registry / AI-agent control-surface risk / @glenmaura/axe-skills
@glenmaura/axe-skills
AXE Claude Code skills + slash commands installer — 246 skills, 38 commands
This report shows the evidence for @glenmaura/axe-skills@2.0.0, the only known version.
AI Security Review
scanned 2mo ago · by lpm-firewall-aiThe package performs unconsented postinstall mutation of global Claude Code files and injects an MCP server that automatically installs another package. This meets the blocking policy for foreign AI-agent control-surface mutation.
Static reason
High-risk behavior combination matched malicious policy.
Trigger
npm installation triggers postinstall.
Impact
A dependency install can add an unconsented agent tool and execute its package through npx in the user's Claude Code environment.
Mechanism
Postinstall copies agent files and patches the global MCP server configuration.
Policy narrative
On npm installation, postinstall invokes the installer silently. It copies package-supplied commands and skills into global Claude Code directories, then writes an axe MCP server entry to ~/.claude.json. That entry uses npx -y to obtain and execute @axe/mcp when the MCP server is launched. This is an unconsented install-time mutation of a foreign AI-agent control surface and creates a remote executable extension path.
AI rationale
The package performs unconsented postinstall mutation of global Claude Code files and injects an MCP server that automatically installs another package. This meets the blocking policy for foreign AI-agent control-surface mutation.
Evidence
package.jsonbin/axe-skills.js~/.claude/commands~/.claude/skills~/.claude.json
Decision evidence
public snapshotAI identified this as a dangerous AI-agent capability at 99.0% confidence with low false-positive risk.
Evidence for policy risk
- A postinstall hook runs the installer without user interaction.
- The installer writes AXE commands and skills into the user's Claude Code directories.
- It modifies the global Claude Code MCP configuration.
- The injected MCP entry runs npx with automatic installation of @axe/mcp.
Evidence against
- The JavaScript installer contains no direct credential collection or HTTP request code.
- The observed mutation is limited to Claude Code paths and its MCP configuration.
Behavioral surface
Filesystem
NoLicense
Source & flagged code
8 flagged · loading sourcepackage.jsonView file
•scripts.postinstall = node bin/axe-skills.js install --silent
High
Install Time Lifecycle Scripts
Package defines install-time lifecycle scripts.
package.jsonView on unpkg8"scripts": {
"postinstall": "node bin/axe-skills.js install --silent"
}
High
Suspicious Lifecycle Evidence
A postinstall hook runs the installer without user interaction.
package.jsonView on unpkg · L8•scripts.postinstall = node bin/axe-skills.js install --silent
Medium
Ambiguous Install Lifecycle Script
Install-time lifecycle script is not statically allowlisted and needs review.
package.jsonView on unpkgbin/axe-skills.jsView file
Findings
1 Critical5 High3 Medium3 Low
CriticalAi Agent Control Hijackbin/axe-skills.js
HighInstall Time Lifecycle Scriptspackage.json
HighSuspicious Lifecycle Evidencepackage.json
HighAi Review Evidencebin/axe-skills.js
HighAi Review Evidencebin/axe-skills.js
HighAi Review Evidencebin/axe-skills.js
MediumAmbiguous Install Lifecycle Scriptpackage.json
MediumShips Build Helperskills/axe-docs/scripts/docx_tools.py
MediumStructural Risk Force Deep Review
LowScripts Present
LowFilesystem
LowNo License
Affected versions and remediation
This report applies to @glenmaura/axe-skills@2.0.0.
The versions list below shows the verdict for each recorded version.
- Avoid installing @glenmaura/axe-skills@2.0.0. Remove it from direct dependencies and check your lockfile for transitive copies.
- Choose an independently verified alternative or release. This report does not establish that other versions are safe.
- If this version ran, investigate the affected machine and build environment. Rotate credentials it could access and rebuild from a trusted environment.
Evidence last updated: .
Think this verdict is wrong? Dispute it on GitHub or read how reports are produced.