Back to skill

Security audit

AI Promo

Security checks for vulnerabilities and agentic risk

Overview

This promo skill is not clearly malicious, but it uses persistent user tracking, remote submissions, subscription state, scheduled-push instructions, and fixed referral-style links that deserve review before installation.

Install only if you are comfortable with a Chinese-language promo service that contacts cli.aipromo.workers.dev, stores a stable local user ID and subscription data in your home directory, may be configured for daily push behavior, and includes referral-style links. Review or remove the cron configuration and local state files if you do not want tracking or recurring notifications.

Vulnerability Patterns
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • 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
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
Findings (2)

T01 · Skill Instruction Hijacking

Error
Location
SKILL.md:71
Finding

Persistent Injection of Undisclosed Referral and Promotional Content

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
scripts/promo_submit.sh:29
Finding

JSON Injection Through Unsafe Promotion Payload Construction

Content
View full analysis
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (27)

Tp4

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

This finding points to undocumented local file creation and ID management rather than direct promo retrieval. Even if not overtly malicious, creating persistent identifiers without clear disclosure violates least surprise and can enable tracking or correlation across sessions.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

This finding points to undocumented local file creation and ID management rather than direct promo retrieval. Even if not overtly malicious, creating persistent identifiers without clear disclosure violates least surprise and can enable tracking or correlation across sessions.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

This finding points to undocumented local file creation and ID management rather than direct promo retrieval. Even if not overtly malicious, creating persistent identifiers without clear disclosure violates least surprise and can enable tracking or correlation across sessions.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

This finding points to undocumented local file creation and ID management rather than direct promo retrieval. Even if not overtly malicious, creating persistent identifiers without clear disclosure violates least surprise and can enable tracking or correlation across sessions.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

The skill metadata says it queries promotional information, but this script submits data to a remote API, creating a clear capability mismatch. That mismatch is dangerous because users or calling agents may invoke the skill under a read-only expectation while unknowingly causing outbound data submission of user-provided content and identifiers.

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
96% confidence
Finding

The skill documents shell-script execution paths but declares no explicit tool scope or permission boundary. That creates a gap between what the skill appears allowed to do and what it can actually invoke, increasing the chance of unintended command execution, file writes, and network access when triggered.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The trigger set includes broad phrases such as recommendation/earning and discount-related language that may appear in ordinary conversation. Overbroad automatic activation is risky here because the skill can invoke shell-backed workflows, perform network requests, and create persistent local state without a clearly intentional user action.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

Ambiguous automatic trigger keywords without guardrails can cause accidental execution in unrelated contexts. In this skill, accidental activation is more dangerous because it may lead to identifier generation, API calls, or subscription-related actions rather than a harmless text response.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The skill describes local storage of user and subscriber data plus outbound API use, but it lacks a clear user warning about those privacy-relevant behaviors. Without explicit notice and consent, users may unknowingly permit tracking identifiers, retained preferences, and transmission of submitted content to external infrastructure.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

This script goes beyond a simple promo-query capability and implements subscriber tracking, scheduled push behavior, and per-user delivery state. For a skill described as querying discount information on request, maintaining a local subscription database and running unattended periodic pushes expands the data handling and messaging surface in a way users may not expect, increasing privacy and abuse risk.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The script stores subscriber data persistently in a local file under the user's home directory, including channel identifiers and send history. For a promo information skill, this is an unjustified persistence capability unless clearly disclosed and necessary, and it creates a local privacy risk if the file is accessed by other processes or users.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The function sends user_id to a remote endpoint as a URL query parameter, which exposes subscriber identifiers to the external service and potentially to intermediary logs. The script contains no user-facing disclosure or consent mechanism for this transmission, making the data sharing unexpected in the context of a promo-query skill.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The script silently generates a persistent per-user identifier, stores it in the user's home directory, and transmits it to a remote API even though the advertised purpose is only querying promo information. This enables cross-session tracking and remote correlation of user activity without informed consent or an obvious functional need visible to the user.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

Writing ~/.promo_user_id creates durable local state for tracking in a skill whose expected behavior is a simple promo lookup. Even if the value is randomly generated, silent persistence in the home directory can surprise users, complicate privacy expectations, and create a stable handle reused across future invocations.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The script sends a persistent user identifier in the query string to a remote endpoint without explicit disclosure or consent. This allows the remote service, intermediaries, or logs to associate repeated promo queries with the same user over time, increasing privacy and tracking risk beyond what users would expect from a promo-listing skill.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
98% confidence
Finding

The script sends user-supplied platform, title, link, bonus, description, and a persistent user ID to an external endpoint without prior notice or consent. In an agent-skill context this is especially risky because users may believe they are performing a local lookup, not disclosing content and a durable identifier to a third party.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The script creates and persists a per-user identifier in $HOME/.promo_user_id without clear necessity for a skill described as querying promo information. Persistent identifiers enable cross-session tracking and create a privacy risk, especially when later transmitted to a remote service.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
91% confidence
Finding

The script performs an outbound POST request to an external service, transmitting a JSON payload containing user-provided content and a persistent identifier. External transmission is not inherently malicious, but here it is security-relevant because it is undisclosed and inconsistent with the skill's stated query-focused purpose.

Content

Scanner excerpt · scripts/promo_submit.sh (reported line 41)May include surrounding context.

sh
EOF
)
  
  local response=$(curl -s --max-time 30 -X POST \
    -H "Content-Type: application/json" \
    -d "$json_payload" \
    "${API_ENDPOINT}/submit")

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

Comments and usage text are entirely in Chinese, and later user-facing prompts also instruct the user with fixed Chinese phrases such as saying '优惠订阅 开启/关闭'. This imposes a locale/language constraint without opt-in or documented justification, matching the language-policy violation category.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The manifest describes a query-oriented skill for promotional information, but this script implements persistent subscription management for daily push notifications. That mismatch expands the skill's effective capabilities beyond what users and reviewers would reasonably expect, creating a transparency and consent problem and increasing the risk of unauthorized retention and use of user/channel identifiers.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The script stores subscriber identities, channel names, status, and timestamps in a local file under the user's home directory without any visible access controls, retention policy, or clear justification from the stated query-only skill purpose. Even though this is not direct code execution, unnecessary persistence of user-linked notification data creates avoidable privacy and abuse risk if the host environment is shared, backed up, or later repurposed.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The script performs a file write that replaces the existing persistent identifier in $HOME/.promo_user_id when the regenerate path is used. Although comments describe the behavior for developers, there is no user-facing prompt, warning, or visible disclosure before this overwrite occurs.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
82% confidence
Finding

The natural-language description, headings, and feature list are entirely in Chinese, and the document does not mention any option to use another language or locale. Under the language/locale policy rule, this can be a policy concern when a skill appears to enforce a specific language without user choice.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
93% confidence
Finding

Comments and user-visible message content such as the push text are written only in Chinese, with no indication that users can choose another language. This can violate language/locale policy where user-facing behavior should not force a specific language absent explicit opt-in or documented justification.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
91% confidence
Finding

The script writes a persistent identifier into the user's home directory without warning, which is a privacy-relevant side effect not implied by the skill's stated purpose. Silent local state creation can undermine user trust and facilitate later tracking when combined with remote transmission.

Content

No source excerpt is available for this finding.