|
|
@@ -0,0 +1,139 @@
|
|
|
+# Pulse - 23rd December 2025
|
|
|
+
|
|
|
+The week before Christmas, and Claude Code got a stocking stuffer: inline code review comments that actually suggest fixes. Meanwhile, Anthropic made the MCP donation to a new foundation official, Simon Willison discovered Claude can navigate Cloudflare dashboards better than most humans, and the GitHub ecosystem continued its explosive growth - the system prompts collection hit 100k stars, which tells you something about how many developers are trying to understand what's actually happening inside these tools.
|
|
|
+
|
|
|
+The throughline this week? The infrastructure for AI-assisted development is maturing fast. We're past the "cool demo" phase and into "this is how work gets done" territory.
|
|
|
+
|
|
|
+---
|
|
|
+
|
|
|
+## The Signal
|
|
|
+
|
|
|
+### [Claude Code v2.0.74 - Inline Code Review Comments](https://github.com/anthropics/claude-code/releases/tag/v2.0.74)
|
|
|
+
|
|
|
+**[anthropics/claude-code](https://github.com/anthropics/claude-code)** - Dec 19, 2025
|
|
|
+
|
|
|
+Here's a feature that sounds simple but changes the code review game: Claude Code now posts inline comments with committable suggestions directly on your PRs. Not just "this looks wrong" - actual diffs you can apply with one click. The team refined this over several commits: suggestions are now limited to 5 lines (because nobody wants a 50-line inline diff), the "Bug:" prefix got dropped (less noise), and there's a prompt encouraging you to use Claude Code for larger fixes rather than trying to cram everything into a suggestion.
|
|
|
+
|
|
|
+What makes this interesting is the philosophy shift. Instead of treating code review as a monolithic pass/fail judgment, it's now a conversation with actionable next steps. The summary comment is skipped when inline comments are posted - no more redundant "I found 3 issues" followed by... 3 issues.
|
|
|
+
|
|
|
+**Pulse insights:** This is worth stealing for our `/review` command. The pattern of "small suggestions inline, big fixes delegated to the agent" is exactly right. Consider updating our review output to match this model - inline for quick wins, explicit "run Claude Code" prompts for anything substantial.
|
|
|
+
|
|
|
+### [MCP Donated to Agentic AI Foundation](https://www.anthropic.com/news/donating-the-model-context-protocol-and-establishing-of-the-agentic-ai-foundation)
|
|
|
+
|
|
|
+**[Anthropic News](https://www.anthropic.com/news)** - Dec 9, 2025
|
|
|
+
|
|
|
+Anthropic made the Model Context Protocol official open infrastructure by donating it to a newly established Agentic AI Foundation. This isn't just a PR move - it's a signal that MCP is meant to be the universal connector between AI tools and external systems, not an Anthropic-proprietary play. The foundation structure means other companies can contribute without worrying about feeding a competitor's moat.
|
|
|
+
|
|
|
+**Pulse insights:** If you've been hesitant to build MCP servers because "what if Anthropic changes direction?" - that concern just got significantly smaller. The protocol is now genuinely open. Time to revisit our mcp-patterns skill and maybe add some server templates.
|
|
|
+
|
|
|
+---
|
|
|
+
|
|
|
+## Official Updates
|
|
|
+
|
|
|
+### [Agent Skills Specification Split](https://github.com/anthropics/skills/commit/f232228)
|
|
|
+
|
|
|
+**[anthropics/skills](https://github.com/anthropics/skills)** - Dec 16-20, 2025
|
|
|
+
|
|
|
+The skills repo got a documentation overhaul: the monolithic agent-skills-spec was split into separate authoring and client integration guides. There's now a dedicated specification website linked from the repo. The doc-coauthoring skill was also added as an example.
|
|
|
+
|
|
|
+**Pulse insights:** If you're writing skills, bookmark the new spec site. The split makes it clearer what skill authors vs. skill consumers need to know.
|
|
|
+
|
|
|
+### [SB53 Transparency Compliance Framework](https://www.anthropic.com/news/compliance-framework-SB53)
|
|
|
+
|
|
|
+**[Anthropic News](https://www.anthropic.com/news)** - Dec 19, 2025
|
|
|
+
|
|
|
+Anthropic published their compliance framework for California's SB53 (Transparency in Frontier AI Act). The interesting bit: they're being proactive about transparency requirements before they're mandated.
|
|
|
+
|
|
|
+**Pulse insights:** Mostly relevant for enterprise deployments and anyone building on Claude who needs to answer compliance questions. The framework doc itself is worth reading if you're pitching Claude to regulated industries.
|
|
|
+
|
|
|
+### [DOE Genesis Mission Partnership](https://www.anthropic.com/news/genesis-mission-partnership)
|
|
|
+
|
|
|
+**[Anthropic News](https://www.anthropic.com/news)** - Dec 18, 2025
|
|
|
+
|
|
|
+Anthropic is working with the US Department of Energy on scientific discovery applications. The "Genesis mission" framing suggests this is about fundamental research acceleration, not just document summarization.
|
|
|
+
|
|
|
+**Pulse insights:** Interesting for the capability signal - if DOE is betting on Claude for serious research, that's validation. Less directly relevant to claude-mods unless you're building science-focused agents.
|
|
|
+
|
|
|
+---
|
|
|
+
|
|
|
+## GitHub Discoveries
|
|
|
+
|
|
|
+### [system-prompts-and-models-of-ai-tools](https://github.com/x1xhlol/system-prompts-and-models-of-ai-tools)
|
|
|
+
|
|
|
+**x1xhlol** - 101,461 stars - System prompts from 30+ AI coding tools
|
|
|
+
|
|
|
+The definitive collection of leaked/extracted system prompts from Claude Code, Cursor, Windsurf, Devin, and basically every AI coding tool that exists. Just crossed 100k stars, making it one of the most popular AI repos on GitHub.
|
|
|
+
|
|
|
+**Pulse insights:** Essential research material. If you're writing agents or skills, seeing how the major tools structure their prompts is invaluable. Warning: some prompts are outdated as tools update.
|
|
|
+
|
|
|
+### [claude-code-router](https://github.com/musistudio/claude-code-router)
|
|
|
+
|
|
|
+**musistudio** - 23,835 stars - Use Claude Code as coding infrastructure foundation
|
|
|
+
|
|
|
+A wrapper that lets you use Claude Code as a foundation while controlling the interaction layer. Useful if you want Claude Code's capabilities but need custom UX or integration points.
|
|
|
+
|
|
|
+**Pulse insights:** Interesting architectural pattern for building on top of Claude Code rather than beside it. Worth exploring if our plugin approach ever feels limiting.
|
|
|
+
|
|
|
+### [wshobson/agents](https://github.com/wshobson/agents)
|
|
|
+
|
|
|
+**wshobson** - 23,253 stars - Multi-agent orchestration for Claude Code
|
|
|
+
|
|
|
+Intelligent automation and multi-agent orchestration specifically designed for Claude Code workflows. The star count suggests this is hitting a real need.
|
|
|
+
|
|
|
+**Pulse insights:** Direct competition/complement to our agent approach. Worth a close look to see what patterns they're using that we're not.
|
|
|
+
|
|
|
+### [claude-mem](https://github.com/thedotmack/claude-mem)
|
|
|
+
|
|
|
+**thedotmack** - 8,789 stars - Session memory plugin
|
|
|
+
|
|
|
+Automatically captures everything Claude does during sessions, compresses it with AI, and injects relevant context back into future sessions. Uses the agent-sdk for compression.
|
|
|
+
|
|
|
+**Pulse insights:** This is basically solving our session continuity problem from a different angle. Instead of explicit save/load, it's automatic capture and smart injection. Worth studying their compression approach.
|
|
|
+
|
|
|
+### [ccusage](https://github.com/ryoppippi/ccusage)
|
|
|
+
|
|
|
+**ryoppippi** - 9,360 stars - Usage analytics CLI
|
|
|
+
|
|
|
+Analyzes Claude Code usage from local JSONL files. See where your tokens are going, what's working, what's not.
|
|
|
+
|
|
|
+**Pulse insights:** Useful for anyone trying to optimize their Claude Code workflow. Could inspire a `/usage` command for claude-mods.
|
|
|
+
|
|
|
+---
|
|
|
+
|
|
|
+## Community Radar
|
|
|
+
|
|
|
+### [Simon Willison](https://simonwillison.net) - The practical AI columnist
|
|
|
+
|
|
|
+Two standout posts this week: "Claude in Chrome extension solves real problems" shows Claude navigating the Cloudflare dashboard to configure CORS headers - a genuinely useful demo of browser agents. "Your job is to deliver code you have proven to work" is a sharp take on professional standards for AI-assisted development.
|
|
|
+
|
|
|
+**Pulse insights:** Willison remains the best source for practical, non-hype AI coverage. The "code you have proven to work" piece should be required reading for anyone using Claude Code professionally.
|
|
|
+
|
|
|
+### [Gemini 3 Flash](https://simonwillison.net/2025/Dec/17/gemini-3-flash/)
|
|
|
+
|
|
|
+**[Google](https://deepmind.google)** - Dec 17, 2025
|
|
|
+
|
|
|
+Google dropped Gemini 3 Flash - faster and cheaper. The multimodal benchmarks are impressive, and the pricing undercuts competitors significantly.
|
|
|
+
|
|
|
+**Pulse insights:** Competition is good. Worth keeping an eye on for our `/conclave` command - Gemini's 2M context window makes it valuable for large codebase analysis.
|
|
|
+
|
|
|
+---
|
|
|
+
|
|
|
+## Quick Hits
|
|
|
+
|
|
|
+- **[n8n-mcp](https://github.com/czlonkowski/n8n-mcp)**: MCP server for building n8n workflows from Claude - 10.9k stars
|
|
|
+- **[obra/superpowers](https://github.com/obra/superpowers)**: Core skills library for Claude Code - 10.9k stars
|
|
|
+- **[claude-code-templates](https://github.com/davila7/claude-code-templates)**: CLI tool for configuring and monitoring - 13k stars
|
|
|
+- **[awesome-claude-code](https://github.com/hesreallyhim/awesome-claude-code)**: Curated commands and workflows - 18.5k stars (still growing)
|
|
|
+
|
|
|
+---
|
|
|
+
|
|
|
+## The Hit List
|
|
|
+
|
|
|
+1. **Try the new inline code review** - Test v2.0.74's suggestion feature on a real PR
|
|
|
+2. **Read the split skills spec** - The authoring guide is now separate and clearer
|
|
|
+3. **Study claude-mem's compression** - Their approach to session memory might improve our /plan system
|
|
|
+4. **Bookmark the system-prompts repo** - 100k stars means everyone else is reading it too
|
|
|
+
|
|
|
+---
|
|
|
+
|
|
|
+*Channeled by Pulse - 23rd December 2025 - 18 sources across the ecosystem*
|
0xDarkMatter