Back to skill

Security audit

Soccer Predict

Security checks for vulnerabilities and agentic risk

Overview

This skill is a disclosed football betting predictor, but it gives gambling recommendations and writes cross-session learning state without enough user control or risk disclosure.

Review this carefully before installing if you might act on betting advice. Treat outputs as uncertain analysis, not financial advice, and avoid using it for real-money wagering without your own controls. If installed, confirm before allowing it to write memory files, periodically inspect/delete its football memory files, and use a pinned or verified installer command rather than the unpinned README example.

Vulnerability Patterns
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • 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 (2)

T08 · Insecure Dependencies

Warning
Location
README.md:26
Finding

Unpinned Package Execution Through npx Installation Command

Content
View full analysis
Remediation
View remediation
install soccer-predict ``` 2. Publish and verify package integrity hashes or signed provenance before execution. 3. Document the expected package registry and discourage installation through untrusted registry mirrors. 4. Recommend installing dependencies from a lockfile with integrity metadata where supported. 5. Periodically audit the pinned installer and update the documented version only after review. 6. Consider providing a non-executing download and verification workflow for security-sensitive environments. ]]>

T02 · Agent Memory Poisoning

Warning
Location
references/review-framework.md:54
Finding

Externally Influenced Prediction State Is Persisted Across Sessions Without Validation or Isolation

Content
View full analysis
Remediation
View remediation
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (15)

Tp2

High
Category
MCP Tool Poisoning
Confidence
85% confidence
Finding

Mixing characters from multiple Unicode scripts in a single identifier is a common technique to create visually ambiguous tool names.

Content

No source excerpt is available for this finding.

Missing User Warnings

High
Category
Not specified by scanner
Confidence
95% confidence
Finding

The description presents the system as a betting recommendation engine without clearly warning users that outputs may influence gambling decisions. In this context, the omission is significant because the skill generates probabilistic recommendations and 'best bets,' which can materially affect user behavior in a high-risk financial domain.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
94% confidence
Finding

The README instructs users to run npx clawhub install soccer-predict without pinning an exact package version. That allows whatever version is current at execution time to be fetched and run, which creates a supply-chain risk if the package is updated maliciously, compromised, or changed incompatibly.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The README presents the skill as generally usable in English, but all invocation examples are Chinese commands such as '预测比赛' and '复盘 比分'. Because no explicit user choice or justified locale limitation is provided, this appears to impose a language-specific interaction pattern, which matches the language/locale policy concern.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
94% confidence
Finding

The Chinese install section repeats the unpinned npx clawhub install soccer-predict command, preserving the same supply-chain execution risk for readers of that section. Because npx resolves and executes remote package code dynamically, lack of version pinning increases exposure to upstream compromise or unexpected updates.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The trigger conditions are broad enough to activate on general football-related prompts, which can cause unintended invocation of a betting-oriented skill. In context, this is more dangerous because the skill performs scraping, produces gambling recommendations, and can persist learned state, so accidental routing could expose users to unsolicited betting advice and unnecessary data collection.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The skill states that match history and updated model weights are persisted across sessions, but this retention is not clearly disclosed up front to users. In context, persistence increases risk because user-provided results, analysis history, and adaptive model state may be stored unexpectedly, creating transparency, privacy, and cross-session data governance concerns.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The framework culminates in 'best betting recommendation', predicted scores, and confidence levels, but includes no warning about gambling risk, financial loss, uncertainty, or responsible-use constraints. In a betting-focused skill, presenting probabilistic advice without risk disclosure can encourage users to over-trust the output and make real-money decisions based on incomplete or unreliable analysis.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The framework explicitly instructs writing learned parameters, league profiles, and match history into persistent memory files under the user's local Claude project directory. This creates stateful behavior across sessions that is not clearly bounded or disclosed in the skill metadata, and can cause silent accumulation of user-derived data and hidden behavior changes over time.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The skill directs persistent writes to local memory files without warning the user that local data will be modified across sessions. Undisclosed file writes can surprise users, retain potentially sensitive or proprietary usage history, and create hard-to-audit behavioral drift as future outputs depend on hidden stored state.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
84% confidence
Finding

This is a markdown file, so SQP-2 applies to omitted warnings in the skill description. The README advertises automated scraping from titan007.com and generation of full HTML reports plus post-match learning/accuracy tracking, but it does not disclose possible privacy, data-retention, or system-impact implications of fetching external data and storing generated outputs or learning state.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
89% confidence
Finding

The document embeds Chinese labels and status strings such as '简洁模式', '可视化模式', '阵容未公布', and '待公布' as part of the prescribed output/instruction flow, but it does not state that the user can choose language or locale. This creates a language-policy concern because the skill appears to force mixed-language output without explicit user opt-in.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
78% confidence
Finding

The input format presents the user-provided result only as a Chinese phrase, which can indicate a language-specific interaction requirement. The file does not offer alternative language options or explain why Chinese is required for this skill.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
89% confidence
Finding

The review flow allows auto-fetching match results from titan007.com but does not disclose that it may make an external network request. Even though the data is sports-related and low sensitivity, undisclosed outbound access can leak usage patterns and violates user expectations around network behavior.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Low
Category
Not specified by scanner
Confidence
74% confidence
Finding

The manifest says the skill supports post-match review and auto weight optimization, but this document specifies an ongoing program of cumulative accuracy tracking and optimization toward a 70%+ target. That turns the skill into a continuously maintained learning system, which is a stronger operational claim than the manifest currently makes.

Content

No source excerpt is available for this finding.