News & Guides

Tips, tutorials, and insights for getting the most out of Claude Code plugins

news

A Plugin's Audit Log Failed Our Security Check. The Check Was Wrong.

The first Python hook submitted to the registry failed our hook-safety check three times: twice falsely, once correctly. The correct failure was an append-only audit log, banned by a blanket no-writes rule that would have forced the author to remove an accountability feature. Here is the real vulnerability the review did find, and why methodology v1.3 narrows the rule instead of enforcing it.

Read more →
Plugin Spotlight

Skills, MCP, and A2A: The Three Layers of the Agent Stack Each Got a Standard in 2026

For two years the agent stack was improvised. In 2026 each of its three layers got a real standard: skills as portable SKILL.md instructions, MCP as the agent-to-tool protocol (now with the Apps and Tasks extensions), and A2A as the agent-to-agent protocol under the Linux Foundation. Here is how the layers fit, what is genuinely new this year, and the trust question each one opens.

Read more →
news

One Open Method, Four Surfaces: What Verified by Sigistry Now Covers

A year ago a Claude Code extension meant a plugin. Now an agent pulls in plugins, portable skills, MCP servers, and MCP Apps that render their own UI, and each fails differently. Sigistry now runs one open, evidence-cited verification method across all four surfaces. Here is what each signal means and where it is honest about its own maturity.

Read more →
tutorial

Publishing Your MCP Server: Namespaces and Claimed Listings

You built an MCP server. Getting it discoverable and claimed across the directories takes three steps: publish to the official registry with a DNS-verified namespace, claim your listings for a verified badge, and publish a maintainer contact at /.well-known/security.txt. Here is the walkthrough, including the nginx gotcha that returns 403 for every well-known file.

Read more →
news

MCP Apps Put a UI Inside the Agent. The Sandbox Is the Floor, Not the Ceiling.

MCP Apps (SEP-1865) is the first official MCP extension: a tool can now return an interactive HTML interface the host renders in a sandboxed iframe. The sandbox is real, but it does not stop your own code from leaking secrets or rendering tool output into an XSS sink. Here is the security surface the sandbox does not cover, and a free checker for it.

Read more →
news

We Built an AI-Discoverability Audit. Then We Pointed It at Our Own Site.

seo-optimizer 1.1.0 added an llms.txt and AI-discoverability audit. The honest test of a tool is running it on yourself, so we ran it against sigistry.com. Here is the before and after across the three things that decide whether ChatGPT, Claude, and Perplexity can read your site, plus the one robots.txt line that had quietly been breaking Google.

Read more →
news

Agents Query Our Catalog Over MCP. Now Everything Else Can, Over REST.

The verified catalog already spoke MCP, which is perfect for agents and invisible to everything else. Now the same data is a public REST API described by an OpenAPI 3.1 spec, so custom GPTs, code generators, CI jobs, and dashboards can read verified plugins, portable skills, and MCP scorecards without an agent in the loop. One dataset, three surfaces, no auth.

Read more →
news

Make Your Site Mobile-Friendly Without Breaking Desktop

The scary part of going responsive is not finding the problems, it is fixing them without breaking the desktop layout that already works. responsive-design-kit audits your site's mobile responsiveness from source and scores every fix by its blast radius, so the safe changes apply automatically and the risky ones wait for your yes.

Read more →
news

ClaudeRegistry Is Now Sigistry

Same catalog, same verification, same team, new name. Why we rebranded, what Sigistry means, the one command existing users need to run, and where the registry goes next.

Read more →
Plugin Spotlight

MCP Server Smith: Scaffold, Audit, and Harden MCP Servers

A new plugin that treats your MCP server as both a protocol artifact and an attack surface: it scaffolds a spec-shaped server, scores an existing one against the stateless core and OAuth model, migrates it off HTTP+SSE, and catches confused-deputy and injection risks, all statically.

Read more →
news

WebMCP, chrome-devtools-mcp, and Your Website in 2026

Three different technologies get called 'AI robots browsing your site' - and they point in opposite directions. Here's what chrome-devtools-mcp, Gemini's auto-browse, and Google's new WebMCP actually do, and the one move that's a real first-mover opportunity.

Read more →