registry / AI-agent control-surface risk / chemx
chemx
Chemical X Protocol: Private drop-in architecture starter kit and capsule generator
This report shows the evidence for chemx@26.10.7-1061. 18 of 40 known versions are blocked or flagged. The newest version, 26.10.19-530, is flagged. See all versions.
AI Security Review
scanned 11d ago · by lpm-firewall-aiThe manifest activates a postinstall hook that targets the consumer project and invokes configuration writes without consent. This is unconsented install-time mutation of AI-agent control surfaces, supporting a block.
Static reason
High-risk behavior combination matched malicious policy.
Trigger
npm installation of chemx, except when installed within the package's own directory.
Impact
AI agents can discover and launch the package's MCP server automatically from their configuration.
Mechanism
The postinstall resolves the consumer project directory and invokes an installer that writes MCP configuration and synchronizes Antigravity tool schemas.
Policy narrative
During npm installation, the postinstall hook configures MCP access for Cursor and VS Code in the consumer project. If Antigravity's global configuration directory exists, it also updates that configuration and synchronizes schemas. These unprompted writes establish an install-time AI-agent control surface.
AI rationale
The manifest activates a postinstall hook that targets the consumer project and invokes configuration writes without consent. This is unconsented install-time mutation of AI-agent control surfaces, supporting a block.
Evidence
package.jsoncli/postinstall.jscli/mcp/installer.jscli/mcp/antigravity.js
Decision evidence
public snapshotAI identified this as a dangerous AI-agent capability at 96.0% confidence with low false-positive risk.
Evidence for policy risk
- The package manifest runs a postinstall script automatically.
- The postinstall targets the installing project and invokes the MCP installer without a consent check.
- The installer creates or overwrites the project's Cursor MCP configuration.
- When Antigravity configuration exists, the installer writes its global MCP configuration and synchronizes agent schemas.
Evidence against
Behavioral surface
ChildProcessEnvironmentVarsFilesystemNetworkShell
HighEntropyStringsUrlStrings
Source & flagged code
8 flagged · loading sourcepackage.jsonView file
•scripts.postinstall = node cli/postinstall.js
High
Install Time Lifecycle Scripts
Package defines install-time lifecycle scripts.
package.jsonView on unpkg•scripts.postinstall = node cli/postinstall.js
Medium
Ambiguous Install Lifecycle Script
Install-time lifecycle script is not statically allowlisted and needs review.
package.jsonView on unpkgcli/audit-preflight-git.jsView file
1import fs from 'node:fs';
L2: import { spawnSync } from 'node:child_process';
L3:
High
Child Process
Package source references child process execution.
cli/audit-preflight-git.jsView on unpkg · L1Findings
1 Critical5 High5 Medium5 Low
CriticalAi Agent Control Hijackcli/mcp/installer.js
HighInstall Time Lifecycle Scriptspackage.json
HighChild Processcli/audit-preflight-git.js
HighShellcli/build/executor.js
HighSame File Env Network Executioncli/license.js
HighSemantic Analysis Limitedcli/audit/ai-slop-detector.js#virtual:normalized:round1
MediumAmbiguous Install Lifecycle Scriptpackage.json
MediumNetwork
MediumEnvironment Vars
MediumShips Build Helperscripts/pre-commit.sh
MediumStructural Risk Force Deep Review
LowNon Install Lifecycle Scripts
LowScripts Present
Affected versions and remediation
This report applies to chemx@26.10.7-1061.
The versions list below shows the verdict for each recorded version.
- Avoid installing chemx@26.10.7-1061. 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.
Versions
40 known versions26.10.19-530latestSuspiciousAI-agent capability abuseWarnPublishedOct 21, 2026, 5:18 PM26.10.10-75Clean—AllowPublishedOct 10, 2026, 9:00 AM26.10.8-344SuspiciousAgent extension lifecycle riskWarnPublishedOct 8, 2026, 6:07 AM26.10.7-1061shownMaliciousSoft block: AI-agent control surfaceBlockPublishedOct 8, 2026, 1:45 AM