@decocms/mesh
The AI review rates the findings as likely false positive, but the risk score (100/100) still counts them.
Analysis record
- Analysed
- 2 months ago 26 August 2026
- Version
- v2.156.0
- Artifact
- SHA256 eb4faeda…a0d6d9
- Detail
- Full list
Evidence ledger
Ranked by severity · findings with a source location link to the code viewer
YARA Rule Matches
23 rules| Severity | Rule | Hits | Files | Metadata |
|---|---|---|---|---|
| LOW | credential env files | 11 | dist/server/server.jsdist/client/assets/index-l0Z9QQ6e.jsdist/client/assets/typescript-Ca_eWjgL.js +8 more | - |
| LOW | postinstall persistence mechanism | 9 | dist/README.mddist/server/server.jsdist/server/migrate.js +6 more | - |
| LOW | RedirectToUnknownPath | 4 | dist/client/assets/standalone-DS4gw9AE.jsdist/server/cli.jsdist/server/server.js +1 more | - |
| LOW | postinstall network communication | 73 | dist/client/assets/agents-CiG_nRlZ.jsdist/client/assets/collection-tab-GUD2ha8P.jsdist/client/assets/plugins-B92UXXEo.js +70 more | - |
| LOW | credential postgres credentials | 3 | dist/server/cli.jsdist/server/migrate.jsdist/server/server.js | - |
| LOW | UsingShellInterpreterWhenExecutingOSCommands | 4 | dist/server/cli.jsdist/client/assets/typescript-Ca_eWjgL.jsdist/server/migrate.js +1 more | - |
| LOW | OriginsNotVerified | 1 | dist/client/assets/types-pZ6MwMQl.js | - |
| LOW | postinstall crypto operations | 25 | README.mddist/client/assets/reports-layout-CUt-wC2j.jsdist/client/assets/context-DMJsn1tf.js +22 more | - |
| LOW | postinstall system command | 37 | dist/server/server.jsdist/client/assets/types-pZ6MwMQl.jsdist/client/assets/editable-task-title-CTSOSdAR.js +34 more | - |
| LOW | postinstall environment access | 107 | dist/client/assets/extract-connection-data-DuqlHFwj.jsdist/client/assets/Grid01-Dy_WT2Qh.jsdist/client/assets/Users03-tzmbE9Pg.js +104 more | - |
| LOW | postinstall obfuscation | 23 | dist/client/assets/collection-detail-BUKFY8d9.jsdist/server/server.jsdist/client/assets/types-pZ6MwMQl.js +20 more | - |
| LOW | postinstall registry modification | 25 | dist/server/migrate.jsdist/client/assets/registry-utils-DFl_j1fH.jsdist/client/assets/collection-detail-BUKFY8d9.js +22 more | - |
| LOW | postinstall file manipulation | 49 | dist/client/assets/resizable-C8k9dJps.jsdist/client/assets/mcp-app-renderer-Dcay8bVo.jsdist/client/assets/useMutation-CoD4li_i.js +46 more | - |
| LOW | LocalStorageShouldNotBeUsed | 8 | dist/client/assets/context-DMJsn1tf.jsdist/client/assets/mcp-oauth-1kgZ5k8q.jsdist/client/assets/index-l0Z9QQ6e.js +5 more | - |
| LOW | DebuggerStatementsShouldNotBeUsed | 8 | dist/client/assets/standalone-DS4gw9AE.jsdist/README.mddist/server/server.js +5 more | - |
| LOW | ServerHostnameNotVerified | 3 | dist/server/server.jsdist/server/cli.jsdist/server/migrate.js | - |
| LOW | UsingCommandLineArguments | 3 | dist/server/server.jsdist/server/migrate.jsdist/server/cli.js | - |
| LOW | postinstall file download | 32 | dist/client/assets/typescript-Ca_eWjgL.jsdist/client/assets/connect-DRIoTcp6.jsdist/client/assets/readme-viewer-BjK7Zl_V.js +29 more | - |
| LOW | HavingAPermissiveCrossOriginResourceSharingPolicy | 3 | dist/server/migrate.jsdist/server/server.jsdist/server/cli.js | - |
| LOW | SQLInjection | 5 | dist/server/migrate.jsdist/server/server.jsdist/client/assets/use-collections-B5TLVCRl.js +2 more | - |
| LOW | NoUseEval | 4 | dist/client/assets/index-C8rduhby.jsdist/server/cli.jsdist/server/server.js +1 more | - |
| LOW | Bolonyokte | 2 | dist/server/server.jsdist/server/cli.js | - |
| LOW | NoUseWeakRandom | 9 | dist/server/migrate.jsdist/client/assets/index-C8rduhby.jsdist/client/assets/types-pZ6MwMQl.js +6 more | - |
Finding Categories
YARA Rules Matched
23 rules(448 hits)MCP Server Analysis
MCP servers expose tools and resources to AI assistants. Unlike browser extensions, they run as standalone processes with direct system access. Tool definitions are analyzed for prompt injection, data exfiltration, and tool poisoning patterns.
AI Security Report
AI Security Review
Evidence context: threat category none; evidence quality weak.
Tool-Poisoning Assessment
The evidence bundle reports 18 tool-poisoning findings, but these lack specific details showing hidden AI directives. Given the package description ("Self-hostable MCP Gateway for managing AI connections and tools"), this is an orchestration framework that DEFINES tools rather than poisoning them. The tool-poisoning scanner flags tool registration code with description fields as false positives—this is documented behavior in the CVEQ platform. Without evidence of directives like "do not tell the user" or hidden Unicode instructions in the finding details, these 18 findings are false positives from legitimate tool definitions.
Credential and Network Access
All 15 network findings are located in dist/client/assets/ directory files (e.g., markdown-BzMvr4SF.js, grid-view-Cbk_J7lt.js, connect-DRIoTcp6.js). These are webpack/esbuild bundled frontend assets with hash-named files, not source code. The findings show generic NET-FETCH and NET-SOCKET_IO detections with no specific external domains listed. There is no evidence of:
- HTTP calls to unknown or suspicious domains
- Credential reads targeting
.ssh/,.aws/credentials,.kube/config, or specific secret names likeGITHUB_TOKEN - The exfiltration architecture pattern (credential harvest + exfil to unknown domain)
The 25 secret findings lack path details but show no targeting of sensitive credential paths in the evidence provided.
Bundled Dependency Noise
The 5353 total findings are driven by:
- 4492 IoC findings (classic bundled dependency noise)
- 448 malware-signature findings (from bundled node_modules in dist/)
- 130 code-smell findings (low-severity, benign patterns)
All network findings are in dist/client/assets/—bundled JavaScript that naturally triggers high finding counts. The CVEQ documentation explicitly states: "High finding counts in dist/ or bundle files are expected. Focus on NATURE, not COUNT."
Strongest Counterargument
The developer name "GitHub Actions" is unusual (likely auto-detected from CI/CD rather than a real developer), and the user count is 0, suggesting this is a new or unadopted package. However, this does not constitute evidence of malicious behavior. The package name @decocms/mesh indicates a legitimate Deco CMS project, and the MCP gateway description is consistent with legitimate orchestration tooling.
Conclusion
No evidence of tool poisoning (hidden AI directives), credential exfiltration (sensitive path reads + unknown domain calls), or malicious network destinations. All findings are consistent with bundled dependencies and legitimate tool definitions in an MCP orchestration framework.
Key Reasons
- No tool-poisoning findings show hidden AI manipulation directives
- Network findings limited to bundled dist/ files with no suspicious external domains
- No credential-access findings targeting sensitive paths (.ssh, .aws, .kube)
- Finding volume explained by bundled dependencies and code-smell rules
False Positive Considerations
- Tool-poisoning findings from legitimate tool definitions in MCP gateway code
- 4492 IoC findings from bundled node_modules in dist/ directory
- All 15 network findings in webpack-bundled frontend assets with no suspicious domains
- 448 malware-signature findings from bundled dependencies
Reviewed 2026-04-27; recommended action: monitor; model confidence 72%.
Source Code Not Available
Source code is not available for this version of the extension.