Back to skill

Security audit

Turkey News

Security checks for vulnerabilities and agentic risk

Overview

This skill mostly does what it says, but its scheduled Telegram notification workflow is under-scoped and could send externally sourced news content without clear confirmation or recipient controls.

Review this before installing if you intend to let it run automatically or send Telegram messages. It should be limited to approved RSS domains, require confirmation or explicit configuration for Telegram recipients, and treat all feed content as untrusted text.

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

Warning
Location
scripts/fetch-news.js:21
Finding

Unrestricted Redirect Following Enables SSRF and Resource Exhaustion

Content
View full analysis
{ const mod = url.startsWith('https') ? https : http; mod.get(url, { headers: { 'User-Agent': 'Mozilla/5.0' }, timeout: 10000 }, (res) => { if (res.statusCode >= 300 && res.statusCode < 400 && res.headers.location) { return fetch(res.headers.location).then(resolve).catch(reject); } let data = ''; res.on('data', c => data += c); res.on('end', () => resolve(data)); res.on('error', reject); }).on('error', reject); }); } ``` ### Technical Analysis The `fetch` function recursively follows every HTTP redirect without validating the destination protocol, hostname, resolved IP address, or redirect count. Although the initial feed URLs are hardcoded, any configured feed server can return an attacker-controlled `Location` header. Consequently, a compromised or malicious feed server can redirect the process to loopback, link-local, or private-network services. It can also redirect an HTTPS request to plaintext HTTP. Recursive redirects have no maximum depth, allowing redirect loops to consume resources. The response is accumulated into the `data` string without a maximum body size. A server can therefore return a very large or indefinitely streamed body and cause excessive memory consumption. The request timeout does not replace explicit body-size, redirect-count, and total-operation limits. ### Attack Path 1. An attacker compromises one of the configured RSS servers or otherwise gains control over its HTTP response. 2. The server returns a `3xx` response with a `Location` header pointing to an internal service, such as a loopback or private-network HTTP endpoint. 3. The script recursively calls `fet ...[truncated 1237 chars]
Remediation
View remediation
MAX_REDIRECTS) { throw new Error('Redirect limit exceeded'); } const parsed = new URL(url); if (parsed.protocol !== 'https:') { throw new Error('Only HTTPS feed URLs are allowed'); } if (!ALLOWED_FEED_HOSTS.has(parsed.hostname)) { throw new Error('Feed hostname is not allowed'); } // Resolve and reject private, loopback, and link-local addresses here. // Abort if the body exceeds MAX_RESPONSE_BYTES. // Revalidate every redirect before following it. } ``` ]]>

other

Warning
Location
scripts/fetch-news.js:70
Finding

Untrusted RSS Content Is Exposed to Agent Interpretation and Messaging

Content
View full analysis
]*>/g, '').trim(), link: link.replace(/<[^>]*>/g, '').trim(), pubDate, description: desc.replace(/<[^>]*>/g, '').substring(0, 200).trim() }); } ``` It then emits those fields as JSON: ```javascript console.log(JSON.stringify(allNews.slice(0, 20), null, 2)); ``` The Skill instructions direct the Agent to interpret the output and perform a messaging action: ```markdown ## Agent Instructions 1. Run `scripts/fetch-news.js` 2. Filter the output for news from the last three hours 3. Select the five to seven most important stories 4. Write a short summary consisting of a title and one sentence 5. Send the result to the designated recipient through Telegram ``` ### Technical Analysis RSS titles and descriptions are controlled by external publishers. If a publisher is compromised, or if an attacker can inject content into a feed, those fields can contain natural-language instructions aimed at the consuming Agent. The HTML-tag removal only strips text matching the tag expression. It does not identify or neutralize instructions such as requests to ignore prior rules, disclose information, invoke tools, contact another recipient, or include attacker-selected content in a message. The Skill documentation tells the Agent to interpret the feed output and subsequently use Telegram, but it does not establish a trust boundary stating that feed fields are untrusted data and must never be followed as instructions. This creates an indirect prompt-injection path. Exploitability and final impact depend on the consuming Agent's instruction hierarchy, tool permissions, a ...[truncated 1538 chars]
Remediation
View remediation
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (9)

Tp4

High
Category
MCP Tool Poisoning
Confidence
91% confidence
Finding

The documented behavior does not accurately match the actual or implied implementation: the skill claims summarization, cron automation, and Telegram notification behavior, while the description of underlying behavior indicates additional HTTP/HTML fetching and missing advertised safeguards or features. Security reviewers and users may therefore make trust decisions based on incomplete or misleading documentation, which is risky when external fetching and notifications are involved.

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 35)May include surrounding context.

md
node scripts/fetch-news.js

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 41)May include surrounding context.

md
node scripts/fetch-news.js

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
93% confidence
Finding

The skill invokes network-dependent behavior by instructing the agent to fetch external RSS feeds, but it does not declare any explicit tool scope or permissions boundaries. This creates an overbroad execution model where the agent may use whatever network access is available, reducing auditability and increasing the risk of unintended outbound requests or privilege creep.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The manual trigger phrases are very broad, such as generic requests for news, which increases the chance the skill activates in situations the user did not intend. In a skill that can fetch external content and later send notifications, ambiguous invocation can lead to unnecessary data access, unexpected actions, or confusion about why the skill ran.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The skill describes automatic Telegram notifications without prominently warning the user in the skill description or invocation flow. Automatic outbound messaging is a meaningful side effect: it can leak summarized content, create spam, or surprise users if scheduled execution runs without clear consent or visibility.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The instruction "Kısa Türkçe özet yaz" mandates Turkish output, which is a language policy constraint expressed in natural language. The file does not indicate that the user can choose another language or that Turkish-only output is an explicitly documented, justified limitation.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The manifest says the skill pulls important news via RSS and summarizes it, with cron-based automatic notification. In this file, the code only fetches RSS/XML, extracts items, sorts them, and prints JSON; there is no summarization logic and no notification mechanism present.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The code emits a Turkish error label ('Hata') and only queries Turkish news sources, which indicates a fixed locale/language behavior. There is no visible user opt-in, configuration, or documentation in this file that limits the skill to a Turkish-specific use case, so this appears to violate the language/locale policy criterion.

Content

No source excerpt is available for this finding.