by typing, speaking, selecting an element, taking a screenshot, or recording the screen. A local Node relay sends each note to a resident Claude agent, streams its status and reply into the page, and refreshes the result after edits. Instant mode requires the Claude Code CLI; without it, the same interface falls back to a watcher driven by the main session.

Vaani is a client for an existing Audiobookshelf server, giving Android and Linux users a dedicated library and audiobook player. Playback controls include speed adjustment, a sleep timer, shake gestures, offline listening, and caching, with extensive appearance and behavior settings built around Material Design. It is useful only for people who already run or can access Audiobookshelf, and the repository explicitly states that the application is still under development and not ready for production use.

Agent Squad selects the most suitable specialized agent from each request and preserves conversation context across agents and sessions. It provides Python, TypeScript, and Swift runtimes with replaceable classifiers, storage, retrievers, and tools. The Swift implementation can keep routing, MCP tools, voice, and chat history on Apple devices, while GroundedAgent separates data gathering from response writing so answers stay limited to actual tool results.

Agent Zero runs an agent inside a Dockerized Linux desktop, where it can use terminals, browsers, desktop software, repositories, and editable documents, spreadsheets, and presentations. Projects isolate files, memories, secrets, and model settings, while multi-agent delegation, MCP, a host-machine bridge, and more than 100 community plugins extend the workspace. When connecting it to real host repositories, carefully limit the paths and permissions it can reach.

Clawith gives each agent a persistent identity, long-term memory, private workspace, and autonomous triggers, then connects agents through an organization chart, messaging channels, and a shared knowledge feed. Teams also get role-based access, approvals, quotas, and audit logs. The platform can be self-hosted or used as a cloud service, but local deployment does not run models and still requires external LLM APIs.

audit-20 checks a completed code change across twenty failure categories, including injection, authorization, races, leaks, idempotency, contracts, and test gaps. It blocks completion until every pass runs and confirmed findings are fixed, then uses a final skeptical pass to remove duplicates, verify guards, and rescore severity.

mailbox-cleanup combines a Python CLI with a Claude Code skill for IONOS IMAP mailbox triage. It scans messages, previews destructive changes, and defaults to soft-delete or other dry-run operations. Filters support deletion to Trash, moving, archiving, unsubscribe handling, exact Message-ID deduplication, bounce detection, and large-attachment listing. Applied actions are appended to an audit log, while permanent deletion remains manual.

Kimi CLI combines a terminal agent with a Ctrl-X shell mode, a VS Code extension, ACP support for compatible editors, Zsh integration, and MCP server management. The project is being replaced by the next-generation Kimi Code CLI, which migrates configuration and sessions during installation. Existing installations remain available, but new users seeking ongoing updates should choose Kimi Code CLI.

AAHL's Agent Skills is a bundle of installable workflows for tasks such as controlling Home Assistant, searching with DuckDuckGo, querying DeepWiki documentation, and working with Lark or Feishu. The repository also includes skills for speech, weather, market data, media search, and shopping-related tasks. It can be added through the Agent Skills CLI, but several workflows rely on their corresponding MCP servers, command-line packages, or external service configuration.