ClawHub
Back to skill

Security audit

CloudArc bounded-memory benchmark

Security checks for vulnerabilities and agentic risk

Overview

This benchmark skill is resource-heavy but its risky actions are disclosed, purpose-aligned, and gated rather than hidden or deceptive.

Install only in a CloudArc development repository where running tests, writing benchmark reports, and consuming large temporary disk are acceptable. Avoid the manual benchmark unless you intentionally want the large 1/5/10 GiB run, and review cleanup commands before running them in an unusual working directory.

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

SkillSpector

Vulnerability Patterns
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Behavioral ASTexec() Call, eval() Call, Dynamic Import
  • Taint TrackingDirect Taint Flow, Variable-Mediated Taint Flow, Credential Exfiltration Chain
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
Findings (5)

Lp3

Medium
Category
MCP Least Privilege
Confidence
94% confidence
Finding
The skill requests only `python` in frontmatter, but the body instructs actions that imply broader capabilities: reading and writing many repository files, invoking shell/PowerShell commands, deleting directories, and potentially interacting with a cloud provider via OAuth-backed remote metadata flows. This mismatch weakens least-privilege controls because an agent or reviewer cannot tell from the declared scope what powers the skill expects, increasing the chance of unintended filesystem, shell, or network use.

Autonomous Decision Making

Medium
Category
Excessive Agency
Content
`pack` returns `skipped`, `skipped_count` and `skipped_by_reason`, and `analyze`
returns `skipped` plus `skipped_summary: {count, by_reason}` (reasons:
`protected` / `system` / `special` / `symlink-target` / `changed` / `vanished`);
the CLI warns on stderr. Never treat a pack as complete without checking that count: protected
subtrees (`.git`, `__pycache__`, `.venv`), leading system paths (without
`--allow-system`) and special files are skipped by design.
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.

subprocess module call

Medium
Category
Dangerous Code Execution
Content
def _run(command: list[str], *, cwd: Path) -> int:
    print("+ " + " ".join(_display_arg(item) for item in command))
    completed = subprocess.run(command, cwd=cwd, check=False)
    return completed.returncode
Confidence
70% confidence
Finding
subprocess module calls execute external commands. Without careful input validation, this enables command injection.

Tainted flow: 'svg' from sys.stdin.read (line 479, user input) → pathlib.Path.write_text (file write)

Medium
Category
Data Flow
Content
return 0
    target = Path(args.output)
    target.parent.mkdir(parents=True, exist_ok=True)
    target.write_text(svg, encoding="utf-8")
    print(f"badge written: {target}")
    return 0
Confidence
65% confidence
Finding
Data from a source is assigned to a variable that is later passed to a sink, creating a variable-mediated taint flow.

Missing User Warnings

Low
Confidence
88% confidence
Finding
The documented cleanup commands perform recursive forced deletion (`Remove-Item -Recurse -Force`) over discovered directories, including temp paths and `__pycache__`. While the examples are somewhat targeted, there is no adjacent warning that they are destructive or guidance to verify the matched paths first, so users may run them blindly and delete unintended data if variables or search roots are broader than expected.

Static analysis

No suspicious patterns detected.