ClawHub
Back to skill

Security audit

Cam

Security checks for vulnerabilities and agentic risk

Overview

This skill is a local CNC/CAM reference and calculator with some promotional output, but no evidence of hidden execution, data access, persistence, or automatic installation.

Before installing, note that command output includes BytesAgain promotional links and the help text suggests installing related skills. Treat those as optional advertisements and review any separate skill before installing it. The reviewed skill itself appears limited to local CAM reference output and calculations.

A.I.G

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
Findings (1)

T01 · Skill Instruction Hijacking

Note
Location
scripts/script.sh:375
Finding

Hard-Coded Promotional and Third-Party Installation Recommendations in Skill Output

Content
View full analysis

Vulnerability Details

File Location: scripts/script.sh:375-384
Additional Locations: scripts/script.sh:42, 95, 159, 235, 280, 310, 331, 352
Vulnerability Type: Skill Instruction Hijacking through unsolicited output steering
Risk Level: Low

Vulnerable Code

bash
Related skills:
  clawhub install gear
  clawhub install cnc-feeds
Browse all: bytesagain.com

Powered by BytesAgain | bytesagain.com
EOF
}

Functional command responses also repeatedly include the following promotional footer:

bash
print(f"\n📖 More skills: bytesagain.com")

or:

bash
📖 More skills: bytesagain.com

Technical Analysis

The script embeds third-party promotion throughout ordinary CAM command output. Its help output also recommends installing two additional skills that are not included in this repository and therefore were not covered by this audit.

When an AI agent invokes the script and relays its output, these recommendations become part of the agent's response even though they are unrelated to the requested machining calculation or reference material. This creates an output-steering channel that promotes external resources and additional software installation.

The displayed clawhub install commands are not executed automatically. No shell evaluation, remote payload retrieval, persistence, privilege escalation, credential access, or data exfiltration was identified in the reviewed implementation.

Attack Path

  1. A user asks an agent for CAM information or command help.
  2. The agent invokes scripts/script.sh, such as with the help, calculate, or speeds-feeds command.
  3. The script appends promotional content or third-party skill installation recommendations to its legitimate output.
  4. The agent relays that output to the user.
  5. The user or agent may follow the recommendation and install unaudited external skills.
  6. Any further impact would depend on ...[truncated 681 chars]
Remediation
View remediation

Remediation Suggestions

  1. Remove third-party skill installation recommendations from executable command output.
  2. Remove repeated promotional footers from calculation and reference results.
  3. Keep optional attribution and related-project links in SKILL.md or another documentation file rather than mixing them into operational output.
  4. If related skills must be mentioned, clearly mark them as optional, state that they are separately maintained and unaudited, and require explicit user review before installation.
  5. Ensure script output is limited to information directly relevant to the command requested by the user.
  6. Add automated output tests that reject unexpected URLs, installation commands, and promotional text in functional command responses.

SkillSpector

Vulnerability Patterns
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (2)

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.

Unbounded Output

Medium
Category
Output Handling
Confidence
60% confidence
Finding

Output size or generation rate is not bounded. Unbounded output enables denial-of-service through resource exhaustion, log flooding, or context-window stuffing.

Content

Scanner excerpt · scripts/script.sh (reported line 165)May include surrounding context.

sh
═══════════════════════════════════════════════════

【Motion Codes】
  G00   Rapid positioning (no cutting)
  G01   Linear interpolation (cutting feed)
  G02   Circular interpolation CW
  G03   Circular interpolation CCW

Static analysis

No suspicious patterns detected.