Back to skill

Security audit

Amazon Review Analyzer

Security checks for vulnerabilities and agentic risk

Overview

This review-analysis skill is mostly coherent, but it has a real credential-safety concern and retains review identity data locally and with a third-party service.

Use this only if you are comfortable giving Reveyes an API key and having fetched tasks/results retained by that service. Do not set REVEYES_BASE_URL or --base-url unless you fully trust the destination; keep output directories private because raw files and reviews.jsonl may contain reviewer names/profile URLs, and delete or redact raw artifacts before sharing anything other than report.html.

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 (2)

T09 · Insecure Skill Coding Practices

Warning
Location
scripts/reveyes_client.py:104
Finding

API Key Disclosure Through an Unrestricted API Base URL

Content
View full analysis
Remediation
View remediation
str: parsed = urlsplit(value) if parsed.scheme != "https": raise ReveyesError("The Reveyes API endpoint must use HTTPS") if parsed.hostname != "server.reveyes.cn": raise ReveyesError("Untrusted Reveyes API hostname") if parsed.username or parsed.password or parsed.fragment: raise ReveyesError("Invalid Reveyes API URL") return value.rstrip("/") ``` ]]>

T09 · Insecure Skill Coding Practices

Note
Location
scripts/review_pipeline.py:390
Finding

Unnecessary Local Retention of Reviewer Identity Data

Content
View full analysis
len(str(existing[field])): existing[field] = normalized[field] ``` ```python # scripts/review_pipeline.py:809-821 def _finalize_run( run_dir: Path, manifest: dict[str, Any], raw_results: list[tuple[dict[str, Any], dict[str, Any], str]], output_root: Path, ) -> None: reviews = normalize_results(raw_results) write_reviews(run_dir / "reviews.jsonl", reviews) stats = compute_statistics(reviews, manifest) write_json(run_dir / "statistics.json", stats) prepare_analysis(run_dir) ``` ```python # scripts/review_pipeline.py:904-905 raw_path = run_dir / "raw" / f"{safe_slug(request_entry['id'])}-{task_id}.json" write_json(raw_path, result) ``` ### Technical Analysis The API returns reviewer names and profile URLs. The public report renderer correctly omits these fields, and the semantic-analysis projection does not require them. Nevertheless, the normalization pipeline copies both fields into normalized review records, merges them when duplicate reviews are encountered, and writes them to `reviews.jsonl`. The pipeline also writes the complete API response under the run's `raw` directory. Consequently, reviewer identity data is retained in at least two local forms even though it is not necessary for statistics, semantic analysis, evidence validation, or report generation. No restrictive file permissions, configurable rete ...[truncated 1505 chars]
Remediation
View remediation
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (14)

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The skill advertises a full review-analysis workflow with cost-aware planning, task reuse, evidence-grounded analysis, validation, and HTML reporting, but the finding indicates the implemented behavior is only a generic API client. This mismatch is dangerous because users may rely on promised safeguards and billing controls that are not actually enforced, leading to unexpected charges, incomplete analysis, insecure handling of outputs, or unsafe operator assumptions about what the skill validates and protects.

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
94% confidence
Finding

The skill requires environment variables, file reads/writes, and network access, but it does not declare any explicit tool scope such as permissions or allowed-tools. That creates an authorization gap: a caller or platform reviewer cannot easily determine or constrain the skill's operational capabilities, increasing the chance of unintended network access, local file exposure, or unsafe execution in environments that rely on manifest-declared boundaries.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The default prompt hard-codes Chinese output ('生成中文 HTML 报告') without any user opt-in or locale negotiation. This can override user expectations or downstream workflow requirements, causing misleading output behavior and reducing user control, though it does not by itself create direct code-execution or data-exfiltration risk.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The HTML document declares lang="zh-CN", and the visible UI text throughout the template is fixed in Chinese. Under the policy, forcing a specific language/locale without user opt-in is a natural-language policy violation unless the constraint is explicitly justified, which is not present here.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

Line L064 requires a Chinese display name and category for each final theme, which imposes a specific language/locale requirement in the skill instructions. The file does not indicate that this is optional, user-selected, or justified as a region-specific constraint.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The example final object hard-codes "language": "zh-CN" and a Chinese report title, which conveys a fixed output-language requirement. In this reference document there is no stated user opt-in or region-specific justification for forcing Chinese, so it conflicts with the policy against unprompted language/locale constraints.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The documentation explicitly states that tasks and results are permanently retained, and the returned data includes personal fields such as user_name and profile_url. Even if the skill advises excluding some fields from public reports, permanent third-party retention without a clear end-user warning or consent flow creates a real privacy risk because submitted queries and fetched review data may remain indefinitely on the vendor's systems.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The routing guidance and intent mapping are built around Chinese trigger phrases such as “分析这个产品” and “快速看看”, and line 13 explicitly references the Chinese word “分析” as the term to interpret. This creates a language-specific behavior without any stated opt-in, alternative locale support, or justification that the skill is intended only for a Chinese-language environment.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

This file contains user-facing error messages in Chinese only, which forces a specific language on users. The file does not offer localization, language selection, or any documented justification that the skill is region-specific.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

This code file contains natural-language and behavior that force a specific locale by default via report_language="zh-CN". The policy allows locale constraints only when the skill offers a user choice or clearly documents and justifies the restriction; this file instead applies Chinese defaults across planning, templates, and CLI arguments.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The argparse configuration sets --report-language to zh-CN by default for the plan command, and the same pattern appears for retrieve. That imposes a language policy on users unless they actively override it, which matches the rule's locale-policy violation criteria.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The retrieve subcommand also defaults --report-language to zh-CN, continuing the forced-language behavior. Because the file does not present this as a documented region-specific skill or ask the user to choose a locale, this is a policy violation.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
92% confidence
Finding

The fallback title is hard-coded as Chinese text ('Amazon 评论分析报告'), which imposes a specific language choice when no override or analysis-provided title is present. The file does not indicate that users can opt into this locale or that the skill is intentionally region-specific.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
84% confidence
Finding

This code creates parent directories and writes an HTML report file, which is a file-system modifying operation. Although the module docstring explains the purpose, there is no confirmation prompt, print/log disclosure, or inline warning near the write operation to make the side effect visible to the user at execution time.

Content

No source excerpt is available for this finding.