Back to skill

Security audit

Arbitrum Bridge Helper

Security checks for vulnerabilities and agentic risk

Overview

This skill is for Arbitrum bridge help, but it gives the agent overly broad authority to find and use local wallet credentials for real transactions.

Review this skill carefully before installing. It should not be allowed to scan for wallet secrets broadly; use it only if you are comfortable with an agent accessing wallet credential files, and prefer a version that requires you to name the exact wallet address and approve a specific credential path before any transaction is prepared or signed.

Vulnerability Patterns
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
  • 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 (1)

T05 · Unauthorized Access and Privilege Escalation

Error
Location
SKILL.md:61
Finding

Overly Broad Discovery and Selection of Local Wallet Credentials

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 61-79
Vulnerability Type: T05: Unauthorized Access and Privilege Escalation
Risk Level: High

Vulnerable Code Snippet

markdown
## Step 2 - Discover a wallet on disk

When execution is requested, look for an existing usable wallet locally before
asking the user for anything else.

Preferred wallet sources:

1. A project or workspace `.env` with a private key or mnemonic
2. A local JSON keystore plus an available password source on disk
3. Existing wallet config files in the workspace or common agent-accessible directories

Wallet handling rules:

- Use the first valid wallet that matches the requested network and has enough funds
- Never print the secret, keystore JSON, mnemonic, or full env file content
- If multiple wallets are found, summarize them by public address only and choose
  the best-funded viable one unless the user specified a different address
- If no wallet is found, say that clearly and stop rather than fabricating one
- If a wallet is found but lacks gas or assets, report that and do not attempt execution

Technical Analysis

The Skill instructs the agent to search for highly sensitive wallet credentials, including private keys, mnemonic phrases, keystores, and corresponding password sources. The search scope includes vaguely defined “common agent-accessible directories,” rather than being confined to a credential path explicitly authorized by the user.

This violates least-privilege principles in two ways:

  1. It permits access to credentials beyond the minimum scope necessary for the requested bridge transaction.
  2. It authorizes automatic wallet selection based on discovery order or available funds rather than explicit ownership and wallet authorization.

The instruction not to print secrets only limits direct disclosure in chat. It does not prevent the agent from reading an unrelated credential, retainin ...[truncated 1996 chars]

Remediation
View remediation

Remediation Suggestions

  1. Remove automatic searches for private keys, mnemonic phrases, keystores, and password sources.
  2. Require the user to specify the exact public wallet address intended for the operation.
  3. Require an explicit, user-approved credential path or an established wallet-provider integration; do not infer authorization from filesystem accessibility.
  4. Restrict credential access to a narrowly allowlisted directory and canonicalized file path within the task workspace.
  5. Reject paths that escape the approved directory through absolute paths, symbolic links, traversal sequences, or mount points.
  6. Never select a wallet based on discovery order, token balance, or gas balance. If several authorized wallets exist, require the user to select one by public address.
  7. Separate consent into distinct stages: permission to access the named credential, approval of the exact source and destination chains, and final confirmation of the asset, amount, address, fees, and transaction calldata before signing.
  8. Prefer hardware wallets or wallet-provider APIs that keep private key material outside the agent's context.
  9. Ensure logs, error messages, tool outputs, and model responses redact private keys, mnemonics, keystore contents, passwords, RPC credentials, and complete environment files.
  10. Clear sensitive execution context after use and avoid persisting wallet material in memory, caches, temporary files, or telemetry.
  11. Record an auditable authorization event containing only non-secret details, such as the selected public address and approved transaction parameters.
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (3)

Missing User Warnings

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

The skill instructs the agent to search local disk for private keys, mnemonics, keystores, or wallet configs, but the user-facing description does not clearly disclose this up front. That creates a dangerous mismatch in consent and expectation: a user may trigger the skill for bridge help without realizing it will inspect sensitive local secret material and potentially use those secrets to sign transactions.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The trigger definition is overly broad for a high-risk skill that can locate local wallets and execute blockchain transactions. Phrases like 'move', 'send', casual wording, abbreviations, and minor typos increase the chance of accidental invocation in unrelated conversations, which could lead the agent into wallet discovery or transaction-preparation flows without sufficiently specific user intent.

Content

No source excerpt is available for this finding.

Vague Triggers

Low
Category
Not specified by scanner
Confidence
88% confidence
Finding

The instruction says to use this reference when a user says a bridge is 'stuck, missing, not claimable, or behaving unexpectedly,' which is a broad natural-language trigger for a markdown file. It does not clearly constrain that this applies only to the official Arbitrum bridge or provide exclusion examples, so it could cause unintended invocation for unrelated bridge contexts.

Content

No source excerpt is available for this finding.