Back to skill

Security audit

Sumo Torrent

Security checks for vulnerabilities and agentic risk

Overview

This torrent-search skill is not clearly malicious, but it asks agents to reuse a personal browser profile/session and can directly add torrents to qBittorrent with weak credential handling.

Review carefully before installing. Use only an isolated browser profile dedicated to this site, do not reuse a personal daily profile, change the qBittorrent password, keep the Web UI bound to localhost, avoid passing passwords on the command line, and require explicit confirmation before adding any magnet or torrent file to a client.

Vulnerability Patterns
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
  • 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)

T05 · Unauthorized Access and Privilege Escalation

Warning
Location
SKILL.md:18
Finding

Use of a Personal Authenticated Browser Profile Violates Least Privilege

Content
View full analysis
", compact=true) // Step 3: 從搜尋結果點進詳情頁 browser(action="act", profile="my-daily-chrome", targetId="", request={"kind": "click", "ref": ""}) ``` ### Technical Analysis The skill deliberately directs the agent to reuse a personal, persistent Chrome profile because that profile contains an authenticated Cloudflare session. This violates least-privilege principles because the torrent-search task only needs a session dedicated to BT4G, not a general browser profile that may contain unrelated cookies, browsing state, saved permissions, or authenticated sessions. Although browser same-origin controls ordinarily prevent one website from directly reading cookies belonging to unrelated domains, operating an automated agent in a broadly privileged profile still increases the exposure surface. The agent can be instructed to navigate, click links, follow redirects, or interact with other resources while retaining all of the profile's existing authentication state. A dedicated, restricted profile would provide the required Cloudflare session wit ...[truncated 1586 chars]
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
add_to_qbittorrent.py:20
Finding

Hardcoded Default Credentials, Plaintext Authentication, and Session Identifier Logging

Content
View full analysis
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • 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 (15)

Context-Inappropriate Capability

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The skill explicitly instructs use of a real browser profile with an existing session cookie to bypass Cloudflare, which leverages authenticated browser state unrelated to the skill’s stated purpose of torrent searching. This creates clear risk of unauthorized use of another user’s session, privacy exposure, and abuse of local authenticated context to access or interact with sites in ways normal tooling cannot.

Content

No source excerpt is available for this finding.

Ssd 3

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

The skill tells the agent to use a browser profile containing another person’s session cookie and continue operating in that authenticated context. This is a strong indicator of credential/session misuse because it instructs the agent to inherit trust and privileges from a real user session rather than obtaining access through approved, isolated means.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The skill is presented as a search utility, but its workflow escalates into facilitating torrent acquisition and direct import into qBittorrent. This mismatch reduces informed user consent and can cause the agent to perform higher-risk actions than the declared scope, including initiating downloads of potentially infringing or unsafe content.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill enables direct side effects—writing output files and adding magnet links to qBittorrent—without a prominent pre-action warning or explicit confirmation gate. In context, this is more dangerous because torrent content may be illegal, malicious, or privacy-sensitive, and direct client submission can immediately begin network activity and content acquisition.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

Using a preexisting browser profile with session cookies exposes private authenticated state to the agent and may allow access, tracking, or actions under another person’s identity. The absence of privacy, consent, and credential-handling safeguards makes this especially dangerous because the skill normalizes using stored session material as a bypass mechanism.

Content

No source excerpt is available for this finding.

Ssd 4

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The workflow chains privileged browser-session use with data extraction and automatic delivery into a torrent client, creating a stepwise path from unauthorized access context to operational action. This kind of guided escalation is dangerous because it reduces friction for misuse and converts sensitive session-derived access into downstream automated behavior.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The documented workflow expands from search and link generation into taking consequential actions such as adding downloads to qBittorrent or writing output files. That exceeds the declared scope and turns a retrieval skill into one that can initiate downloads or persist potentially risky content without explicit safety boundaries.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
87% confidence
Finding

Directly adding torrent links to a torrent client or writing them to disk introduces side effects that can trigger downloads, persistence, and potential distribution of infringing or harmful content. Without warnings, confirmation steps, or destination constraints, the skill encourages unsafe automation beyond simple informational output.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
87% confidence
Finding

The add_magnet function submits a magnet link to qBittorrent, which can start downloading content and contacting trackers/peers, but there is no confirmation prompt or explicit user-facing warning about this side effect. For a code file, this is a safety-relevant operation affecting network activity and local downloads, and the surrounding messages only report success after the action has already been taken.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
85% confidence
Finding

The add_torrent_file function reads a local torrent file and posts it to qBittorrent, which can cause the client to begin downloading and interacting with the torrent network. Although the code logs success or failure afterward, it does not provide any prior disclosure or confirmation about the operational impact on the user's system and network activity.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The file’s user-facing description and runtime messages are written only in Traditional Chinese, with no indication that users may choose another language. The policy explicitly calls out language or locale constraints as violations when they are forced without opt-in or justification.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
93% confidence
Finding

The document is written as a Chinese-only skill specification and does not indicate that the user can choose another language or locale. Under the policy rule, forcing a specific language without opt-in is a natural-language policy violation unless the constraint is explicitly justified.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Low
Category
Not specified by scanner
Confidence
87% confidence
Finding

The document explicitly says all /torrent requests should be handed off to a central manager and that other instances should not search on their own. However, the file also contains detailed operational steps for opening BT4G, bypassing Cloudflare, and extracting magnet hashes, which conflicts with that delegation-only instruction.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Low
Category
Not specified by scanner
Confidence
91% confidence
Finding

The module docstring and usage text describe the skill as a bridge for adding magnet links to qBittorrent. However, the code also treats the positional argument as a local file path and uploads torrent files via add_torrent_file, which is a broader capability than the documentation states.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
79% confidence
Finding

The module docstring presents the skill description and usage entirely in Chinese, and the rest of the user-facing strings also assume that locale. Under the policy, forcing a specific language without user opt-in can be a natural-language policy issue when no alternative or justification is provided.

Content

No source excerpt is available for this finding.