Back to skill

Security audit

web123

Security checks for vulnerabilities and agentic risk

Overview

This skill is a Web3 skill recommender, but it broadly routes wallet and trading requests into unpinned installs for transaction-capable skills without enough risk disclosure.

Install only if you understand that this is a recommender for Web3 skills, not a safety guarantee. Before running any suggested install command, review the exact downstream repository, prefer pinned reviewed revisions, and be especially careful with skills involving trades, exchange API keys, wallet connections, signatures, approvals, payment links, or leverage.

Vulnerability Patterns
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • 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
  • 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)

T03 · Remote Payload Retrieval and Execution

Warning
Location
SKILL.md:237
Finding

Unpinned Remote Skill Installation Allows Post-Audit Payload Changes

Content
View full analysis

Vulnerability Details

File Location: SKILL.md:237-251
Vulnerability Type: Unpinned installation of mutable remote Skill repositories
Risk Level: Medium

Vulnerable Code

bash
# Install web123 itself / 安装 web123
openclaw skill install https://github.com/AntalphaAI/web123

# Single skill / 安装单个
openclaw skill install https://github.com/AntalphaAI/web3-trader

# 新手包 / Beginner pack
openclaw skill install https://github.com/AntalphaAI/wallet-balance https://github.com/AntalphaAI/web3-trader https://github.com/AntalphaAI/transaction-receipt

# 交易套装 / Trading pack (v1.1 new)
openclaw skill install https://github.com/AntalphaAI/web3-trader https://github.com/AntalphaAI/cex-trader https://github.com/AntalphaAI/TA-Radar

# 数据情报套装 / Data intel pack (v1.1 new)
openclaw skill install https://github.com/AntalphaAI/smart-money https://github.com/AntalphaAI/crypto-social-intel https://github.com/AntalphaAI/TA-Radar

Equivalent unpinned commands also appear in README.md:17, README.md:60-78, README.md:127, README.md:170-188, and throughout references/skills.json:47-344.

Technical Analysis

The Skill generates installation commands that retrieve other Skills directly from mutable GitHub repository URLs. The URLs do not specify immutable commit hashes, signed release tags, version constraints, or integrity digests.

Consequently, the content installed by a command can differ from the content that existed when this navigator was reviewed. Although all listed URLs use the same stated GitHub organization and no malicious upstream content was present in the audited project, the external repositories are not included in the audit scope and their future contents cannot be guaranteed.

This is best classified as remote payload retrieval because execution of the suggested installation command fetches externally controlled Skill content whose effective payload may change after review.

Attack Path

...[truncated 1544 chars]

Remediation
View remediation

Remediation Suggestions

  1. Pin every recommended repository to a reviewed immutable commit hash or cryptographically signed release.
  2. Extend references/skills.json with explicit version, commit, and integrity-digest fields instead of storing only mutable repository URLs.
  3. Verify the resolved revision and integrity digest before installation, and reject content that does not match the approved metadata.
  4. Display the repository, revision, requested permissions, and verification status before installation.
  5. Require explicit user confirmation before installing any remote Skill, particularly trading, wallet, payment, API-key, or signing-related Skills.
  6. Maintain an allowlist of reviewed repository revisions and periodically re-audit dependencies before approving newer versions.
  7. Prefer a trusted registry or release-artifact mechanism that supports signatures and reproducible integrity verification.
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 (7)

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The README uses very broad natural-language trigger examples such as trading, wallet balance, and beginner help that closely resemble ordinary user requests. In an agent ecosystem, this can cause unintended skill activation or over-recommendation in contexts where the user did not explicitly mean to invoke this navigator, increasing the chance of routing users into sensitive financial or wallet-related workflows.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The README recommends skills for trading, wallet inspection, contract scanning, and exchange use without any caution about financial loss, phishing, privacy leakage, or the limits of automated security analysis. In Web3, users may interpret these recommendations as safe-by-default guidance and proceed with fund-affecting actions or disclose wallet data without understanding the risks.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The Chinese section repeats the same issue by presenting common conversational phrases like buying BTC, checking wallet balances, and scanning contracts as triggers. Because these are normal user utterances, they can blur invocation boundaries and cause accidental activation for high-sensitivity Web3 actions in multilingual settings.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The Chinese README likewise promotes wallet, trading, and contract-scanning workflows without warning about loss of funds, privacy exposure, false negatives in scam detection, or exchange/account risks. The multilingual presentation broadens the affected audience and can normalize risky actions without informed consent.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The trigger conditions are broad enough to activate on generic Web3-related requests such as trading, wallet balance checks, beginner questions, or category browsing. This can cause the skill to intercept user queries that were not intended for this navigator, increasing the chance of inappropriate routing, unsolicited install-command suggestions, or overshadowing more specific and safer skills.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

This catalog recommends and packages skills that can trade assets, connect wallets, request signatures, and generate payment links, but it does not present any explicit warnings about financial loss, irreversible transactions, approvals, leverage risk, or signing malicious payloads. In a navigator that helps users choose and install Web3 skills, omission of these warnings can directly increase the chance that users install and use powerful transaction-capable tools without understanding the risks.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
97% confidence
Finding

All category labels, descriptions, scenarios, and examples are presented in Chinese, and the file does not indicate that this locale is optional or region-specific. This can violate language-choice policy when the skill set is not explicitly limited to Chinese-speaking users.

Content

No source excerpt is available for this finding.