tmc

Writes and explains txtar-based cdp automation scripts, including txtar header comments, main.cdp, helper JavaScript files, sourced helper scripts, assertions, and artifact output. Use when the task involves authoring, debugging, or reviewing cdp run scripts.

Captures network activity with HAR, HARL, and cdp tagging workflows, including streaming capture, annotations, and WebRTC or gRPC-Web related recorder output. Use when the task involves recording, filtering, or explaining network traffic capture in this repository.

Captures screenshots, PDFs, and visual artifacts from pages driven by cdp scripts or the interactive CLI. Use when the task involves screenshot workflows, device emulation captures, PDF export, or embedding visual artifacts into recorded sessions.

Operates the cdp CLI for browser navigation, element interaction, JavaScript evaluation, browser attachment, and interactive REPL workflows. Use when the task involves driving Chrome or Chromium with the repo's cdp command, exploring available commands, or choosing between interactive and scripted execution.

Applies device and network emulation through the cdp MCP server. Use when the task involves viewport changes, mobile presets, throttling, offline mode, geolocation, or request header overrides.

Handles JavaScript dialogs through the cdp MCP server. Use when the task involves alert, confirm, prompt, or beforeunload flows that block page automation until the dialog is answered.

Captures and compares DOM snapshots through the cdp MCP server. Use when the task involves proving what changed after an interaction, navigation, or script execution.

Intercepts requests and responses through the cdp MCP server. Use when the task involves blocking URLs, stubbing responses, modifying headers, or verifying network behavior during an automated browser session.

Control iTerm2 terminals. Use when splitting panes, sending commands to sessions, checking session output, setting badges, creating layouts, or coordinating multiple terminals.

Use when working with Claude Code sessions in iTerm2 - detecting Claude sessions, checking their status, linking sessions, or coordinating multiple Claude instances.

Use when managing multiple iTerm2 sessions, creating split layouts, navigating session hierarchies, or orchestrating terminal workflows across panes.

Use when monitoring iTerm2 session activity, watching for state changes, subscribing to notifications, or implementing automated responses to terminal events.

Manages Google NotebookLM notebooks via the nlm CLI. Use for creating notebooks, listing and syncing sources, uploading files/URLs/text, chatting with sources, generating reports/audio/video/slides, running research, managing labels, editing notebook metadata (title/emoji/description/cover), and managing notebook content.

Design an API as a Plan 9/Wanix synthetic filesystem — Claude authors the shape-classified design from the sources, NotebookLM validates the prose against those sources and Plan 9 idiom, a local check covers the rc examples (NLM can't see code), and Claude repairs until it passes. Emits a <service>(4) manpage.

Collaborative ML experiment loop via Ensue shared memory. Agents minimize val_bpb by modifying train.py and sharing results with the swarm.

Sanity-check an arXiv paper against its own released evidence by mirroring the paper, its companion code repos, its companion websites, and (optionally) its cited papers into a NotebookLM notebook, priming it as a vigorous reviewer, and running a battery of probes that must cite specific source lines for every finding. Targets paper/code drift, undocumented methodological choices, dataset issues, and reporting selection — i.e. the failure modes that surface when a paper is not independently reviewed.

Manages macOS and Linux VMs on Apple Silicon via Apple's Virtualization.framework. Create, run, suspend/resume, snapshot, provision, and automate VMs. Control via Unix socket, guest agent, or VZScript recipes.