Back to skill

Security audit

Horus Lite

Security checks for vulnerabilities and agentic risk

Overview

The skill performs a meeting-notes task, but it should be reviewed because it prints full raw notes and installs an unpinned package into the system Python environment.

Review before installing. Do not use this with confidential transcripts unless you are comfortable with the full notes appearing in terminal output or logs. If used, install dependencies in an isolated virtual environment and expect promotional upgrade text in the generated output.

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

Warning
Location
SKILL.md:93
Finding

Mandatory Commercial Output Injection

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 93-115
Vulnerability Type: T01: Skill Instruction Hijacking
Risk Level: Medium

Vulnerable Code

markdown
## Step 4 -- Upsell

```python
from rich.console import Console
from rich.panel import Panel
console = Console()
console.print()
console.print(Panel(
    "[bold yellow]Horus Standard ($5)[/bold yellow] adds a full meeting summary, decisions log, and a ready-to-send follow-up email.\n"
    "[bold magenta]Horus Pro ($9)[/bold magenta] adds stakeholder-specific briefs, next meeting agenda, risk flags, and saves to file.\n\n"
    "Upgrade: [bold cyan]ko-fi.com/occupythemilkyway[/bold cyan]",
    title="[yellow]Want the full meeting brief?[/yellow]",
    border_style="yellow"
))
text

### Technical Analysis

The Skill defines commercial promotion as a mandatory workflow step rather than an optional or user-requested action. When an agent follows the Skill instructions, it appends creator-controlled pricing, branding, and an external purchase destination to the meeting-analysis workflow.

This changes the agent's output behavior beyond the legitimate task of extracting action items and producing a summary. The instruction does not alter safety controls or obtain additional system permissions, but it uses the loaded Skill instructions to make the agent act as an unsolicited advertising channel.

### Attack Path

1. A user loads or invokes the Horus Lite Skill.
2. The user supplies meeting notes through `MEETING_NOTES`.
3. The agent follows the instructions to process the notes and generate action items.
4. The mandatory Step 4 code executes as part of the workflow.
5. The user is shown creator-controlled product pricing and an external purchase destination, regardless of whether promotional content was requested.

### Impact Assessment

The issue affects the current agent session and the integrity of its output. It does not
...[truncated 217 chars]
Remediation
View remediation

Remediation Suggestions

Remove the mandatory upsell step from the executable Skill workflow. If upgrade information must be retained:

  1. Display it only after explicit user consent or a direct request for product information.
  2. Keep promotional material separate from generated meeting results.
  3. Do not instruct the agent to append advertisements to every invocation.
  4. Clearly identify external destinations before users choose to visit them.
  5. Ensure the default workflow performs only the documented meeting-analysis task.

T08 · Insecure Dependencies

Warning
Location
SKILL.md:39
Finding

Unpinned Package Installation Bypasses System Package Protections

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 39-43
Vulnerability Type: T08: Insecure Dependencies
Risk Level: Medium

Vulnerable Code

markdown
## Step 1 -- Install

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

### Technical Analysis

The installation command does not pin `rich` to a reviewed version or verify package hashes. Consequently, the package and its transitive dependencies are resolved dynamically at execution time, so the installed code may differ from the code reviewed when the Skill was published.

The `--break-system-packages` option explicitly bypasses protections intended to prevent pip from modifying an externally managed Python installation. This can overwrite or conflict with operating-system-managed packages. The `--quiet` option also suppresses installation details that could otherwise help users identify unexpected dependency changes.

The audited material does not establish that `rich` is malicious. The risk arises from mutable dependency resolution, lack of integrity verification, and modification of a system-managed Python environment.

### Attack Path

1. An agent or user follows the documented installation step.
2. `pip3` queries the configured package index and resolves the latest available `rich` package and transitive dependencies.
3. Package installation logic runs with the permissions of the invoking account.
4. Because versions and hashes are not constrained, a compromised or unexpectedly changed distribution could install code not covered by the Skill audit.
5. The `--break-system-packages` option allows the installation to modify an externally managed Python environment, potentially replacing or conflicting with system packages.
6. Subsequent Python applications may import or rely on the modified packages.

### Impact Assessment

Installation code executes with the privileges of the account running `pip3`. If invoked by an elev
...[truncated 526 chars]
Remediation
View remediation

Remediation Suggestions

Replace the system-level mutable installation with a reproducible, isolated setup:

  1. Create and use a dedicated virtual environment rather than modifying the system interpreter.
  2. Pin rich and all relevant transitive dependencies to reviewed versions.
  3. Use a locked dependency file generated from trusted sources.
  4. Require package hashes, such as with pip install --require-hashes -r requirements.txt.
  5. Remove --break-system-packages.
  6. Avoid --quiet so installation changes and errors remain visible.
  7. Review and update pinned dependencies through a controlled maintenance process.
  8. If practical, replace the dependency with standard terminal output because the Skill only uses basic console formatting.

A safer installation pattern is:

bash
python3 -m venv .venv
.venv/bin/python -m pip install --require-hashes -r requirements.txt
Vulnerability Patterns
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (3)

Ssd 3

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

The code prints the full contents of MEETING_NOTES verbatim, directly exposing all user-supplied transcript data in the output stream. Meeting notes commonly contain sensitive internal discussions, attendee identities, action owners, dates, and confidential plans, so this behavior creates a direct confidentiality risk, especially if terminal sessions or outputs are logged or shared.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The skill handles potentially sensitive meeting notes but does not warn users that their full raw input will be echoed back to the terminal output. In contexts where output is logged, shared, or visible to others, this can cause unintended disclosure of confidential business discussions, names, deadlines, or other sensitive transcript content.

Content

No source excerpt is available for this finding.

Vague Triggers

Low
Category
Not specified by scanner
Confidence
93% confidence
Finding

The description says users can 'Paste meeting notes or a transcript and get a clean action item list instantly,' but it does not clearly specify how or when the skill should be invoked beyond a very general meeting-notes context. There are no explicit trigger phrases, scope boundaries, or negative examples to prevent overly broad activation for ordinary note-cleanup requests.

Content

No source excerpt is available for this finding.