Developer Tools: Automation

Browser and File Automation

• Enable chrome://inspect/#remote-debugging to use Google's WebMCP to control your main Chrome browser instance directly, without sandboxed Playwright (from lightning fast chrome browser control)
• dev-browser CLI (npm i -g dev-browser) lets agents control browsers by writing code -- faster than screenshot-based computer use because code is precise, repeatable, composable (from dev browser sawyerhood)
• MarkItDown (Microsoft, MIT, 87K stars) converts PDF/PPTX/DOCX/Excel/images/audio/YouTube to clean Markdown -- LLMs reason better on Markdown because they were trained on it (from markitdown microsoft file converter)
• MarkItDown ships as MCP server for Claude Desktop integration; single command: markitdown path-to-file.pdf > document.md (from markitdown microsoft file converter)
• Obsidian Headless now supports Publish and Sync without the desktop app -- enables server-side vault automation (from obsidian headless publish sync)

Real-Time Issue Tracking

• Use LogRockets to identify customer issues in near-real-time, then fix via markdown-based workflow without planning decks (from real time customer issue tracking at rippling)

• Goose has strong underlying technical capabilities being overlooked despite interface limitations — Jack Dorsey calls it a "superpower" tool with the team actively pushing rapid improvements (from goose ai development tool jack endorsement)

• Paperclip Desktop is now available as a free Mac app, packaging the Paperclip orchestration platform into a standalone desktop interface (from paperclip desktop mac app launch)

• Orca IDE v1.0.80+ adds Notion-style markdown editing — described as "Notion and Obsidian had a baby" — enabling faster spec reviews and natural doc editing without fighting the editor during agentic workflows (from orca ide notion style markdown editing)

• Git repositories have a 5GB size limit; knowledge bases that grow to 2.3GB+ require migrating to SQLite backends — plan for this transition early in long-running git-based knowledge systems (from garry tan openclaw git wiki gstack)

• agent-browser (Vercel Labs, 26K+ GitHub stars) lets AI agents control a dedicated Chrome browser for web scraping — handles JavaScript-heavy sites, pages behind logins, and dynamic content; uses 82% fewer tokens than Playwright MCP (from nick spisak shared link)

• brain-ingest CLI processes video/audio locally (no data leaves machine) and generates structured Obsidian notes with frontmatter and wikilinks — extracts 12-18 claims from a 90-minute talk vs. raw transcript noise (from nyk builderz shared link)

• Two MCP servers for Obsidian integration: smart-connections (semantic search across vault) and qmd (structured queries and metadata operations) — together they give Claude both fuzzy and precise retrieval over a knowledge graph (from nyk builderz shared link)

Productivity Skills

• Allie K. Miller's /ss screenshot skill is the reference implementation for visual input — lists newest files in your screenshots folder, grabs the most recent or N most recent (/ss 4), and acts on the trailing argument (huh/fix/do this); saves ~1 hour/week (from claude screenshot skill visual processing)

• Codex's hidden Chronicle feature analyzes computer-usage patterns and offers direct productivity feedback — invoked by an explicit prompt asking what the user has been doing inefficiently (from codex app chronicle productivity analysis)

• A strong /goal prompt should specify one measurable outcome, repo/context, constraints, priority, plan, done-when criteria, verification steps, output expectations, and stop rules — turning vague "make no mistakes" instructions into a bounded mission contract (from goal command prompt engineering structure)

• Stop rules are an explicit anti-scope-creep mechanism: halt on high-impact ambiguity, surface ranked proposals before acting, and stop expanding once the stated goal is satisfied (from goal command prompt engineering structure)

• The official claude-code-setup plugin scans a project for hooks, skills, MCP servers, subagents, and automations, then configures them step-by-step — setup automation is becoming part of the Claude Code product surface, not a community-only chore (from claude code setup plugin enhancement)

• "Codex maxxing" is emerging as a named discipline around daily primitives and workflows used by Codex team members, signaling that agent productivity now has repeatable operating practices worth documenting and critiquing (from codex maxxing primitives draft)

• claude-smart stores actionable local learnings from corrections, such as replacing a hanging npm test watch command with npm test -- --run, so repeated repo-specific mistakes become reusable automation rules (from claude smart self improving plugin)

• Systematic network troubleshooting should start with baseline speedtest-cli, then check DNS resolution, MTU, packet loss, Wi-Fi interference, stale network profiles, bandwidth-heavy background processes, and mDNS before running before/after validation (from systematic network troubleshooting methodology)

Interactive HTML and Live Collaboration

• Claude Code's interactive HTML skill launches a localhost server that allows in-browser commenting on generated HTML pages — comments trigger automatic content updates and guided tours of changes, creating a Google Docs-style collaborative workflow for HTML (from claude code interactive html comments)

ML Research Automation

• ml-intern automates the post-training research loop end-to-end: arXiv reads, citation walking, HF dataset pulls, HF Jobs training, run monitoring, failure diagnosis, retraining — beat Claude Code on GPQA (32% vs 22.99% in <10h); CLI plus phone/desktop web app, with $1k GPU + Anthropic credits provisioned for early users (from ml intern automated research agent)

Trading and Finance Tooling

• See Ai Trading for the full set of AI-driven quant/trading repos (Vibe-Trading, FinceptTerminal, OpenBB, daily_stock_analysis, Kronos, qlib, freqtrade) — finance has its own dedicated topic.

Symphony and Open-Source Orchestration

• Symphony assigns a Codex agent to every open issue in a task tracker — turns issue trackers into always-on agentic systems, shifting humans from doing to reviewing and directing; open-source orchestrator for Codex (from symphony codex agent orchestrator)

agent-browser-automation

• Trick from @jlongster: have an AI agent control a browser while recording all network requests to a HAR (HTTP Archive) file, then use that HAR to derive a direct API client for the site. (from har derived api clients from browser agents)
• Once an API client is derived from a HAR file, subsequent interactions with that website can skip browser automation entirely, making repeated tasks much more efficient than driving a browser each time. (from har derived api clients from browser agents)

agent feedback loops

• The tool runs entirely locally via a small server (cli.js, server.js, sdk.js, chrome-client.js, markdown.js) with no account, database, or API key required, and no data uploaded to the cloud—addressing privacy concerns for editing sensitive docs like PRDs and product copy. (from human review visual editor skill)
• Agent waits for feedback in 10-minute windows and receives all edits/comments/deletions grouped by page immediately upon clicking Send, rather than processing feedback line-by-line during a chat conversation. (from human review visual editor skill)

agent PR tooling

• Coding agents can add before/after screenshots or videos to GitHub PR descriptions by POSTing media to GitHub's user-attachments upload endpoint, then embedding the returned URL directly in the PR body—no need to commit binary files or host media externally. (from github user attachments pr media)

web-scraping-infra

• Obscura is a Rust-based open-source headless browser engine: 70MB binary, ~85ms page load, ~30MB RAM usage, versus a Chrome-based equivalent needing ~300MB disk, ~200MB memory, and ~2s startup. (from obscura rust headless browser)
• Obscura's stealth mode randomizes per-session fingerprints (GPU, canvas, audio, battery), blocks 3,520 tracker domains, and fully masks navigator.webdriver with cleanly patched native functions to evade bot detection. (from obscura rust headless browser)
• Obscura runs real JavaScript via V8 and speaks the Chrome DevTools Protocol natively, making it a drop-in binary replacement for Puppeteer and Playwright without code changes. (from obscura rust headless browser)

automated-adversarial-testing

• A browser-based adversarial testing suite runs massively parallel adversarial checks against each new release, attempting to break it before ship, at a cost described as 'pennies.' (from parallel adversarial browser testing)