GitHub Repo Briefing
meetily
A local meeting assistant that records or imports calls, transcribes speech, and produces searchable notes and summaries without requiring a hosted service.
A first test
Run it on a short test recording with consent from everyone involved. Compare the transcript with the audio, check where files are stored, and delete the recording after the test.
codex-plugin-cc
A Claude Code plugin that adds Codex commands for review, adversarial review, rescue, and task transfer inside a Claude Code session.
A first test
Install it in a disposable repository and run /codex:review on one small diff. Compare the findings with your tests and inspect the plugin permissions before enabling more commands.
DesktopCommanderMCP
wonderwhy-er/DesktopCommanderMCP
An MCP server that gives an agent terminal control, filesystem search, file reads, diffs, and edits from a connected desktop environment.
A first test
Run it in a sandbox folder with list, search, and diff requests only. Confirm the allowed path and command policy before enabling file writes or shell execution.
system_prompts_leaks
A reference corpus of collected system-prompt files from several AI products and agents. It is useful for comparing instruction layers, but the files are not official guarantees.
A first test
Read one prompt as a text sample and label its source and date. Compare its rules with the product's public documentation, and do not copy private or product-specific text into a deployed system.
herdr
A terminal agent multiplexer that runs several coding agents in parallel and provides a shared control surface for their sessions.
A first test
Run it in a disposable repository with two read-only tasks. Check each session's working directory and process before allowing one agent to edit or commit files.
CubeSandbox
A RustVMM and KVM-based sandbox for AI agents with hardware isolation, concurrent startup, snapshots, rollback, and compatibility with E2B-style SDKs.
A first test
Launch one disposable sandbox and run a harmless command. Measure startup and cleanup, test snapshot rollback, and check the host boundary before running untrusted code.
astryx
A React 19 and StyleX design system with reusable packages, customizable tokens, and conventions intended to be understandable by coding agents.
A first test
Install the packages and rebuild one button from the examples. Check token values, generated CSS, and keyboard states before using it as a product design system.
strix
An AI-assisted penetration-testing tool that scans a local application, explains possible vulnerabilities, and helps investigate fixes.
A first test
Run strix --target against a deliberately vulnerable local app. Reproduce one finding manually, record the request and fix, and use it only on systems you are authorised to test.
claude-video
A Claude Code skill that downloads a video, extracts frames, transcribes speech, and gives the combined evidence to Claude through the /watch command.
A first test
Run /watch on a short local or permitted video and ask one timestamped question. Check the extracted frames and transcript against the source before using the answer.
OmniRoute
A local AI gateway that exposes one endpoint for many model providers and translates requests for clients such as Claude Code, Codex, Cursor, and Copilot.
A first test
Run it on the local port with one provider and one low-risk prompt. Inspect the selected model, request log, costs, and fallback behavior before adding credentials for other providers.
orca
A desktop and mobile workspace for running several coding agents in parallel, with separate worktrees and support for agents such as Codex, Claude, OpenCode, and Pi.
A first test
Run two agents on separate worktrees for the same small issue. Compare their diffs and merge one manually; check that neither agent can modify the other's branch.
page-agent
A JavaScript in-page GUI agent that controls web interfaces with natural-language instructions without a browser extension or headless browser.
A first test
Add it to a local test page and ask it to read one field. Inspect the generated actions and block submit or delete actions until selectors and confirmation checks are explicit.