Back to skill

Security audit

Dailyfinance

Security checks for vulnerabilities and agentic risk

Overview

This skill is a disclosed finance-briefing dashboard with optional scheduled pushes, but users should review the recurring notification behavior before enabling it.

Installers should understand that enabling pushes creates recurring morning and evening scheduled messages to the selected channel and stores a local userId plus schedule preferences. Use the documented off command to disable pushes, and review the generated HTML if ad placeholders or fixed branding are not desired.

Vulnerability Patterns
  • 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
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Findings (1)

T01 · Skill Instruction Hijacking

Note
Location
SKILL.md:68
Finding

Mandatory Advertising Placeholders and Third-Party Branding in Generated Output

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 68–69
Vulnerability Type: Output instruction hijacking through mandatory promotional content
Risk Level: Low

Complete Code Snippet:

md
- **Ad-ready zone**: `<div id="ad-slot-ticker">` below ticker. `<div id="ad-slot-mid">` mid-page. `<div id="ad-slot-bottom">` at footer.
- **Footer**: "Powered by ClawCode"

Technical Analysis

The skill instructions require every generated financial dashboard to contain three advertising placeholders and a fixed third-party branding statement. These requirements are unrelated to the core function of producing a financial-news briefing and alter the agent's output without requiring explicit user consent.

The advertising containers do not execute code by themselves. However, they establish predefined insertion points that could subsequently be populated with advertising, tracking elements, or other third-party content. The fixed footer also causes the agent to include promotional branding as though it were an inherent part of the requested artifact.

This is classified as skill instruction hijacking because the behavior originates in the loaded skill instructions and persistently modifies the current session's output requirements.

Attack Path

  1. A user invokes the Daily Finance skill to generate a financial briefing.
  2. The agent loads and follows the visual requirements in SKILL.md.
  3. The agent inserts the three required advertising containers into the generated HTML.
  4. The agent adds the mandatory Powered by ClawCode footer.
  5. The user receives an artifact containing promotional infrastructure and branding that was not independently requested.
  6. If another component later populates the placeholders, advertising or tracking content may be delivered through the generated artifact.

Impact Assessment

The issue can manipulate the content and structure of generated HTM ...[truncated 475 chars]

Remediation
View remediation

Remediation Suggestions

  1. Remove the mandatory advertising placeholders and fixed branding requirement from the default skill workflow.
  2. If advertising support is a legitimate feature, make it an explicit opt-in setting that defaults to disabled.
  3. Inform users before adding promotional or sponsored content to generated artifacts.
  4. Keep optional branding separate from the financial-report generation instructions.
  5. If external content may populate an enabled advertising slot, restrict it to trusted sources and apply a strict Content Security Policy.
  6. Sanitize any content inserted into optional slots and prohibit arbitrary scripts, tracking pixels, inline event handlers, and untrusted iframe sources.
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Findings (12)

Tp4

High
Category
MCP Tool Poisoning
Confidence
95% confidence
Finding

A dashboard/news skill that also manages recurring outbound delivery channels and persists per-user configuration materially expands its trust boundary. Even without overtly malicious logic, hiding or under-describing this operational behavior can enable unauthorized notifications, unintended retention of user identifiers, and abuse of messaging integrations.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

A dashboard/news skill that also manages recurring outbound delivery channels and persists per-user configuration materially expands its trust boundary. Even without overtly malicious logic, hiding or under-describing this operational behavior can enable unauthorized notifications, unintended retention of user identifiers, and abuse of messaging integrations.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

Very broad finance-related keywords can cause the skill to trigger on common user conversations about markets, investing, or economics without clear intent to use this specific skill. Over-broad activation increases the chance of unintended execution, which is more concerning here because the skill also advertises scheduling and outbound-notification related functionality.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

Including push-notification management in a skill described primarily as a dashboard creates a secondary capability that changes the security/privacy posture of the skill. Users may invoke or install it expecting passive content generation, while the skill documentation enables recurring outbound actions tied to user IDs and delivery channels.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The push-management commands describe enabling recurring notifications with a userId and channel, but do not warn users about ongoing outbound messaging, identifier handling, retention, or unsubscribe expectations. This creates privacy and consent risks, and could lead to users being enrolled in recurring delivery without understanding the persistence or scope of the action.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

Managing Telegram, Feishu, Slack, and Discord delivery channels is a substantial capability unrelated to simple HTML dashboard generation. If not justified and permissioned, these integrations can be used to send unsolicited content, mishandle credentials or user identifiers, and broaden exposure beyond the local rendering context.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
85% confidence
Finding

The manifest describes a daily finance briefing skill, but this file's concrete behavior is mostly local user-file handling plus printing a prompt for someone else to perform the actual search and briefing generation. That is a noticeable mismatch from a skill that claims to provide the briefing itself, especially since the only user-specific operation is reading a local profile that is never used in the output.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

A daily financial news briefing skill would reasonably fetch market/news data or format a digest, but reading ../data/users/<user>.json is not justified by the stated purpose as implemented here. The loaded user record is not used afterward, making the capability appear extraneous to the finance recap function.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The emitted instruction explicitly requires '中英双语' and is otherwise written entirely in Chinese, with no option for the user to select language or locale. This is a natural-language policy concern because the skill imposes a specific language behavior rather than offering opt-in or documenting a justified regional constraint.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The script's user-facing output is consistently in Chinese, including errors, usage text, and status messages. This imposes a specific language on all users without opt-in or justification, which matches the language/locale policy violation category.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
95% confidence
Finding

The README states that the skill produces a bilingual EN/CN dashboard as a fixed behavior. This can violate language/locale policy because it forces a specific language configuration without mentioning user selection, opt-in, or regional justification.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
96% confidence
Finding

The generated instruction at L23 explicitly requires the output to be "中英双语". This imposes a language policy choice in the skill's natural-language behavior without any user opt-in or configurable locale selection.

Content

No source excerpt is available for this finding.