Issue #3 | Wednesday, 11 March 2026

Weekly Claws #3: GPT-5.4 & The Rise of Agent-Readable Skills

โŠฑโ€•โŠฐ

Hey everyone!

This week: OpenClaw shipped a double release with ACP provenance and GPT-5.4 support, GPT-5.4โ€™s system card gave us a peek behind the safety curtain, and TanStack Intent dropped a fresh take on agent-readable skills.

Plus: Googleโ€™s new Workspace CLI, a benchmark for testing AI agents, and a community video on leveling up your OpenClaw setup.

Letโ€™s dive in. ๐Ÿพ


๐Ÿš€ Releases & News

OpenClaw Updates

  • ๐Ÿ“ฆ OpenClaw 2026.3.8 & 2026.3.7 โ€” Busy week: ACP provenance (agents know whoโ€™s talking), backup command, GPT-5.4 + Gemini 3.1 Flash-Lite support, ACP bindings that survive restarts, slim Docker multi-stage builds, SecretRef for gateway auth, pluggable context engines, Telegram fixes, HEIF image support, and 12+ security fixes.

Community & Sponsorship

Stay tuned for upcoming community spotlights!


๐Ÿ”Œ Around the Web

AI News

  • ๐Ÿ“„ GPT-5.4 Thinking System Card โ€” OpenAIโ€™s latest reasoning model with enhanced cybersecurity mitigations, building on GPT-5.3 Codex approaches. The system card is 50+ pages of โ€œhereโ€™s how we tried not to break thingsโ€ โ€” worth skimming if you want to see how frontier labs think about safety (or at least how they write about it).
  • ๐Ÿฆ TanStack Intent (alpha) โ€” Ship agent-readable skills inside npm packages. Auto-discovered from node_modules, knowledge syncs with npm update, staleness detection built in. Skills travel with your library โ€” agents automatically know how to use your code.

Essays & Analysis

  • โœ๏ธ How to Kill the Code Review โ€” Ankit Jain makes the case for ending manual code reviews in the age of AI-generated code. The future? Spec-driven development, automated verification, and human judgment upstream โ€” not staring at 500-line diffs.

Tools & Resources

  • ๐Ÿ› ๏ธ Google Workspace CLI โ€” One CLI for Drive, Gmail, Calendar, Sheets, Docs, Chat, Admin, and more. Dynamically built from Google Discovery Service with 40+ agent skills included.
  • ๐Ÿ› ๏ธ PinchBench โ€” โ€œClaw-some AI Agent Testingโ€ โ€” A success rate leaderboard for benchmarking AI agent performance.

๐ŸŒŸ Community Highlights


๐Ÿ˜‚ Fun

Software engineer, product manager, and designer as Spider-Man pointing meme โ€” "I don't need you anymore"


See you next week!

โŠฑโ€•โŠฐ