ClawHub
Back to skill

Security audit

PMC Harvest

Security checks for vulnerabilities and agentic risk

Overview

The skill mostly matches its PubMed Central harvesting purpose, but it exposes broader network access than it documents.

Review before installing in environments with sensitive internal network access. Prefer a version that does not export a generic HTTP helper, restricts requests and redirects to eutils.ncbi.nlm.nih.gov and pmc.ncbi.nlm.nih.gov, and enforces request timeouts and response-size limits.

A.I.G

Vulnerability Patterns
  • 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
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
Findings (1)

T09 · Insecure Skill Coding Practices

Warning
Location
lib/api.js:25
Finding
Unrestricted HTTPS Requests and Unvalidated Redirect Destinations<![CDATA[ ## Vulnerability Details **File Location**: `lib/api.js`, lines 25–65; exported at lines 246–255 **Vulnerability Type**: Server-Side Request Forgery (SSRF) and unbounded network response handling **Risk Level**: Medium ### Vulnerable Code ```javascript function httpGet(url, redirects = 0) { return new Promise((resolve, reject) => { if (redirects > 5) { reject(new Error('Too many redirects')); return; } const parsedUrl = new URL(url); const options = { hostname: parsedUrl.hostname, path: parsedUrl.pathname + parsedUrl.search, method: 'GET', headers: { 'Accept': '*/*', 'Accept-Encoding': 'gzip, deflate', 'User-Agent': 'PMC-Harvest/1.0' } }; https.get(options, (res) => { if (res.statusCode >= 300 && res.statusCode < 400 && res.headers.location) { const location = res.headers.location; const redirectUrl = location.startsWith('http') ? location : new URL(location, parsedUrl.origin).href; return httpGet(redirectUrl, redirects + 1).then(resolve).catch(reject); } let stream = res; const encoding = res.headers['content-encoding']; if (encoding === 'gzip') { stream = res.pipe(zlib.createGunzip()); } else if (encoding === 'deflate') { stream = res.pipe(zlib.createInflate()); } let data = ''; stream.on('data', chunk => data += chunk); stream.on('end', () => { if (res.statusCode >= 200 && res.statusCode < 300) { resolve(data); } else { reject(new Error(`HTTP ${res.statusCode}`)); } }); stream.on('error', reject); }).on('error', reject); }); } ``` The unrestricted helper is also exposed as part of the public module API: ```javascript module.exports = { searchPMC, getSummaries, fetchFullText, fetchAbstract, parseJATS, harvestJournals, httpGet }; ``` ### Technical Analysis Th ...[truncated 2821 chars]
Remediation
<![CDATA[ ## Remediation Suggestions 1. **Remove the generic helper from the public API** - Do not export `httpGet` unless arbitrary outbound requests are an explicit, documented requirement. - Expose only task-specific methods such as `searchPMC` and `fetchFullText`. 2. **Enforce an exact destination allowlist** - Permit only `https:`. - Permit only the documented hosts: - `eutils.ncbi.nlm.nih.gov` - `pmc.ncbi.nlm.nih.gov` - Reject URL credentials, unexpected ports, IP literals, malformed hostnames, and hostnames that merely end with an allowed string. 3. **Validate every redirect** - Resolve relative redirects with `new URL(location, currentUrl)`. - Apply the same scheme, hostname, port, and address validation before following each redirect. - Prefer rejecting cross-origin redirects unless they are explicitly required and allowlisted. 4. **Restrict private and special-purpose addresses** - Resolve the destination hostname and reject loopback, link-local, private, multicast, and other special-use IPv4 and IPv6 ranges. - Account for DNS rebinding by verifying the address actually used for the connection. 5. **Bound resource consumption** - Add connection and response timeouts. - Abort the request after a fixed compressed or decompressed byte limit. - Handle decompression errors and oversized decompressed output. - Destroy the response stream before following a redirect. 6. **Construct requests with the validated URL** - Pass the validated URL object to `https.get` or explicitly include the validated protocol and port. - Revalidate immediately before each outbound connection. ]]>

SkillSpector

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)

Lp3

Medium
Category
MCP Least Privilege
Confidence
87% confidence
Finding
The skill clearly instructs users to invoke Node.js code that performs outbound network access to NCBI endpoints, but the manifest does not declare any explicit tool scope such as permissions or allowed-tools. This creates a least-privilege and transparency gap: an agent or reviewer cannot tell from the top-level declaration that executing the skill will make network requests, which can lead to unintended data egress or policy bypass in environments that rely on manifest-declared capabilities.

Static analysis

No suspicious patterns detected.