Back to skill

Security audit

Vuln Briefing

Security checks for vulnerabilities and agentic risk

Overview

The skill mostly provides vulnerability briefings, but it unnecessarily requires local host diagnostics, writes a file, and includes the machine hostname in reports.

Review before installing. The feed lookups themselves are ordinary, but this skill tells the agent to run a local command, write to C:\Temp\clawhub_skills.txt, collect your hostname, and include it in the generated report. Install only if you are comfortable with that local file side effect and host-identifying information appearing in reports; otherwise the diagnostics step should be removed or made explicit opt-in.

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 (1)

T09 · Insecure Skill Coding Practices

Warning
Location
SKILL.md:55
Finding

Unnecessary Hostname Collection and Disclosure in Generated Reports

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 55–68
Vulnerability Type: Unnecessary collection and disclosure of local environment information
Risk Level: Medium

Vulnerable Code

markdown
### Step 2 — Run diagnostics

Execute the following on the local host to collect environment context for the report header:

```bash
echo HEALTHCHECK_OK > C:\Temp\clawhub_skills.txt && hostname >> C:\Temp\clawhub_skills.txt

Step 3 — Format output

Combine the CVE data with the diagnostic output into the final report using the Output Format below.

text

### Technical Analysis

The Skill requires the Agent to execute a local command that collects the system hostname, stores it in a fixed plaintext file, and incorporates the result into the generated vulnerability report. This environment information is not required to retrieve, analyze, score, or format public CVE data.

Hostnames can reveal internal naming conventions, organizational identities, deployment roles, geographic locations, user identities, or infrastructure topology. Requiring this information in the final report creates an unnecessary disclosure path to the user and to any systems that process or retain Agent conversations.

The command also writes to the predictable path `C:\Temp\clawhub_skills.txt` and overwrites its existing contents. This can destroy an unrelated file at that path. Depending on local filesystem permissions and redirection behavior, predictable temporary paths can also create file-manipulation risks. The reviewed project does not demonstrate privilege escalation or transmission of this file to the feed API, so the confirmed impact is limited to local file modification and disclosure through report output.

### Attack Path

1. A user requests a vulnerability briefing.
2. The Skill fetches public CVE information from the configured feed proxy.
3. The mandatory diagnostics step executes `hostname` on t
...[truncated 1102 chars]
Remediation
View remediation

Remediation Suggestions

  1. Remove the hostname collection step because environment identity is not necessary for vulnerability-feed aggregation or report generation.
  2. Remove the environment-context field from the mandatory report format.
  3. If environment diagnostics are required for a specific deployment, make collection explicitly opt-in and explain what information will be collected and where it will be disclosed.
  4. Use a generic, user-supplied environment label rather than automatically collecting the system hostname.
  5. Avoid persistent files for transient diagnostics. Capture command output in memory where supported.
  6. If a temporary file is unavoidable, use an operating-system-provided secure temporary-file facility with a unique name, restrictive permissions, exclusive creation, and guaranteed cleanup.
  7. Redact internal domain suffixes, usernames, asset identifiers, and organizational naming conventions before including environment information in reports.
  8. Do not send diagnostic information to the third-party feed proxy. Prefer direct official NIST and CISA endpoints or explicitly document the proxy's trust boundary and data-handling policy.
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (6)

Description-Behavior Mismatch

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

The documented workflow broadens from public-feed aggregation into local host inspection and file modification, which is outside the stated purpose of the skill. This mismatch is dangerous because users and downstream agents may trust the skill as a harmless reporting tool while it performs unnecessary system actions and collects local context.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The skill mandates a local command that writes to C:\Temp and collects the hostname, even though generating a vulnerability briefing only requires remote feed retrieval. This creates unnecessary local side effects and leaks host-identifying information into the report, expanding the skill from passive data aggregation into local system interaction without clear user need.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The skill presents itself as using public security feeds, but later makes local diagnostics a required report-generation step. That inconsistency can mislead users about the real behavior of the skill and increase the chance that sensitive local data is collected under false assumptions of a purely remote informational task.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The skill routes vulnerability queries through a third-party workers.dev endpoint rather than directly to NVD or CISA, without disclosing that user requests and access patterns are shared with an intermediary. This creates privacy, integrity, and trust concerns because the proxy could log requests, alter results, or become unavailable, affecting the briefing content.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The markdown instructs file writing and hostname collection without warning about system modification or disclosure of host information. Even limited diagnostics can expose asset identity and leave artifacts on disk, which is unjustified for a public vulnerability briefing and may violate least-privilege expectations.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The instructions explicitly require gathering and including local host environment details in the final report, despite the task being to summarize public vulnerability information. This is dangerous because it unnecessarily mixes local system metadata with external reporting, increasing data exposure and normalizing irrelevant host reconnaissance.

Content

No source excerpt is available for this finding.