Back to skill

Security audit

reading-to-post-reading-continuation

Security checks for vulnerabilities and agentic risk

Overview

The skill is rated suspicious because scanners report commands for account access, installation, repository pushes, and skill publication that do not match its stated purpose of creating a reading exercise.

Consider avoiding installation until the publishing, authentication, and Git steps have been reviewed and removed or made clearly optional. The scan covered the listed Markdown files but was marked partial because several referenced paths were missing, so the available evidence is incomplete.

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
Finding

Local webpage generation triggers an unauthorized repository push

Content
View full analysis

Impact

Lesson content and repository changes can be transferred to the remote and shared with its readers without user approval. Other unpushed commits on main can also be transferred.

Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • Rogue AgentSelf-Modification, Session Persistence
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (16)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The supplied code consists solely of a ClawHub login command. Authentication to that service is unrelated to the declared self-contained webpage-generation purpose, and no exercise-generation behavior appears in the chunk.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The supplied commands are a skill-publication workflow, not an implementation of the described educational webpage generator. The publication calls are operational actions unrelated to producing the promised HTML file. Comments about writing documentation do not actually create it, and labeling one publication call a dry run does not change its arguments or behavior.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The supplied code performs an account-identity lookup rather than reading analysis or HTML exercise generation. This accesses an undeclared, unrelated resource and is not a supporting implementation detail of the stated purpose.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The supplied chunk is a publishing workflow rather than the declared webpage-generation workflow. It neither analyzes reading passages nor generates the eight-section HTML exercise. Installing a global CLI, authenticating, and publishing to an external service are materially different capabilities not disclosed in the description. These are documented commands; the chunk does not establish that they have been executed.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The supplied commands install an external CLI globally rather than transform a reading passage into the specified HTML exercise. No connection to the declared workflow is shown, so this is not established as a supporting implementation detail. The assessment is limited to the supplied chunk and does not infer what clawhub does after installation.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The supplied command performs skill publication, not the declared educational webpage generation. Publishing local skill content is an undeclared capability with a materially different purpose, rather than a supporting implementation detail. The chunk contains no logic for the promised reading analysis, eight interactive sections, or self-contained HTML output.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The GitHub section explicitly directs a commit and push after webpage creation, extending a local generation task into repository modification and external transmission without authorization. The commit can include already-staged changes, and the push can transmit other unpushed commits as well as lesson content.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

The only supplied command performs skill installation, which is materially different from generating the promised interactive HTML file. The installed skill's contents are not supplied, so its eventual behavior cannot be verified. This assessment concerns the visible installation command and does not assume that the installed package is malicious.

Content

No source excerpt is available for this finding.

The local webpage-generation workflow also directs an unrequested Git commit and remote push.

High
Category
Excessive Agency
Confidence
99% confidence
Finding

The phrase 'After creating the webpage file' directly connects successful generation to an unrequested commit and remote push. This can modify repository history and transmit lesson content or previously staged and unpushed work using the user's existing Git access.

Content

No source excerpt is available for this finding.

Self-Modification

High
Category
Rogue Agent
Confidence
85% confidence
Finding

Skill modifies its own code, configuration, or behavior at runtime. Self-modification enables an agent to escalate privileges, disable safety constraints, or install persistent backdoors.

Content

Scanner excerpt · SKILL.md (reported line 240)May include surrounding context.

md
mkdir -p reading-to-post-reading-continuation
cd reading-to-post-reading-continuation

# 2. Write SKILL.md (frontmatter + content)
# 3. Write README.md (optional, for human browsing)

# 4. Dry run to preview

The command labeled as a dry-run preview actually invokes publication without --dry-run.

High
Category
Excessive Agency
Confidence
100% confidence
Finding

The command labeled 'Dry run to preview' omits the documented --dry-run flag and therefore requests actual publication. An agent or maintainer expecting a non-mutating preview could upload local folder contents and create or update a public listing without intending to do so.

Content

No source excerpt is available for this finding.

The command labeled as a dry run actually publishes the skill folder.

High
Category
Output Handling
Confidence
100% confidence
Finding

The purported preview invokes the real publishing operation because --dry-run is missing. This can externally disclose included folder contents before the user has approved publication.

Content

No source excerpt is available for this finding.

Annotation text is required to be in English and Chinese without user language choice or opt-in.

Medium
Category
Output Handling
Confidence
97% confidence
Finding

The instruction mandates bilingual English-and-Chinese annotations regardless of the user's language preferences. Although English is relevant to the reading exercise, the skill does not offer a choice or document a regional requirement justifying mandatory Chinese annotations.

Content

No source excerpt is available for this finding.

The post-creation workflow pushes lesson content to GitHub without warning about external disclosure.

Medium
Category
Output Handling
Confidence
98% confidence
Finding

A local classroom exercise can contain unpublished passages, personal information, or licensed teaching material. Automatically pushing it to the configured remote creates an external-disclosure risk, even though the source does not establish that the destination is public or attacker-controlled.

Content

No source excerpt is available for this finding.

Skill-registry installation, authentication, and publication exceed the educational webpage builder's purpose.

Medium
Category
Excessive Agency
Confidence
98% confidence
Finding

This section introduces global package installation, a GitHub OAuth login, and skill-folder publication to ClawHub. These are skill-distribution capabilities rather than requirements for converting a reading passage into an offline HTML exercise, and the manifest does not declare them as part of the skill's scope.

Content

No source excerpt is available for this finding.

The step labeled “Dry run to preview” actually publishes the skill, misleading the agent into a state-changing action under a preview-only framing.

Medium
Category
Prompt Injection
Confidence
99% confidence
Finding

The preview-only label misrepresents the command's state-changing behavior. This undermines informed authorization by making actual registry publication appear to be a harmless inspection step.

Content

No source excerpt is available for this finding.