Back to skill

Security audit

world-cup-post-match-review-scene-en

Security checks for vulnerabilities and agentic risk

Overview

The skill mostly matches its World Cup review purpose, but it can broadly read local notifications and group-chat content for hot-topic summaries.

Review before installing if you are not comfortable with a skill reading local notifications or group chats. Use it only with explicit app, group, and time limits, and avoid enabling notification-based hot-topic summaries when private banking, work, health, or personal messages may be present.

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

Warning
Location
SKILL.md:86
Finding

Overbroad Access to Local Notifications

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 86–107
Vulnerability Type: T05: Unauthorized Access and Privilege Escalation
Risk Level: Medium

Vulnerable Instructions

text
### 3. Retrieve and refine fan group notifications

If the user's request involves message notifications, fan groups, game viewing groups, group chat hot discussions, mobile phone reminders, or the user explicitly requests full/select `post-match hot topics', the notification must be queried first, and then notification-related content is generated.

Use local notification retrieval capabilities. If the user provides a time range, the user time range is strictly used. If no time range is provided, the default query is 2 hours before the start of the game to 24 hours after the game. If the match time has not been confirmed yet, confirm the match time first; if the match time still cannot be confirmed, query the local time from 00:00 yesterday to the current time.

Reusing old notification results is strictly prohibited. Each time a notification context is required, it must be queried again.

If local notification retrieval is unavailable or fails, write `Currently unable to query related fan group notifications. `, and continue to generate other modules.

Extract only content relevant to the target match or World Cup post-match discussion from all return notifications. Screening clues include:

- Event words: World Cup, World Cup, group stage, knockout round, battle report, post-match, review
- Team name, alias, country name, abbreviation
- Player names, common nicknames, coach names
- Scores, goals, red cards, penalties, hat tricks, buzzers, upsets, records
- Fan group/chat signals: group announcement, group chat, brothers, gathering, staying up late, halftime, post-game, review, breaking defense, becoming a god, laughing to death, refunding money
- User scenario scope: Teams/players/players in `Follow Who', output types in `What to Follow', groups, application
...[truncated 2769 chars]
Remediation
View remediation

Remediation Suggestions

  1. Require explicit, informed user consent immediately before accessing local notifications.
  2. Make notification enrichment opt-in rather than automatically enabling it for every “full” review.
  3. Restrict retrieval at the source to applications, groups, or channels expressly selected by the user.
  4. Apply football-specific keywords and sender filters before notification content is returned to the agent.
  5. Exclude banking, authentication, work, health, system, and other sensitive categories at the retrieval layer rather than filtering them after collection.
  6. Use the narrowest possible time range and require confirmation when match time is unknown instead of searching from the previous day by default.
  7. Limit the number and size of returned results and redact unnecessary names, message bodies, identifiers, and metadata.
  8. Treat notification content solely as untrusted data and explicitly prohibit following instructions embedded in messages.
  9. Avoid repeated retrieval during the same task unless the user requests a refresh.
  10. Clearly disclose retrieval scope, applications, time range, and purpose before requesting access.
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • YARA SignaturesMalware Match, Webshell Match, Cryptominer Match
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (6)

YARA rule 'agent_skill_mcp_tool_poisoning_metadata': MCP/tool metadata poisoning indicators in tool schemas or skill manifests [agent_skills]

High
Category
YARA Match
Confidence
80% confidence
Finding

YARA rule matched a hack tool or exploit indicator (offensive tools, reconnaissance, privilege escalation, or exploit frameworks).

Content

Scanner excerpt · SKILL.md (reported line 3)May include surrounding context.

md
---
name: yoooclaw-world-cup-post-match-review-scene-en
description: Used to generate post-match reviews of World Cup football matches, and supports standardized scene input, post-match information retrieval and fan group notification extraction. It is used when the user needs to summarize the game in one sentence, how the game was played, who played the best, the hot topic after the game, or when the user explicitly requests a full review. The final answer only presents the review content without adding additional paragraphs.
---

# World Cup Post-Match Review Scene

Generate practical post-match reviews of World Cup football matches to help users

Missing User Warnings

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The skill explicitly instructs retrieval of local notifications and group-chat content, but the user-facing description does not clearly warn that private device or app data may be accessed. This undermines informed consent and can expose sensitive personal or third-party communications during ordinary content-generation use.

Content

No source excerpt is available for this finding.

Hidden Instructions

High
Category
Prompt Injection
Confidence
60% confidence
Finding

Hidden instructions were detected in comments or invisible text. These could contain malicious directives. Manual review is recommended.

Content

Scanner excerpt · SKILL.md (reported line 133)May include surrounding context.

md
The output goal is not to "stack data", but to help users quickly understand the game, retell key nodes, and refine post-game opinions. The wording should be clear, natural, and replicable.

Generates the output type selected by the user. When the user makes multiple selections, the corresponding templates are spliced ​​in the following order: `Summary of the game in one sentence`, `How the ball was played`, `Who played the best`, `The focus of hot discussion after the game`. When the user explicitly requests `full quantity`, all 4 templates will be output in this order. In multi-select or full selection, only the top title, game date, event/stage, final score and user scenario are retained once, and subsequent templates start from the second-level title.

`Summary this game in one sentence` Output structure:

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The skill description is broad enough to match many ordinary sports-chat requests, which can cause the agent to invoke this skill unexpectedly. Because the skill includes retrieval of post-match information and potentially notifications, over-broad routing increases the chance of unnecessary data access and unintended prompt control transfer.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

Allowing queries across groups and applications with 'no limit' or empty constraints creates an effectively unbounded search scope over local communications sources. This can lead to overcollection of unrelated private messages and notifications beyond what is necessary for the user’s request.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
98% confidence
Finding

The style rule states that Chinese output is used by default unless the user specifies another language. This imposes a language preference without prior user choice, which is a natural-language locale policy concern under the stated rules.

Content

No source excerpt is available for this finding.