Back to skill

Security audit

Network Pharmacology

Security checks for vulnerabilities and agentic risk

Overview

The skill is coherent and not deceptive, but its mandatory HTML report has a real browser-script injection risk from report input data.

Install only if you trust the data being rendered or can patch the renderer first. The SciMiner credential handling is appropriately constrained, but generated network_report.html should be treated as active web content, especially if nodes.csv or edges.csv can come from untrusted sources or if the report will be hosted on a shared origin.

Vulnerability Patterns
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
Findings (2)

T09 · Insecure Skill Coding Practices

Warning
Location
scripts/render_network_report.py:117
Finding

Stored Cross-Site Scripting Through Unescaped Network Metrics

Content
View full analysis
`
${k.replaceAll('_',' ')}: ${typeof v==='object'?JSON.stringify(v):v}
`).join(''); if(!metrics.nodes||!metrics.edges){const outcome=document.getElementById('outcome');outcome.hidden=false;outcome.textContent='No eligible evidence network was produced. This is a recorded negative or partial result, not evidence that no biological relationship exists.';} ``` ### Technical Analysis Values originating in the input edge table, including `source_ref`, `source`, `evidence_type`, and `evidence_tier`, are incorporated into the `metrics` object without validation or HTML encoding. The generated report subsequently converts these values to strings and inserts them into the document through `innerHTML`. Escaping the report title and replacing `
Remediation
View remediation
{ const metric = document.createElement('div'); metric.className = 'metric'; const label = document.createElement('b'); label.textContent = `${key.replaceAll('_', ' ')}: `; const text = document.createTextNode( typeof value === 'object' ? JSON.stringify(value) : String(value) ); metric.append(label, text); container.appendChild(metric); }); ``` ]]>

T08 · Insecure Dependencies

Note
Location
scripts/render_network_report.py:115
Finding

Remote JavaScript Dependency Loaded Without Subresource Integrity

Content
View full analysis
{safe_title} ``` ### Technical Analysis Every generated report retrieves and executes Cytoscape from a third-party CDN. Although the dependency version is pinned to `3.30.4`, the script element does not specify a Subresource Integrity hash. Consequently, the browser has no cryptographic mechanism to verify that the downloaded resource is the exact dependency reviewed and expected by the project. This also conflicts with the renderer's description of the output as standalone because opening the report requires a live third-party resource. The report's behavior and availability therefore depend on the external CDN. ### Attack Path 1. The renderer creates `network_report.html` containing the external script reference. 2. A user opens the generated report while connected to the network. 3. The browser requests Cytoscape from the CDN. 4. If the CDN account, hosted resource, or delivery path is compromised, the browser receives modified JavaScript. 5. Because no integrity hash is present, the browser executes the modified response in the report's context. This attack requires compromise or malicious control of the external dependency delivery path; the project itself does not contain evidence that the CDN is currently compromised. ### Impact Assessment A successful supply-chain attack could execute arbitrary JavaScript in the generated report's browser context. This could manipulate reported evidence, capture report-origin data, redirect users, or initiate additional network requests. The dependency does not directly grant operating-system privileges. The practical scope is limited by the browser sandbox and the privileges of the origin fro ...[truncated 148 chars]
Remediation
View remediation
``` ]]>
Vulnerability Patterns
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (1)

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
84% confidence
Finding

The skill explicitly instructs reading local reference files and running a local renderer script, which implies file read/write capability, but it does not declare any tool scope or allowed-tools boundary. Without an explicit permission model, an agent runtime may grant broader filesystem access than intended, increasing the chance of unauthorized reads of adjacent workspace files or unintended writes/overwrites during report generation.

Content

No source excerpt is available for this finding.