Back to skill

Security audit

research-assistant

Security checks for vulnerabilities and agentic risk

Overview

This skill is a disclosed research knowledge-base tool, but it can fetch any HTTP URL, read broad local inputs, and persist extracted content indefinitely without strong scoping or cleanup controls.

Install only if you are comfortable with a local research tool that can contact any URL you give it and permanently store extracted content. Do not use it on confidential documents, internal URLs, localhost/private-network services, or PII unless you add your own network restrictions and cleanup process.

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 (2)

T09 · Insecure Skill Coding Practices

Error
Location
research-assistant.js:145
Finding

Arbitrary URL Fetching Enables Server-Side Request Forgery

Content
View full analysis
{ const isHttps = url.startsWith('https'); const client = isHttps ? https : http; const req = client.get(url, (res) => { if (res.statusCode !== 200) { reject(new Error(`HTTP ${res.statusCode}`)); return; } let data = ''; res.on('data', chunk => data += chunk); res.on('end', () => { // Strip HTML tags const text = data.replace(/<[^>]*>/g, ' ').replace(/\s+/g, ' ').trim(); const extracted = extractEntities(text); resolve({ source: url, summary: text.substring(0, 200), entities: extracted.entities, relations: extracted.relations, wordCount: text.split(/\s+/).length, extractedAt: getToday() }); }); }); req.on('error', reject); req.setTimeout(10000, () => { req.destroy(); reject(new Error('Timeout')); }); }); } ``` The URL is reached from the CLI without validation: ```javascript if (target.startsWith('http')) { entry = await extractFromUrl(target); } else { entry = extractFromFile(target); } addToKB(entry); ``` ### Technical Analysis The skill passes a user-controlled URL directly to Node.js `http.get()` or `https.get()` without validating the destination address. There is no protocol parser with a strict scheme check, hostname allowlist, DNS resolution check, or rejection of loopback, link-local, private, reserved, and IPv6-local address ranges. Consequently, a caller can make the process issue GET requests to services reachable from the host even when those services are not externally accessible. Relevant destinations can include: - Loopback services such as `127.0.0.1` or `[::1 ...[truncated 2750 chars]
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
research-assistant.js:145
Finding

Unbounded HTTP Response Buffering Allows Memory Exhaustion

Content
View full analysis
{ const isHttps = url.startsWith('https'); const client = isHttps ? https : http; const req = client.get(url, (res) => { if (res.statusCode !== 200) { reject(new Error(`HTTP ${res.statusCode}`)); return; } let data = ''; res.on('data', chunk => data += chunk); res.on('end', () => { // Strip HTML tags const text = data.replace(/<[^>]*>/g, ' ').replace(/\s+/g, ' ').trim(); const extracted = extractEntities(text); resolve({ source: url, summary: text.substring(0, 200), entities: extracted.entities, relations: extracted.relations, wordCount: text.split(/\s+/).length, extractedAt: getToday() }); }); }); req.on('error', reject); req.setTimeout(10000, () => { req.destroy(); reject(new Error('Timeout')); }); }); } ``` ### Technical Analysis Every response chunk is appended to the `data` string without a maximum response-size limit. The implementation does not inspect or cap `Content-Length`, count streamed bytes, restrict content types, or stop reading after enough content has been collected. The ten-second request timeout is not an adequate size control. A server capable of transmitting a large response within that interval can cause substantial allocation. Moreover, Node.js request timeouts generally concern socket inactivity rather than a guaranteed total transaction deadline, so a server that continues sending data may keep the operation active. After buffering the response, the code creates additional large values and collections through HTML replacement, whitespace normalization, reg ...[truncated 1324 chars]
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (5)

Tp4

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

The skill description materially overstates safety and capability boundaries while omitting critical operational scope in the manifest, including filesystem persistence and arbitrary outbound fetches. This mismatch can cause agents to trust and route sensitive data into a tool that stores content indefinitely and can access attacker-controlled or internal URLs, increasing the risk of SSRF, data retention, and unintended disclosure.

Content

No source excerpt is available for this finding.

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 32)May include surrounding context.

md
the `extractEntities()` function does not match email patterns (see line 90 of `research-assistant.js`: `// NOTE: Email addresses are intentionally NOT extracte

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
90% confidence
Finding

The skill performs sensitive operations—outbound HTTP requests, local file reads/writes, and likely environment access—without declaring any tool scope or permissions in the manifest. Even though the markdown warns about some behavior, the absence of explicit permission boundaries makes it easier for an agent or user to invoke network and persistence capabilities without informed consent or policy enforcement.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
86% confidence
Finding

The skill writes extracted content and indexes to persistent workspace files without an explicit notice or consent flow, so invoking extraction/add operations modifies local state silently. In an agent setting, silent persistence can surprise users, retain sensitive data longer than expected, and pollute shared workspace memory with attacker-controlled or confidential content.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The skill fetches arbitrary user-supplied URLs over HTTP/HTTPS without any confirmation, allowlist, or clear warning that it will contact external hosts. In an agent context, this creates an SSRF-style capability and can leak network metadata, trigger requests to internal services, or exfiltrate information through outbound requests, especially because the tool is explicitly designed to ingest remote content into persistent storage.

Content

No source excerpt is available for this finding.