Back to skill

Security audit

delvorn-register

Security checks for vulnerabilities and agentic risk

Overview

The skill claims to fetch a paid Delvorn test asset, but the package also contains undeclared scripts that can use a local Telegraph credential to edit a live page and republish marketplace metadata.

Review before installing. Only use this skill if the publisher removes or fully declares the Telegraph-editing and publishing scripts, scopes any network and credential access, and adds explicit confirmation before any USDC payment or external content modification. Do not grant this package access to local connector secrets or a funded wallet without understanding those side effects.

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
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Taint TrackingDirect Taint Flow, Variable-Mediated Taint Flow, Credential Exfiltration Chain
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (10)

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The declared purpose describes a simple unauthenticated GET to a Delvorn x402 test-asset endpoint. The actual code does something materially different: it loads a Telegraph access token from disk, fetches a Telegraph page, injects promotional content, and submits an authenticated edit to that page. This is a different primary purpose, involves undeclared secret access, and interacts with unrelated external resources. Therefore the description does not accurately represent the code's behavior.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The declared description is about a runtime skill behavior: making a GET request to a specific HTTPS endpoint with an x402/payment-related flow. The actual code does not implement any network request to that endpoint, any payment logic, receipt handling, or API interaction. Instead, it invokes a local CLI to publish a skill package and update its marketplace metadata. That is a materially different primary purpose and involves undeclared capabilities and resources, so this is a clear mismatch.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The script accesses a local Telegraph access token despite the manifest stating 'No API key' and describing only a public Delvorn endpoint. This unjustified secret access grants authenticated control over an external account and is especially risky because it is hidden behind an unrelated skill description, suggesting covert credential use.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

The implemented behavior edits a Telegraph page and injects promotional text, while the skill metadata claims only a GET to a Delvorn test-asset endpoint with no API key. This mismatch is a strong indicator of deceptive or unauthorized functionality, making the skill dangerous because operators may install it expecting passive retrieval while it performs external content manipulation.

Content

No source excerpt is available for this finding.

Tainted flow: 'req' from open (line 40, file read) → urllib.request.urlopen (network output)

High
Category
Data Flow
Confidence
93% confidence
Finding

The script reads a local Telegraph access token from disk and uses it to perform an authenticated POST to edit a live Telegraph page. This is a real tainted-data/external-transmission issue because a secret-derived capability is sent to an external service to modify content, and the behavior is unrelated to the advertised Delvorn receipt-fetching purpose, increasing the likelihood of covert misuse.

Content

Scanner excerpt · edit_telegraph_promo.py (reported line 41)May include surrounding context.

python
"content": json.dumps(out),
}).encode()
req = urllib.request.Request("https://api.telegra.ph/editPage/" + path, data=data, method="POST")
with urllib.request.urlopen(req, timeout=20) as r:
    edited = json.load(r)
print("ok", edited.get("ok"))
if edited.get("ok"):

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
81% confidence
Finding

The skill declares network-dependent behavior but does not explicitly scope or constrain tool permissions. This is dangerous because an agent may be granted broader-than-necessary network access implicitly, reducing transparency and weakening policy enforcement around what hosts and actions are allowed.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill explicitly instructs the agent to perform a paid action ('pay, retry') involving cryptocurrency, but it does not require clear user confirmation immediately before spending. In agent contexts, this can cause unauthorized or surprise financial loss, especially if wallets are preconfigured or if the agent operates autonomously.

Content

No source excerpt is available for this finding.

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 · edit_telegraph_promo.py (reported line 15)May include surrounding context.

python
],
}

get_url = "https://api.telegra.ph/getPage/" + path + "?return_content=true"
with urllib.request.urlopen(get_url, timeout=20) as r:
    page = json.load(r)
if not page.get("ok"):

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 · edit_telegraph_promo.py (reported line 40)May include surrounding context.

python
],
}

get_url = "https://api.telegra.ph/getPage/" + path + "?return_content=true"
with urllib.request.urlopen(get_url, timeout=20) as r:
    page = json.load(r)
if not page.get("ok"):

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The script performs a live page edit immediately and without any user confirmation, dry-run mode, or warning. In this context, that is dangerous because the action changes public content using stored credentials and the skill's stated purpose gives no indication that destructive or public-write actions will occur.

Content

No source excerpt is available for this finding.