Back to skill

Security audit

Learning Pack

Security checks for vulnerabilities and agentic risk

Overview

The pack is mostly a disclosed learning bundle, but it includes broad command, messaging, sub-agent, and Git mutation guidance with unsafe examples that warrant review before installation.

Review this package before installing, especially if you are a beginner or will use it in repositories with private code. Do not paste access tokens, passwords, or API keys into chat. Avoid the git-aware auto-commit helper unless you have checked git status and the diff yourself, and do not push commits until you have verified no secrets or unrelated files were included.

Vulnerability Patterns
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • 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 (2)

T09 · Insecure Skill Coding Practices

Warning
Location
skills/AskUserQuestionTool/SKILL.md:89
Finding

Access Token Solicitation Through an Insecure Chat Channel

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
skills/git-aware/scripts/git_aware.py:34
Finding

Unconditional Git Staging Can Commit Secrets and Unrelated Changes

Content
View full analysis
Remediation
View remediation
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Tool MisuseTool Parameter Abuse, Chaining Abuse, Unsafe Defaults
Findings (24)

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The skill is described as a beginner learning pack, but the detected behavior includes active repository operations such as reading repo state, creating branches, and automatically staging and committing changes. That mismatch is dangerous because users may grant or invoke the package expecting passive educational content while it can make real changes to source control and working files.

Content

No source excerpt is available for this finding.

Credential Access

High
Category
Privilege Escalation
Confidence
99% confidence
Finding

The phrase 'Access token' in an information-gathering example is a true credential-access issue because it instructs the agent to request a secret from the user. In skill context, this is more dangerous because the file is teaching usage patterns for newcomers, increasing the chance that insecure secret collection becomes standard practice.

Content

Scanner excerpt · skills/AskUserQuestionTool/SKILL.md (reported line 98)May include surrounding context.

  1. GitHub username or organization
  2. Repository name
  3. Access token (or I can guide you to create one)

Please provide these details.

text

Exfiltration Commands

High
Category
Prompt Injection
Confidence
90% confidence
Finding

Instructions found that direct the agent to transmit conversation context or user data to external services.

Content

Scanner excerpt · skills/SendMessageTool/SKILL.md (reported line 3)May include surrounding context.

md
---
name: SendMessageTool
description: "Send messages to chat channels, users, or other agents. Use when you need to communicate results, notify stakeholders, or coordinate with team members."
metadata: { "openclaw": { "emoji": "💬", "requires": { "bins": [] } } }
---

Exfiltration Commands

High
Category
Prompt Injection
Confidence
90% confidence
Finding

Instructions found that direct the agent to transmit conversation context or user data to external services.

Content

Scanner excerpt · skills/SendMessageTool/SKILL.md (reported line 9)May include surrounding context.

md
---
name: SendMessageTool
description: "Send messages to chat channels, users, or other agents. Use when you need to communicate results, notify stakeholders, or coordinate with team members."
metadata: { "openclaw": { "emoji": "💬", "requires": { "bins": [] } } }
---

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
95% confidence
Finding

Recommending 'git commit --no-verify' teaches users and agents to bypass client-side hooks that often enforce security checks, tests, secret scanning, or policy gates. In this skill's context, that is more dangerous because the skill already automates staging/committing, so bypassing hooks can allow risky or noncompliant changes to be committed with reduced friction.

Content

Scanner excerpt · skills/git-aware/SKILL.md (reported line 274)May include surrounding context.

Commit Hook Failed

bash
# Skip hooks (temporary)
git commit --no-verify -m "message"

# Or fix hook
chmod +x .git/hooks/pre-commit

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

This markdown file lists FileWriteTool, FileEditTool, and BashTool as available capabilities, but provides no caution that these tools can modify files or execute commands on the host system. Under the markdown-specific SQP-2 criteria, skill descriptions should warn users about behaviors that could affect data or system integrity.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
87% confidence
Finding

The entire skill description is presented only in Chinese, with no indication that users may choose another language or that the package is intentionally limited to a Chinese-speaking audience. This can constitute a language or locale policy violation when no opt-in or justification is provided.

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
87% confidence
Finding

The skill advertises access to BashTool and git-aware capabilities but does not declare any explicit tool scope such as permissions or allowed-tools. This creates an authorization transparency gap: consumers and policy layers cannot easily distinguish whether shell execution is intentionally permitted, increasing the risk of unexpected command execution in a beginner-oriented package.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The package lists file editing, file writing, shell execution, and messaging capabilities without warning users that these tools can modify local data, run commands, or send content outward. In a beginner-focused skill pack, missing impact disclosure increases the chance of unsafe use by inexperienced users who may not understand the consequences of invoking these capabilities.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The manifest description says to use this skill whenever the agent needs user input, confirmation, or clarification, which are very common conversational situations. Although the markdown later gives some usage guidance, the top-level activation description does not define narrow trigger phrases or clear scope constraints, increasing the risk of unintended invocation.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
98% confidence
Finding

The skill’s example explicitly asks the user to provide a GitHub access token in chat, which directly contradicts its own security guidance not to request sensitive data through chat. This can normalize insecure credential collection and lead to token exposure in conversation logs, transcripts, or downstream integrations.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The description and 'When to Use' section say to use this skill whenever you need to 'communicate results,' 'notify stakeholders,' or 'coordinate with team members,' which are very broad, common activities. Although there are a few negative examples, the trigger scope is still not narrowly constrained to specific contexts or explicit invocation phrases, increasing the chance of unintended activation.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The skill encourages automatic staging and committing of all repository changes without clearly warning that it will modify repo state. In an agent context, this can cause unintended inclusion of unrelated files, secrets, generated artifacts, or partially reviewed changes, especially because the script uses 'git add -A' before commit.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

Documenting auto-push behavior without a privacy/integrity warning is risky because pushing sends repository contents and history to a remote system. In an agent skill, this can exfiltrate proprietary code, credentials accidentally committed earlier, or unreviewed changes to external remotes with lasting impact.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The markdown claims support for branch switching/merging, review operations, diff operations, amend commits, typed commits, and auto-push/sync workflows, but the embedded git_aware.py only implements status, commit(message, commit_type='feat'), and branch create. This is not merely incomplete documentation; the documented usage actively represents unavailable behavior as supported commands.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The code calls git(..., stderr=subprocess.DEVNULL), implying the helper supports forwarding arbitrary subprocess options, but the helper defined at L169-L175 only accepts positional git arguments and always invokes subprocess.run(..., capture_output=True, text=True). As written, the documented implementation would fail at runtime instead of safely suppressing stderr for the upstream-branch check.

Content

No source excerpt is available for this finding.

subprocess module call

Medium
Category
Dangerous Code Execution
Confidence
70% confidence
Finding

subprocess module calls execute external commands. Without careful input validation, this enables command injection.

Content

Scanner excerpt · skills/git-aware/scripts/git_aware.py (reported line 10)May include surrounding context.

python
def git(*args):
    """Run git command and return output."""
    result = subprocess.run(['git'] + list(args), capture_output=True, text=True)
    if result.returncode != 0:
        print(f"Git error: {result.stderr}", file=sys.stderr)
        sys.exit(1)

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The script performs repository-modifying actions such as branch creation, staging, and commits despite being presented as a beginner learning pack. In that context, users may reasonably expect instructional or low-risk behavior, so hidden or easy-to-trigger write operations increase the chance of unintended repository changes and trust abuse.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The commit helper stages all changes with git add -A, then commits them automatically. This can capture unrelated work, secrets, generated files, or accidental modifications across the repository, making the action overly broad and dangerous for novice users in particular.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
98% confidence
Finding

Auto-staging and committing without prior warning or confirmation creates a direct integrity risk by persisting potentially unintended repository state. In a beginner-oriented skill, this is especially hazardous because users may not understand that invoking the command will immediately record all local changes.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
83% confidence
Finding

A natural-language policy violation can occur when a skill or its documentation forces a specific language without offering user choice. This quickstart is fully presented in Chinese and does not mention that the language is fixed, optional, or intended for a specific locale or audience.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
82% confidence
Finding

This markdown file describes included skills such as BashTool, FileEditTool, and FileWriteTool, which can execute commands and modify files, but it provides no warning about effects on user data or system integrity. For a skill description, the absence of any disclosure about these behaviors can leave users unaware of operational risks.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
93% confidence
Finding

The visible natural-language fields, including display name, description, headings, and author-facing text, are all in Chinese. Per the policy, forcing a specific language without user opt-in can be a language/locale policy violation when no alternative language option or justification is provided.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
95% confidence
Finding

The document switches from English to Chinese in headings, comments, and workflow steps, but does not state that the skill is intended for Chinese-speaking users or provide an opt-in language choice. This can violate language/locale policy when a skill implicitly forces a specific language for part of its instructions.

Content

No source excerpt is available for this finding.