Back to skill

Security audit

Unbrowse for Siriusxm (unofficial)

Security checks for vulnerabilities and agentic risk

Overview

The skill is domain-scoped to SiriusXM, but it exposes remote browser, login autofill, and dynamic tool-creation capabilities beyond the advertised read-only tools.

Review this skill before installing if you only want simple SiriusXM search or page reading. Use it only if you are comfortable connecting to Unbrowse, sending SiriusXM tasks through that service, and allowing a cloud browser/password-manager workflow; require explicit confirmation before any sign-in, tool compilation, purchase, account change, deletion, or other mutating action.

Vulnerability Patterns
  • 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
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • 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
Findings (5)

Rp1

Medium
Category
MCP Rug Pull
Confidence
88% confidence
Finding

This is a duplicate instance of the same issue: the documentation references npx skills add without any version constraint. In a supply-chain attack scenario, users following the README could fetch and execute a newer malicious package or dependency than the author originally intended.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The skill markets the MCP as holding only siriusxm.com-scoped tools, but later documents generic Unbrowse capabilities such as task execution, scraping, mapping, and recorded browser actions that can expand capabilities dynamically. This mismatch weakens operator trust and can lead an agent to invoke broader, potentially state-changing functionality under a misleadingly narrow safety model.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The tool descriptions repeatedly frame the integration as read-only, yet the later operational guidance acknowledges actions like posting, buying, sending, or deleting. That contradiction can cause an agent or operator to treat the skill as non-mutating when it may actually perform side-effecting actions, increasing the risk of unauthorized account or transaction changes.

Content

No source excerpt is available for this finding.

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
88% confidence
Finding

The workflow encourages the agent to proceed autonomously with tool selection, resume interrupted runs, and use password-manager autofill for sign-in, while only requiring user go-ahead for a subset of obviously mutating actions. In context, this increases the chance of the agent authenticating to accounts or continuing workflows without sufficiently granular user consent, especially given the presence of browser-based and dynamically compiled capabilities.

Content

Scanner excerpt · SKILL.md (reported line 67)May include surrounding context.

md
1. Pick the tool whose title matches the job and pass its inputs. Leave the site's recorded parameters out unless the user asked for something they control.
2. Read `status`. `succeeded` carries the verified `result`. `input_required` is not a failure: answer the open fields with `unbrowse.resume {runId, answers}` on the same run.
3. No tool fits: `unbrowse.run {task}`. It fails with `no_capability` when nothing is compiled for that yet: do it once with `unbrowse.browse.open` on siriusxm.com → `unbrowse.browse.act` → `unbrowse.browse.finish`, which compiles a new tool this server lists next time. Anything off siriusxm.com is refused here (`not_in_scope`): use the general Unbrowse server for that.
4. Signing in: never ask the user for a password. `unbrowse.browse.act` with `action: "autofill"` fills siriusxm.com's login from their Unbrowse password manager.
5. Report only what a `succeeded` result says. `outcome_unknown` means a change may have happened: check before retrying. Posting, buying, sending or deleting needs the user's go-ahead.

## Without MCP

External Transmission

Medium
Category
Data Exfiltration
Confidence
60% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · SKILL.md (reported line 75)May include surrounding context.

Every tool is also a REST call:

sh
curl -s https://unbrowse.ai/api/v1/sites/siriusxm.com/call/siriusxm_com__get_search \
  -H "authorization: Bearer $UNBROWSE_API_KEY" -H "content-type: application/json" \
  -d '{"query":"<query>"}'