Back to skill

Security audit

skill-release-audit

Security checks for vulnerabilities and agentic risk

Overview

This is a disclosed static skill-audit tool with an optional, explicit dependency install mode; I found no hidden persistence, exfiltration, publishing, deletion, or deceptive behavior.

Install is reasonable for users who want a local skill-release auditor. Treat --auto-install as a separate consent decision: it can modify the active Python environment and may fetch packages based on imports in the skill being audited, so avoid that flag for untrusted skills unless you intend to install those dependencies.

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
  • 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
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
Findings (28)

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

This second mismatch reinforces that the skill is not purely audit-only in all modes, because --auto-install changes the environment. In agent ecosystems, misleading capability descriptions can cause unsafe automation decisions, especially when users expect audits to be read-only and non-mutating.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

This second mismatch reinforces that the skill is not purely audit-only in all modes, because --auto-install changes the environment. In agent ecosystems, misleading capability descriptions can cause unsafe automation decisions, especially when users expect audits to be read-only and non-mutating.

Content

No source excerpt is available for this finding.

Anti-Refusal Statement

High
Category
Anti-Refusal
Confidence
85% confidence
Finding

Skill instructs the agent to omit warnings, disclaimers, or ethical commentary. Stripping safety caveats hides risk from the user and is a common jailbreak preamble.

Content

Scanner excerpt · SKILL.md (reported line 148)May include surrounding context.

md
refs (`~/...`, `$HOME/...`, absolute paths incl. Windows drive letters) —
the latter resolve in the *user's* environment at runtime, so existence is
not a package-completeness problem: resolvable paths are skipped silently,
unresolvable ones emit INFO (never WARN).

## Report language

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 171)May include surrounding context.

md
| 3. Edge cases | try/except coverage, HTTP timeout, response-status handling, Bash `set -e` | `scripts/check_edges.py` |

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/safe-paths.md (reported line 17)May include surrounding context.

md
|-----------|----------|
| User config | `.skill-data/<skill-name>/config.json` |
| Cache | `.skill-data/<skill-name>/cache/` |
| Token / credentials | `.skill-data/<skill-name>/credentials.json` |
| Logs | `.skill-data/<skill-name>/logs/` |
| Runtime state | `.skill-data/<skill-name>/state.json` |

Env Variable Harvesting

High
Category
Data Exfiltration
Confidence
70% confidence
Finding

Code enumerates, copies, or searches environment variables for secrets. Bulk environment access can collect credentials unrelated to the skill's stated purpose.

Content

Scanner excerpt · scripts/check_deps.py (reported line 275)May include surrounding context.

python
def extract_declared_env(skill_dir: Path) -> "set[str]":
    """Collect env vars declared in frontmatter under the ClawHub schema:
    metadata.openclaw.requires.env / primaryEnv / envVars[].name (aliases:
    metadata.clawdbot / metadata.clawdis).
    """

Description-Behavior Mismatch

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

The dependency checker performs package installation even though the skill description says it is a pure reporter that never edits files. That mismatch is dangerous because users may trust the tool as non-mutating, while it can actually alter the runtime environment, pull code from external registries, and create side effects during an audit step that should be safe and deterministic.

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
70% confidence
Finding

Without declared permissions the skill's intent is opaque and cannot be validated.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

Claiming 'never edits your files' and 'pure reporter' while also documenting dependency installation creates a misleading safety boundary. Even if files are not edited, mutating the execution environment can affect reproducibility, introduce supply-chain risk, and violate user expectations about a read-only audit tool.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

Because this is a markdown file, trigger specificity applies. Phrases like "pre-release check", "release gate", and "skill safety check" are generic enough to match ordinary discussion about releases rather than a clear, narrowly scoped invocation, and the file does not provide exclusion conditions or negative examples to bound activation.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The docs frame the skill as non-mutating, then later introduce an optional mode that mutates the environment. That inconsistency can be exploited indirectly by causing a user or orchestration layer to approve the skill under false assumptions, increasing the chance of unintended package installation or network access.

Content

No source excerpt is available for this finding.

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
75% confidence
Finding

Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.

Content

Scanner excerpt · SKILL.md (reported line 53)May include surrounding context.

md
with its specific description.
2. **Suggest fixes per finding, but only as suggestions, not actions.**
3. **Wait for explicit user confirmation** before touching any file.
4. **Never modify SKILL.md, scripts/, references/, or any other file** without
   confirmation.
5. **Never delete files**, even ones that look like leftovers — always ask first.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The file states that report language falls back to zh as a legacy default, while only zh and en are supported. This creates a language-policy issue because a specific language is forced when detection fails, rather than prompting the user or using a neutral opt-in mechanism.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
70% confidence
Finding

npx commands without a version suffix (e.g. @1.0.0) create a rug-pull risk if the upstream server is compromised and publishes a malicious update.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
70% confidence
Finding

npx commands without a version suffix (e.g. @1.0.0) create a rug-pull risk if the upstream server is compromised and publishes a malicious update.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
70% confidence
Finding

npx commands without a version suffix (e.g. @1.0.0) create a rug-pull risk if the upstream server is compromised and publishes a malicious update.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The docstring says user confirmation is baked into the main flow, but this file contains unattended installation logic once auto_install is enabled. That inconsistency weakens operator understanding and can lead to unintended code execution or environment modification if higher-level callers pass the flag without an actual confirmation boundary.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

This Python file contains natural-language documentation in Chinese for field support and behavior notes. Under the policy, forcing a specific language without user choice is a locale/language policy violation, and there is no indication here that the user opted into Chinese or that the constraint is region-specific.

Content

No source excerpt is available for this finding.

subprocess module call

Medium
Category
Dangerous Code Execution
Confidence
96% confidence
Finding

This code invokes pip installation via subprocess, which changes the local environment and executes package installation logic. In a skill advertised as a static auditor and 'pure reporter,' this creates an integrity and supply-chain risk: if enabled, the skill can install unreviewed packages from package indexes and mutate the host system.

Content

Scanner excerpt · scripts/check_deps.py (reported line 351)May include surrounding context.

python
"""Attempt pip install. Returns (success, message)."""
    pip_name = pip_package_name(pkg_name)
    try:
        result = subprocess.run(
            [sys.executable, "-m", "pip", "install", pip_name, "--quiet"],
            capture_output=True,
            text=True,

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The docstring says this check applies only when scripts have third-party imports, but the code merely checks whether any Python files exist in scripts/ and then requires a dependencies section regardless of whether those files use third-party packages. This is an active intent/behavior contradiction in inline documentation, not just an incomplete comment.

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 · scripts/check_logic.py (reported line 45)May include surrounding context.

python
def check_bash_syntax(script_path: Path) -> list[dict]:
    issues = []
    try:
        result = subprocess.run(
            ["bash", "-n", str(script_path)],
            capture_output=True,
            text=True,

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The manifest describes the skill as a non-modifying reporter with 'no network by default', but this entrypoint exposes --auto-install and passes auto_install=True into the dependency checker, enabling package installation. Even though it is opt-in and framed as off by default, it is still a behavior beyond a strict read-only auditor/pure reporter description.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The docstring frames the tool as non-modifying in normal use, yet the same module advertises and implements an installation mode that changes the environment. This is an intent-level contradiction because the top-level documentation presents the tool as an auditor while the code also acts as an installer.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

A release-audit/checking tool is expected to inspect and report, not mutate the runtime environment. Adding pip installation introduces side effects and likely network/package execution behavior that is outside the core purpose unless explicitly declared as part of scope.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

This code file contains user-facing natural language that says the report output defaults to zh, which imposes a specific language unless the user overrides it or sets an environment variable. The stated policy forbids forcing a specific language without user opt-in, and this line does not justify the locale constraint as region-specific.

Content

No source excerpt is available for this finding.