Sigrid Axis Agent Integrations

A Claude Code plugin marketplace for Sigrid Axis (documentation). Sigrid Axis brings Sigrid's analysis to AI coding assistants, agents, and other MCP-based tools:

• Guardrails: Sigrid checks the code an assistant writes, so it doesn't introduce security or other quality issues.
• Auto-fix agents: agents use Sigrid's data to fix existing quality issues at scale.

The axis plugin

The plugin autoconfigures the Sigrid Axis MCP server and ships the auto-fix skills. The bracketed argument picks the model, and phrasing is flexible: /axis:change-feedback for security problems works too.

Skill	What it does
/axis:setup	Writes the repo's Sigrid profile (.sigrid/profile.md: customer, system, conventions). Run this first.
/axis:diagnose [maintainability|architecture]	Reports the system's state for a metrics-based model and names the one fix most worth making. Never changes code
/axis:triage-findings [security|reliability|open-source]	Goes over a list of findings: false positive, accepted, needs a person, or will fix. Never changes code
/axis:autofix [maintainability|architecture|open-source|security|reliability]	Fixes what was planned, as local commits on a branch
/axis:architecture-drift	Checks your diff against Sigrid's dependency graph for new dependencies, bypassed facades, and cycles. Same check as /axis:change-feedback architecture
/axis:explore-architecture	Answers questions about the codebase structure with the architecture-explorer agent
/axis:change-feedback [maintainability|architecture|open-source|security]	Sigrid's verdict on local changes before you push: Sigrid CI run locally, or architecture drift against Sigrid's graph

Plan with diagnose or triage-findings, then fix with autofix, straight away or later. The plan is saved as a handover in .sigrid/handovers/. The skills stop at local commits and Sigrid statuses. Pushing, change requests, and issues are up to you or your pipeline.

Guardrails nudge

Before Claude reports a task done, the plugin nudges it to run the guardrails check on the production code it changed and to fix what the change made worse. It's on by default. To turn it off: /plugin → Installed → axis → Configure options → Nudge to run guardrails.

The plugin doesn't add an architecture check for you. To have Claude run one before every push, add this to your CLAUDE.md:

Before pushing code or opening a pull request, run the architecture-drift skill.
If it reports drift, fix it, or explain why the new dependency is intended.

Prerequisites

• Claude Code
• A Sigrid API token (sigrid-says.com)

Install

1. Add the marketplace:

   /plugin marketplace add Software-Improvement-Group/agent-integrations
   

2. Install the plugin. On first use you're asked for your Sigrid API token, which is stored in your system keychain.

   /plugin install axis@sigrid
   

3. Run setup in each repository. It detects your Sigrid system from the repo, asks for what it can't tell, and writes .sigrid/profile.md. Commit that file so your team shares it.

   /axis:setup
   

4. Enable auto-update: in /plugin, go to Marketplaces → sigrid → Enable auto-update.

Usage

For MCP usage, see the Sigrid Axis MCP tools documentation. Each skill has a README.md under axis/skills/.

Customization

Run /axis:setup or edit .sigrid/profile.md by hand. Besides the Sigrid settings, the profile takes plain-language guidance on how the skills should behave in the repo, such as off-limits code or commit message conventions.

Upgrading from the sigrid plugin

The plugin is now axis, in a new marketplace. Remove the old one, then follow Install:

/plugin uninstall sigrid@sigrid-ai-toolkit
/plugin marketplace remove sigrid-ai-toolkit

What changed:

• Skills are called as /axis:<skill>.
• The skills stop at local commits and Sigrid statuses. They no longer push, open change requests, or create issues.
• The per-user profile is replaced by .sigrid/profile.md in each repository. /axis:setup offers to copy your old one.
• MCP tool permissions in your settings move from mcp__plugin_sigrid_sigrid__* to mcp__plugin_axis_axis__*.
• The hook options are now Nudge to run guardrails and Nudge to use architecture-explorer. If you had turned one off, turn it off again.