Back to skill

Security audit

Omni Learning Anything

Security checks for vulnerabilities and agentic risk

Overview

The skill’s cloud publishing, account connections, and optional lesson schedules are disclosed and tied to learner-selected settings.

Cloud delivery uploads course text, images, and agreed source excerpts to your chosen provider. Review its account grants and destination, keep private course archives outside the installation, and use current patched dependencies and reviewed installer versions. Verify scheduled access on your actual host; live integrations and resolved dependency versions were not tested here.

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
Vulnerability Patterns
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (80)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

The supplied code exposes installation-management operations rather than course creation or learning support. These are substantive, undeclared capabilities involving local filesystem resources, not ordinary implementation details of the described educational workflow. The underlying omni_installation module is not shown, so the precise filesystem effects and checksum validation cannot be confirmed; however, the installation-related command routing is explicit.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

The code is a DingTalk publishing and verification adapter rather than a course-generation component. Checking lesson text, source links, and illustrations supports the declared educational purpose, but uploading documents and accessing authenticated cloud resources introduce a material external-service capability not disclosed in the description. The supplied chunk does not show credential theft or unrelated command execution; the mismatch is the undeclared authenticated publishing integration.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

The declared purpose covers course creation and learning support but does not disclose publishing course materials to Feishu or modifying an external document repository. This chunk primarily implements authenticated external document delivery and verification, rather than course construction or instruction. Although it could support a broader course workflow, uploading local content and creating remote documents are substantive capabilities not represented in the supplied description.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

The supplied chunk implements skill installation lifecycle management rather than course creation or study support. Its filesystem changes include writing Python scripts and skill configuration, replacing existing managed installations, and creating tracking metadata. These are substantive administrative capabilities not disclosed in the learning-focused description, rather than ordinary storage of course content or learner progress. The implementation includes checksum, path, symlink, and integrity safeguards; no network download or execution of archived scripts is shown, but those safeguards do not resolve the description–behavior mismatch.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
95% confidence
Finding

The supplied commands perform software installation and update operations, which are not disclosed in the educational description. Managing deployed skill software is a separate operational capability, rather than an implementation detail of lesson generation or course planning. The underlying Python implementation is not shown, so the assessment is limited to the operations requested by these commands.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
95% confidence
Finding

Authentication could support access to learner materials, but the description does not disclose a Tencent Docs integration or account authorization with document-write access. This chunk implements that additional capability rather than merely processing supplied learning sources. It restricts requests to Tencent's HTTPS domain and protects credentials with an OS keyring, but those safeguards do not resolve the disclosure mismatch. The code requests document permissions; it does not itself read or modify documents.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
94% confidence
Finding

Illustrated Markdown could support the declared learning workflow, and the absence of course-generation logic in this individual module is not itself a mismatch. However, the code adds an externally visible publishing capability: it sends local artifacts to Notion and creates or modifies workspace content. The description does not disclose Notion integration, cloud publication, or remote workspace writes. These are substantive capabilities beyond merely formatting or verifying course materials.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
93% confidence
Finding

The supplied command performs diagnostics using a private staging configuration. Nothing in the chunk connects that operation to course creation or delivery, so it is not established as a supporting implementation detail. The command alone does not establish the configuration's contents or any further effects.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
93% confidence
Finding

The content-validation checks support illustrated, source-grounded lessons, but the chunk's primary operation is authenticated third-party document publishing rather than course construction or study support. Uploading learner or course material and creating a remote document are externally consequential capabilities not disclosed in the description. The code does not demonstrate credential theft or arbitrary file collection; the mismatch is the undeclared cloud publishing capability.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
93% confidence
Finding

The declared purpose concerns educational content and study workflows, while the supplied commands perform host diagnostics. No connection to course creation or delivery is established that would make these checks ordinary supporting implementation details. The commands do not establish what diagnostic data is collected or whether any data is transmitted.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
92% confidence
Finding

Publishing course materials could support the declared learning workflow, but this chunk implements a substantial external-account integration with persistent cloud writes. The description does not disclose Tencent Docs access, document uploads, or management of cloud files and folders. These externally visible capabilities go beyond incidental lesson-building implementation details. The code includes account, upload-host, and content-verification safeguards; those safeguards do not resolve the description mismatch.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
90% confidence
Finding

The supplied commands perform global skill management rather than course creation or learning support. They inspect installed skills and change the installed skill version, capabilities not disclosed in the description. Unlike an ordinary course-generation implementation detail, this chunk operates on the skill installation environment and contains no learning-related behavior.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
90% confidence
Finding

These adapters could support a larger course workflow, but authenticated cloud storage access, remote modification, and file uploading are substantive capabilities not disclosed by the description. The endpoint restrictions, identity checks, and Feishu upload-path restriction constrain those capabilities but do not make them declared. This chunk exposes these operations; it does not establish that any particular upload or modification has occurred.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
88% confidence
Finding

Accessing learner-provided materials could support course creation, but this chunk goes beyond reading supplied sources: it provisions and manages personal workspace integrations and initiates an authentication flow. Those account and connection-management capabilities are not disclosed in the declared purpose. The commands contain placeholder destinations, and there is no evidence here of document retrieval, exfiltration, or course generation.

Content

No source excerpt is available for this finding.

Credential Access

High
Category
Privilege Escalation
Confidence
70% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · references/delivery.md (reported line 92)May include surrounding context.

md
A paused course can inspect delivery state but cannot publish an undelivered daily lesson. Completion stops the host schedule; API timeouts never skip a day. Remote edits made later by the learner are outside the local archive's immutability contract.

Tencent `mode: mcp` stores a non-secret `auth_profile` and verified `account_id`; OS keyring stores OAuth credentials. The explicit home root is `destination.folder_id: "/"`. HTML is a standalone image-embedded local artifact, with the same manifest/receipt requirements. Legacy generic host/import configurations remain readable for existing courses; they are not advertised as native support for unspecified platforms.

Credential Access

High
Category
Privilege Escalation
Confidence
70% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · scripts/omni_readiness.py (reported line 114)May include surrounding context.

python
modules['requests'] = 'requests'
        if target['platform'] == 'tencent' and target['mode'] == 'api': modules['qcloud_cos'] = 'cos-python-sdk-v5'
    if target and target['platform'] == 'tencent' and target['mode'] == 'mcp':
        modules['keyring'] = 'keyring'
    for module, package in modules.items():
        try:
            available = importlib.util.find_spec(module) is not None

Credential Access

High
Category
Privilege Escalation
Confidence
70% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · scripts/omni_readiness.py (reported line 114)May include surrounding context.

python
modules['requests'] = 'requests'
        if target['platform'] == 'tencent' and target['mode'] == 'api': modules['qcloud_cos'] = 'cos-python-sdk-v5'
    if target and target['platform'] == 'tencent' and target['mode'] == 'mcp':
        modules['keyring'] = 'keyring'
    for module, package in modules.items():
        try:
            available = importlib.util.find_spec(module) is not None

Credential Access

High
Category
Privilege Escalation
Confidence
70% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · scripts/omni_setup.py (reported line 10)May include surrounding context.

python
from omni_delivery import MODES,default_formats

PACKAGES={'tzdata':'tzdata>=2025.1','docx':'python-docx>=1.1,<2','PIL':'Pillow>=10,<13','pypdf':'pypdf>=5,<7',
          'pypdfium2':'pypdfium2>=4,<6','requests':'requests>=2.32,<3','keyring':'keyring>=25.0,<26','qcloud_cos':'cos-python-sdk-v5>=1.9,<2'}
TOOLS={'notion':('ntn','https://developers.notion.com/cli/get-started/installation'),
       'feishu':('lark-cli','https://github.com/larksuite/cli'),
       'dingtalk':('dws','https://github.com/DingTalk-Real-AI/dingtalk-workspace-cli/releases')}

Credential Access

High
Category
Privilege Escalation
Confidence
70% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · scripts/omni_setup.py (reported line 10)May include surrounding context.

python
from omni_delivery import MODES,default_formats

PACKAGES={'tzdata':'tzdata>=2025.1','docx':'python-docx>=1.1,<2','PIL':'Pillow>=10,<13','pypdf':'pypdf>=5,<7',
          'pypdfium2':'pypdfium2>=4,<6','requests':'requests>=2.32,<3','keyring':'keyring>=25.0,<26','qcloud_cos':'cos-python-sdk-v5>=1.9,<2'}
TOOLS={'notion':('ntn','https://developers.notion.com/cli/get-started/installation'),
       'feishu':('lark-cli','https://github.com/larksuite/cli'),
       'dingtalk':('dws','https://github.com/DingTalk-Real-AI/dingtalk-workspace-cli/releases')}

Credential Access

High
Category
Privilege Escalation
Confidence
70% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · scripts/omni_setup.py (reported line 29)May include surrounding context.

python
if 'pdf' in formats:modules+=['pypdf','pypdfium2'];tools.append({'name':'XeLaTeX','next_action':'Check the existing official TeX distribution; compile and inspect the actual plan to verify packages/fonts.'})
    if material_format=='pdf' and 'pypdf' not in modules:modules.append('pypdf')
    if mode in ('api','mcp'):modules.append('requests')
    if platform=='tencent' and mode=='mcp':modules.append('keyring')
    if platform=='tencent' and mode=='api':modules.append('qcloud_cos')
    if mode=='cli':
        name,url=TOOLS[platform];binary=shutil.which(name)

Credential Access

High
Category
Privilege Escalation
Confidence
70% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · scripts/omni_tencent_auth.py (reported line 1)May include surrounding context.

python
"""Official Tencent Docs public OAuth device client; secrets stay in OS keyring."""
import json,re,time
from urllib.parse import urlsplit,urlencode
from urllib.request import Request,build_opener,HTTPRedirectHandler

Credential Access

High
Category
Privilege Escalation
Confidence
70% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · scripts/omni_tencent_auth.py (reported line 39)May include surrounding context.

python
class KeyringStore:
    def __init__(self):
        try:
            import keyring
            backend=keyring.get_keyring()
            module=type(backend).__module__
            if module not in ('keyring.backends.macOS','keyring.backends.Windows','keyring.backends.SecretService','keyring.backends.libsecret'):raise RuntimeError('Select an OS credential store; plaintext keyring backends are not supported.')

Credential Access

High
Category
Privilege Escalation
Confidence
70% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · scripts/omni_tencent_auth.py (reported line 40)May include surrounding context.

python
def __init__(self):
        try:
            import keyring
            backend=keyring.get_keyring()
            module=type(backend).__module__
            if module not in ('keyring.backends.macOS','keyring.backends.Windows','keyring.backends.SecretService','keyring.backends.libsecret'):raise RuntimeError('Select an OS credential store; plaintext keyring backends are not supported.')
            self.keyring=keyring

Credential Access

High
Category
Privilege Escalation
Confidence
70% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · scripts/omni_tencent_auth.py (reported line 40)May include surrounding context.

python
def __init__(self):
        try:
            import keyring
            backend=keyring.get_keyring()
            module=type(backend).__module__
            if module not in ('keyring.backends.macOS','keyring.backends.Windows','keyring.backends.SecretService','keyring.backends.libsecret'):raise RuntimeError('Select an OS credential store; plaintext keyring backends are not supported.')
            self.keyring=keyring

Credential Access

High
Category
Privilege Escalation
Confidence
70% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · scripts/omni_tencent_auth.py (reported line 42)May include surrounding context.

python
import keyring
            backend=keyring.get_keyring()
            module=type(backend).__module__
            if module not in ('keyring.backends.macOS','keyring.backends.Windows','keyring.backends.SecretService','keyring.backends.libsecret'):raise RuntimeError('Select an OS credential store; plaintext keyring backends are not supported.')
            self.keyring=keyring
        except ImportError:raise RuntimeError('Install requirements-tencent-mcp.txt in the selected Python environment for OS keyring authorization.') from None
    def get(self,name):