Back to skill

Security audit

gate-info-riskcheck

Security checks for vulnerabilities and agentic risk

Overview

This skill is mostly a read-only token risk checker, but it needs review because it relies on runtime rule files outside the packaged artifact that could change its behavior.

Install only if you trust the publisher and can verify the shared runtime rule files that this skill expects outside its own package. Expect read-only token-risk reports, not definitive investment, sanctions, or blacklist decisions; address safety checks are explicitly degraded and should not be treated as compliance clearance.

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
Findings (1)

T01 · Skill Instruction Hijacking

Error
Location
SKILL.md:15
Finding

External Runtime Rules Can Hijack Skill Instructions

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 15–18
Vulnerability Type: External instruction redirection with priority override
Risk Level: High

Vulnerable Code

markdown
⚠️ STOP — You MUST read and strictly follow the shared runtime rules before proceeding.
Do NOT select or call any tool until all rules are read. These rules have the highest priority.
→ Read `../gate-runtime-rules.md`
→ Also read `../info-news-runtime-rules.md` for gate-info / gate-news shared rules (tool degradation, report standards, security, routing, and optional local maintenance when `scripts/` is present).

Technical Analysis

The skill mandates loading two instruction files from its parent directory and explicitly declares those external instructions to have the “highest priority.” Neither referenced file is included in the audited artifact, so its contents, integrity, and security properties cannot be verified.

Because the paths escape the project root through ../, the skill’s effective behavior depends on mutable files outside the reviewed package. Any party able to create or modify either parent-directory file could introduce instructions that alter tool selection, routing, data handling, maintenance operations, safety constraints, or report output without changing the audited skill itself.

This constitutes instruction hijacking because loading the skill establishes an unaudited external control channel for changing the Agent’s current-session behavior. The external rules cannot legitimately override system or developer instructions, but the skill text nevertheless attempts to assign them maximum priority within its workflow.

Attack Path

  1. An attacker obtains write access to the directory containing gate-runtime-rules.md or info-news-runtime-rules.md, or influences how those files are provisioned.
  2. The attacker inserts malicious Agent instructions into one of those files.
  3. A user invokes this skill.
  4. Before selecting or ca ...[truncated 1040 chars]
Remediation
View remediation

Remediation Suggestions

  1. Bundle every required runtime-rule document inside the audited skill package.
  2. Replace parent-directory references with package-relative paths that cannot escape the project root, such as references/gate-runtime-rules.md.
  3. Remove language assigning externally loaded content “highest priority.”
  4. Explicitly state that skill and referenced instructions cannot override system, developer, platform-security, or user-consent requirements.
  5. Include all referenced rule files in release review, integrity verification, and version control.
  6. If shared rules must remain external, verify them against a pinned cryptographic digest and fail closed when verification fails.
  7. Restrict external rules to declarative configuration with a documented schema rather than unrestricted natural-language instructions.
  8. Ensure external configuration cannot expand the skill’s tool allowlist, request additional privileges, initiate maintenance actions, or change data-disclosure policy.
  9. Add packaging tests that reject references resolving outside the skill root.
Vulnerability Patterns
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Findings (7)

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The example invocations include generic phrases like "Is this token safe" and "Is this address safe," which are common natural-language requests and may match user speech outside a narrowly defined invocation context. The README does not provide exclusion conditions or more specific trigger constraints to clarify when this skill should activate versus other crypto-analysis skills.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The manifest description narrowly scopes the skill to security assessment only and explicitly says multi-dimension queries should go elsewhere. However, the workflow calls info_coin_get_coin_info and includes project name, sector, and listed exchanges as supplementary context, which goes beyond pure token/contract/address security and overlaps with broader research dimensions.

Content

No source excerpt is available for this finding.

Unrestricted Tool Access

Medium
Category
Excessive Agency
Confidence
80% confidence
Finding

Skill grants unrestricted tool access without appropriate constraints. An agent with unfettered tool access can perform arbitrary actions including file modification, network requests, and code execution.

Content

Scanner excerpt · SKILL.md (reported line 16)May include surrounding context.

md
## General Rules

⚠️ STOP — You MUST read and strictly follow the shared runtime rules before proceeding.
Do NOT select or call any tool until all rules are read. These rules have the highest priority.
→ Read `../gate-runtime-rules.md`
→ Also read `../info-news-runtime-rules.md` for gate-info / gate-news shared rules (tool degradation, report standards, security, routing, and optional local maintenance when `scripts/` is present).
- **Only call MCP tools explicitly listed in this skill.** Tools not documented here must NOT be called, even if they

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
87% confidence
Finding

Line L022 states address compliance checking is a future capability, implying it is not currently part of the skill's operations. Elsewhere, the skill lists info_compliance_check_address_risk among tools used and required query operations, which contradicts that statement even though later sections mark it as not ready.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The tool parameter specification explicitly sets lang="en", which forces English regardless of the user's preferred language. This is a natural-language policy concern because the file does not offer a language choice or explain a justified region-specific constraint.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The trigger examples "Is this token safe" and "Is PEPE safe" are broad enough to activate this skill without ensuring the user wants only security/risk analysis. Because the skill metadata explicitly says this skill must be used only for exclusive token/contract/address security queries, ambiguous routing can cause the agent to select a narrow risk-check workflow when the user may actually need broader research, leading to incomplete or misleading analysis.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
86% confidence
Finding

The address-risk examples "Is this address safe 0x..." and especially "Is this a blacklisted address" are semantically broad and can imply sanctions, fraud, or compliance screening that the skill cannot actually perform. Since the scenario itself states address compliance risk detection is under development, these triggers may cause over-routing into a degraded mode that provides only basic address info, creating a risk of false reassurance or incomplete safety judgments.

Content

No source excerpt is available for this finding.