Back to skill

Security audit

value-pricing

Security checks for vulnerabilities and agentic risk

Overview

This pricing research skill needs review because its crawler can be tricked into accessing internal services despite its stated public-web restriction.

Keep the crawler disabled or run it with network controls that block private, loopback, and link-local destinations until connection-level address enforcement is fixed. Review and minimize the exact prompts before approving external model processing, which may incur charges. Correct the elasticity calculation before relying on it for pricing decisions. The network concern is supported by code inspection; actual exploitation depends on DNS and network conditions.

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
scripts/crawl_site.py
Finding

Separate DNS validation and connection allow DNS-rebinding SSRF

Content
View full analysis

Impact

Unauthorized internal HTTP GET requests are possible. Internal HTML returned with status 200 can be written into the study corpus. Automatic disclosure to the attacker or code execution is not established.

Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Behavioral ASTexec() Call, eval() Call, Dynamic Import
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Findings (7)

Skill declares no tool scope ('permissions' or 'allowed-tools') but code capabilities were detected: env, file_read, file_write, network, shell.

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.

Ae4

Medium
Category
analysis-evasion
Confidence
80% confidence
Finding

Suspicious Unicode normalization or mixed-script content

Content

No source excerpt is available for this finding.

Ae4

Medium
Category
analysis-evasion
Confidence
80% confidence
Finding

Suspicious Unicode normalization or mixed-script content

Content

No source excerpt is available for this finding.

The reported transaction-quantity elasticity is actually the slope of the latent, untruncated count mean.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

The reported elasticity uses beta_segment, which is the log-price slope of the latent negative-binomial mean, not the elasticity of expected quantity conditional on a positive transaction. Truncation changes that expectation, so the label and interval can materially mislead pricing decisions even though the separate scenario calculations account for truncation; the observational-use warnings do not correct this estimand mismatch.

Content

No source excerpt is available for this finding.

Ae4

Medium
Category
analysis-evasion
Confidence
80% confidence
Finding

Suspicious Unicode normalization or mixed-script content

Content

No source excerpt is available for this finding.

Ae4

Medium
Category
analysis-evasion
Confidence
80% confidence
Finding

Suspicious Unicode normalization or mixed-script content

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/llm_batch.py (reported line 71)May include surrounding context.

python
def call_cli(prompt, system, model):
    allowed_env = ("PATH", "HOME", "TMPDIR", "LANG", "LC_ALL", "LC_CTYPE", "TERM", "CLAUDE_CONFIG_DIR")
    env = {k: os.environ[k] for k in allowed_env if k in os.environ}
    p = subprocess.run(["claude", "-p", "--model", model, "--output-format", "json", "--tools", "", "--system-prompt", system,
                        "--no-session-persistence", "--setting-sources", "", "--strict-mcp-config"],
                       input=prompt, capture_output=True, text=True, timeout=900, env=env, cwd="/tmp")
    if p.returncode: