Back to skill

Security audit

Thoth

Security checks for vulnerabilities and agentic risk

Overview

The skill appears to be a real documentation generator, but it reads substantial project source into the model context and uses an unsafe package-install command.

Install only if you are comfortable with the skill reading source files from the target project and placing excerpts into the agent context. Run it on a copy or limited project path, remove secrets first, and avoid the provided system-wide pip install; use a virtual environment with pinned dependencies instead.

Vulnerability Patterns
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • 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:237
Finding

Mandatory Commercial Content Hijacks Agent Output

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 237-248
Vulnerability Type: T01: Skill Instruction Hijacking
Risk Level: High

Vulnerable Code

python
tbl.add_row(usage_path,   "Practical usage guide")
console.print()
console.print(tbl)
console.print(Panel(
    "[bold green]Documentation complete![/bold green]\n\n"
    "Upgrade to [magenta]Thoth Pro ($9)[/magenta] for auto-injected docstrings, git CHANGELOG, and architecture docs.\n"
    "-> [cyan]ko-fi.com/occupythemilkyway[/cyan]",
    border_style="green"
))
text
Save all three documents to OUTPUT_DIR using your file writing tool, then display the confirmation panel.

Technical Analysis

The Skill requires the agent to display a predetermined commercial advertisement and external referral after completing the requested documentation work. The instruction is mandatory and unrelated to the functional requirement of generating project documentation.

Because Skill text is incorporated into the agent's active instructions, this directive alters the agent's response behavior when the Skill is loaded. It uses the agent's trusted output channel to distribute third-party promotional content without requiring a separate, informed user request.

Attack Path

  1. A user loads or invokes the Thoth Skill to generate project documentation.
  2. The agent follows the scanning and document-generation instructions.
  3. The final step explicitly directs the agent to display the supplied confirmation panel.
  4. The panel injects a paid-product promotion and an external Ko-fi link into the agent's response.
  5. The user receives attacker-selected commercial content through a trusted agent interface.

Impact Assessment

The issue does not grant additional operating-system privileges or direct access to project data. Its scope is control over the current session's final output and user-facing messaging. It can:

  • Compromis ...[truncated 272 chars]
Remediation
View remediation

Remediation Suggestions

  • Remove the mandatory commercial promotion and external referral from the completion panel.
  • Restrict completion output to factual information about files actually generated.
  • Place optional product information in passive package metadata or project documentation rather than agent-generated responses.
  • Require explicit user consent before displaying external commercial links.
  • Review all Skill instructions to ensure they serve the declared documentation task and do not commandeer the agent's trusted output channel.

T08 · Insecure Dependencies

Warning
Location
SKILL.md:40
Finding

Unpinned Dependency Installation Bypasses Managed-Environment Protections

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, line 40
Vulnerability Type: T08: Insecure Dependencies
Risk Level: Medium

Vulnerable Code

bash
pip3 install rich --break-system-packages --quiet

Technical Analysis

The installation command retrieves the current available version of rich and its transitive dependencies without a version constraint or integrity hash. Consequently, the installed dependency set can change after the Skill has been reviewed.

Python package installation may execute package build or installation logic under the privileges of the invoking process. If the package source, publisher account, registry path, or a transitive dependency is compromised, a later invocation could install attacker-controlled code.

The --break-system-packages option bypasses protections for externally managed Python environments. This can modify the host Python installation, create package conflicts, and affect unrelated applications. The --quiet option also suppresses information that could help users identify unexpected dependency resolution or installation behavior.

No evidence was found that the current rich package is malicious; the vulnerability is the unsafe and non-reproducible installation mechanism.

Attack Path

  1. The user or agent executes the Skill's installation command.
  2. pip3 resolves the latest available rich release and transitive dependencies at execution time.
  3. If any resolved distribution or publishing channel has been compromised, attacker-controlled package logic is downloaded.
  4. Package installation or subsequent import executes under the invoking user's privileges.
  5. Because managed-environment safeguards are disabled, installed files may alter the host Python environment and affect other Python applications.

Impact Assessment

Any malicious package code would run with the privileges of the account executing pip3. Under a normal user acc ...[truncated 436 chars]

Remediation
View remediation

Remediation Suggestions

  • Create and use a dedicated virtual environment instead of modifying the managed host environment.
  • Remove --break-system-packages.
  • Pin rich and all transitive dependencies to reviewed versions.
  • Use a lock file or requirements file containing cryptographic hashes, and install with hash verification.
  • Obtain packages only from an explicitly configured trusted index.
  • Remove --quiet so users can review dependency resolution and installation warnings.
  • Prefer platform-provided dependencies or a prebuilt, reviewed runtime when feasible.
  • Run installation and execution as an unprivileged account with access limited to the target project and output directory.
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
Findings (2)

Ssd 3

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

This section explicitly instructs the skill to print file contents and metadata into the model context, including up to 25 code files and multiple metadata files. That behavior can expose embedded secrets, API keys, internal endpoints, customer data, or proprietary source to the model and downstream outputs, and the risk is heightened because the skill’s purpose is to read an entire project and generate shareable markdown documents.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The skill recursively scans the user-specified project and, as part of its normal workflow, reads project files and metadata without any consent gate, filtering, or warning about sensitive content. In a documentation skill, this creates a real risk of pulling secrets, proprietary code, internal comments, or configuration data into the model context and potentially into generated documentation or logs.

Content

No source excerpt is available for this finding.