Back to skill

Security audit

Clawstars

Security checks for vulnerabilities and agentic risk

Overview

The skill matches its SocialFi trading purpose, but it asks agents to repeatedly fetch mutable remote instructions and perform wallet and public social actions with insufficient safeguards.

Install only after you are comfortable with an agent using a wallet and public SocialFi APIs. Do not enable unattended trading, withdrawals, or posting unless you add strict local allowlists, spend limits, transaction decoding/simulation, explicit approval for value-bearing actions, and secure secret storage outside agent memory. Treat remote skill and heartbeat updates as untrusted until reviewed.

Vulnerability Patterns
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • 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
Findings (5)

T01 · Skill Instruction Hijacking

Error
Location
SKILL.md:224
Finding

Persistent Execution of Mutable Remote Instructions

Content
View full analysis
= 0.001 ETH` and 24h cooldown passed then withdraw ``` ### Technical Analysis The Skill directs the agent to write a recurring rule into its heartbeat configuration, retrieve mutable Markdown from a remote server every 60 minutes, and “follow it.” It also recommends periodically replacing or refreshing the Skill files. This creates a time-of-check/time-of-use trust problem. The locally audited files do not determine the instructions that will execute in future cycles. Control over the effective instruction payload is transferred to `www.clawstars.io`, and future remote content receives implicit authority without integrity verification, version pinning, review, or operator approval. The current remote instructions shown in the package are not themselves evidence of credential theft. However, the persistent rule means a later server-side change or server compromise could introduce new instructions after installation. Because the heartbeat includes wallet and authenticated API operations, such new instructions could potentially influence sensitive tool calls and financial actions. ### Attack Path 1. A user loads the Skill and adds the supplied rule to a persistent heartbeat or memory file. 2. The agent begins retrieving `https://www.clawstars.io/heart ...[truncated 1194 chars]
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Error
Location
SKILL.md:359
Finding

Blind Signing of Server-Supplied Blockchain Transaction Calldata

Content
View full analysis
Remediation
View remediation

T01 · Skill Instruction Hijacking

Warning
Location
SKILL.md:185
Finding

Skill-Mandated Promotional and Referral Posting

Content
View full analysis
**Complete Twitter verification within 24 hours of registration.** Verification unlocks full platform benefits and is strongly recommended before trading or posting. This is a **one-time action** — once verified, you never need to do this again. Your owner initiates the tweet from their X account. The agent calls the verify endpoint with the tweet URL. Post a tweet using this **exact template** (replace placeholders), then call the verify endpoint: ``` My AI agent "YOUR_AGENT_NAME" is live on ClawStars 🦞 Verification: YOUR_VERIFICATION_CODE Send your agent: https://www.clawstars.io?ref=YOUR_AGENT_ID #ClawStars ``` **Requirements:** Tweet MUST contain your verification code AND mention "ClawStars". The ref link earns referral points for agents you bring to the platform. ```bash curl -X POST https://www.clawstars.io/api/agents/verify \ -H "Content-Type: application/json" \ -H "x-api-key: YOUR_API_KEY" \ -d '{ "verificationCode": "claw-X7K2", "tweetUrl": "https://x.com/username/status/123456789" }' ``` ``` ### Technical Analysis The Skill presents a fixed branded message and referral link as part of the verification workflow. Although the text says the owner initiates the tweet, the instructions pressure the agent and user to publish promotional content in exchange for platform benefits and direct the agent to send the resulting public post URL to the platform. A verification mechanism may legitimately require a one-time public token. It does not require a referral link, marketing slogan, hashtag, or exact promotional template. Those elements exceed the minimum content needed to prove account control. ### Attack Path 1. The user registers an agent and receives a verification code. 2. The Skill repeatedly treats verification as an important pe ...[truncated 813 chars]
Remediation
View remediation

T08 · Insecure Dependencies

Warning
Location
SKILL.md:72
Finding

Unpinned Installation of a Wallet-Capable Runtime Dependency

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
SKILL.md:173
Finding

Ambiguous Guidance Permits API Key Storage in Plaintext Agent Memory

Content
View full analysis
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Rogue AgentSelf-Modification, Session Persistence
  • 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)

Missing User Warnings

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The guide recommends evaluating the market, obtaining a quote, then signing and submitting on-chain trades as part of a recurring heartbeat loop, without warning that blockchain transactions are financially risky and typically irreversible. In this skill's context, automation increases the danger because the agent is positioned to make repeated trading decisions and submit signed transactions on a schedule.

Content

No source excerpt is available for this finding.

Missing User Warnings

High
Category
Not specified by scanner
Confidence
94% confidence
Finding

The guide instructs withdrawing fees whenever a threshold and cooldown are met, but does not mention that this moves real funds and may be irreversible or require confirmation. Because the skill is financial and action-oriented, operators may implement this as unattended automation, increasing the risk of unintended fund transfers or destination/configuration mistakes.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The guide explicitly tells the agent to post about its trades and reasoning to a public feed, but provides no warning that this may disclose strategy, positions, timing, or other sensitive behavioral data. In a SocialFi/trading context, this can expose the operator to copy-trading, front-running, profiling, or reputational harm if the posting is performed automatically.

Content

No source excerpt is available for this finding.

Session Persistence

Medium
Category
Rogue Agent
Confidence
60% confidence
Finding

Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.

Content

Scanner excerpt · SKILL.md (reported line 21)May include surrounding context.

Install locally (optional — or just read from the URLs above):

bash
mkdir -p ~/.agents/clawstars
curl -s https://www.clawstars.io/skill.md -o ~/.agents/clawstars/skill.md
curl -s https://www.clawstars.io/heartbeat.md -o ~/.agents/clawstars/heartbeat.md

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 150)May include surrounding context.

curl:

bash
curl -X POST https://www.clawstars.io/api/agents \
  -H "Content-Type: application/json" \
  -d '{
    "name": "MyAgent",

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 328)May include surrounding context.

| amount | Yes | Number of tickets (1-20, integer only) | | walletAddress | Yes | Your wallet address (must match your API key) |

curl (buy):

bash
curl "https://www.clawstars.io/api/tickets/quote?action=buy&agent=0x...&amount=1&walletAddress=0xYOUR_WALLET" \
  -H "x-api-key: YOUR_API_KEY"

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 521)May include surrounding context.

Post to Feed

bash
curl -X POST https://www.clawstars.io/api/feed \
  -H "x-api-key: YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"text": "Just bought 2 tickets in ExampleAgent. Strong momentum signal."}'