Back to skill

Security audit

Model Pricing Sync

Security checks for vulnerabilities and agentic risk

Overview

The skill is mostly aligned with pricing collection and Lark syncing, but it hides stealth web-scraping behavior and can replace remote spreadsheet tabs, so users should review it before installing.

Install only if you are comfortable with automated scraping of vendor pricing pages, including concealed browser automation, and with uploading generated pricing data to Feishu/Lark. Use a least-privileged Lark identity, review the configured source URLs and CSV contents before push, pin dependencies, and avoid running it in an environment with unrelated secrets.

Vulnerability Patterns
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • 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
Findings (2)

T09 · Insecure Skill Coding Practices

Warning
Location
scripts/_shared.py:257
Finding

Spreadsheet Formula Injection Through Unsanitized Pricing Data

Content
View full analysis
str: parts = [str(value).strip() for value in values if str(value or "").strip()] return "; ".join(dict.fromkeys(parts)) def feature_summary(item: dict[str, Any]) -> str: explicit = str(item.get("plan_summary", "")).strip() if explicit: return explicit features = item.get("features", []) if isinstance(features, list): return "; ".join(str(feature).strip() for feature in features if str(feature).strip()) if isinstance(features, str): return features.strip() return "" ``` The values are inserted into records without spreadsheet-specific sanitization: ```python expanded.append( { "vendor_key": item.get("vendor_key", ""), "region": item.get("region", ""), "product_name": item.get("product_name", ""), "model_name": item.get("model_name", ""), "price_dimension": price.get("price_dimension", ""), "source_price_amount": price.get("source_price_amount", ""), "currency": price.get("currency", ""), "unit_basis": price.get("unit_basis", ""), "original_price_text": price.get("original_price_text", ""), "notes": combine_notes(item.get("notes", ""), price.get("notes", "")), "source_url": price.get("source_url", "") or item.get("source_url", ""), } ) ``` CSV serialization preserves arbitrary strings as supplied: ```python def write_csv_rows(path: Path, rows: list[dict[str, Any]], fields: list[str]) -> None: path.parent.mkdir(parents=True, exist_ok=True) with path.open("w", encoding="utf-8", newline="") as handle: writer = ...[truncated 4014 chars]
Remediation
View remediation
Any: if not isinstance(value, str): return value if value.startswith(DANGEROUS_CELL_PREFIXES): return "'" + value return value ``` 2. Apply the function to every string field in `write_csv_rows`, not only fields currently known to contain remote text. 3. If the Feishu/Lark API supports explicit cell types, submit remote and descriptive fields as literal text rather than allowing automatic formula inference. 4. Sanitize both current and historical tables because previously collected values are copied into history and change records. 5. Validate extraction JSON before building. Reject or flag formula-like values in descriptive fields unless an operator explicitly approves them. 6. Add regression tests for values beginning with: - `=` - `+` - `-` - `@` - Tab - Carriage return - Leading whitespace followed by a dangerous prefix 7. Consider adding provenance metadata or review warnings when extraction fields contain formula syntax, URLs, or unexpected control characters. ]]>

T08 · Insecure Dependencies

Note
Location
requirements.txt:4
Finding

Unpinned Executable Dependencies Create Supply-Chain Exposure

Content
View full analysis
=1.40.0 # Installation steps: # 1. pip install -r requirements.txt # 2. playwright install chromium # 3. npm install -g @larksuiteoapi/cli (for Feishu/Lark integration) ``` ### Technical Analysis The Playwright requirement uses an open-ended lower bound, allowing installation of any future compatible or incompatible release. The documented setup also installs the latest available `@larksuiteoapi/cli` globally without a fixed version or integrity constraint. These components are executable dependencies: - Playwright launches a downloaded Chromium binary and controls network browsing. - `lark-cli` is located through the process `PATH` and executed during synchronization. - The Lark CLI operates using the selected authenticated identity and receives spreadsheet data through standard input. The project does not provide a lock file, package hashes, a pinned Chromium revision policy, or a required CLI version check. Consequently, a future compromised release, registry compromise, or unexpected dependency update could change the behavior after the Skill itself has been reviewed. No evidence was found that the currently named packages are typosquatted or malicious. The issue is the mutable and unverified installation process. ### Attack Path 1. A dependency account, package registry, release pipeline, or future package version is compromised. 2. An operator follows the documented installation instructions. 3. The package manager resolves `playwright>=1.40.0` or the unversioned global Lark CLI to the compromised release. 4. The dependency executes during browser installation, collection, or spreadsheet synchronization. 5. A compromised Lark CLI could operate with the configured Lark identity, inspect submitted pricing data, alter API requests, or access cred ...[truncated 1036 chars]
Remediation
View remediation
``` 2. Generate and commit a lock file containing exact transitive versions. 3. Use package hashes, such as pip hash-checking mode, to verify downloaded artifacts. 4. Pin the Lark CLI installation to a reviewed version: ```bash npm install -g @larksuiteoapi/cli@ ``` 5. Record and verify the expected Chromium revision installed by Playwright. 6. Resolve `lark-cli` from a controlled installation directory rather than relying solely on ambient `PATH` ordering. 7. Before execution, verify the resolved CLI path and version against an allowlist. 8. Run collection and synchronization in a restricted environment with: - Minimal filesystem access. - No unrelated environment secrets. - A least-privileged Lark identity. - Network access limited to configured pricing sites and official Lark API endpoints. 9. Establish a dependency-update process requiring review, integrity verification, and security testing before version changes are accepted. ]]>
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Behavioral ASTexec() Call, eval() Call, Dynamic Import
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
Findings (29)

Tp4

High
Category
MCP Tool Poisoning
Confidence
94% confidence
Finding

This mismatch is more serious because the analyzed behavior reportedly includes browser automation, stealth/UA spoofing, webdriver hiding, retries, and dynamic expansion that are not disclosed in the skill description. Undisclosed anti-bot evasion and automation can expose operators to policy, legal, and security risk, especially when combined with network and shell access.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

This mismatch is more serious because the analyzed behavior reportedly includes browser automation, stealth/UA spoofing, webdriver hiding, retries, and dynamic expansion that are not disclosed in the skill description. Undisclosed anti-bot evasion and automation can expose operators to policy, legal, and security risk, especially when combined with network and shell access.

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
92% confidence
Finding

The skill requests or implies powerful capabilities (file read/write, network, shell) but does not declare any tool scope or permission boundaries. This creates an over-privileged, opaque execution model where an operator may invoke scraping, local file mutation, and command execution without clear review of what the skill is allowed to do.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The skill instructs users to push synchronized data to Lark Sheets but provides no user-facing warning or confirmation about external data transmission. Even if the intended data is pricing information, silent export to an external SaaS destination can leak internal annotations, notes, or unintended content from local artifacts.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The short description is written entirely in Chinese, indicating a fixed language choice in the skill’s natural-language interface without offering the user a language or locale option. This can violate language/locale policy when the skill is presented in a broader environment where users are not explicitly opting into Chinese.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The fixed spreadsheet title is hard-coded in Chinese ("模型价格追踪"), which imposes a specific language choice in the skill behavior. Under the language/locale policy, locale-specific output should either be user-selectable or clearly justified as 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

The raised RuntimeError message is written only in Chinese, which forces a specific language for user-visible behavior. The file does not indicate user opt-in or a documented regional justification for this locale restriction.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

These RuntimeError strings are user-visible and written only in Chinese, which enforces a single language/locale across failure paths. No surrounding documentation in this file offers language selection or explains a required China-specific scope.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
98% confidence
Finding

This code emits natural-language runtime errors in Chinese, such as the message at L0462, with no indication that the skill is Chinese-only or that users can choose their preferred language. That creates a language/locale policy issue because the skill enforces a specific language in user-facing output without opt-in.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The function writes to persistent files like current_api.csv and history_api.csv, and similar writes continue later in the file, but there is no confirmation prompt, user-facing log/print, or explanatory docstring/comment disclosing that stateful data will be overwritten or appended. For a code file, these file modifications qualify as safety-relevant operations lacking visible disclosure in the file itself.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

These lines append or overwrite multiple tracking artifacts and run metadata files, but the code provides no print/log message, confirmation, or top-level documentation warning users that running the build mutates durable records. Because these are file writes in a code file, they meet the missing-warning criterion when no disclosure is present in the file.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The hard-coded user agent is a forged browser fingerprint used to impersonate a normal Chrome user rather than an automation client. Combined with the rest of the collection flow, this facilitates deceptive access to third-party pages and can help evade anti-bot defenses, which is risky in a pricing-sync scraper context.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The code intentionally injects anti-detection JavaScript to hide Playwright automation by altering navigator.webdriver, platform, language, and window.chrome. In a data collection skill, this crosses from ordinary browsing into stealth scraping behavior that can bypass bot-detection controls and violate site access restrictions, increasing legal, compliance, and abuse risk.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The code injects a stealth script that masks automation signals and overrides browser language/platform properties, and later uses it during page collection. This is a safety-relevant network interaction behavior, but the file provides no confirmation prompt, logging, or explanatory comment/docstring warning that requests may be sent with spoofed browser attributes.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The browser context is hard-coded to en-US and America/New_York while also using a forged user agent and Accept-Language headers, creating a fabricated client profile. In this scraping skill, that increases the deceptive nature of collection and may produce region-specific content inconsistent with the actual operator environment, undermining integrity and potentially bypassing localization-based controls.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
84% confidence
Finding

The collector saves merged page content to markdown files and later writes CSV/JSON artifacts for collected pages and issues. Although file output is central to collection, this file does not include any visible warning, print/log statement, or explanatory comment indicating that remote content and metadata will be persisted under the artifacts directory.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The collector adaptively changes behavior when scraping fails: it retries, disables stealth selectively, and even launches a visible browser specifically for OpenAI pricing capture. This shows purpose-built evasion and resilience against access controls rather than simple page fetching, making misuse or policy-violating scraping more likely.

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/sync_feishu_sheet.py (reported line 86)May include surrounding context.

python
raise RuntimeError("Unable to locate lark-cli in PATH.")

    def _run(self, args: list[str], input_text: str | None = None) -> dict[str, Any]:
        result = subprocess.run(
            [self.cli_command, *args],
            input=input_text,
            capture_output=True,

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The code deletes existing managed sheets via a batch update request, which is an irreversible remote operation affecting user data. While there are status prints after replacement, there is no confirmation prompt, pre-action warning, or explanatory comment/docstring at the deletion point describing that existing sheets will be removed.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
82% confidence
Finding

The function writes local CSV rows to remote Feishu/Lark sheets using API calls, which both transmits local data over the network and changes remote user data. Although there are progress prints, they do not disclose the privacy/integrity impact or warn that synchronization will push local data to the external service.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
86% confidence
Finding

In code files, network transmission of user or system data should have some visible disclosure unless it is clearly surfaced to the user. This script calls run_sync to send data to Feishu, but this file provides no confirmation prompt, warning message, or explicit user-facing notice at the point of execution.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
84% confidence
Finding

This config mixes English and Chinese product names, and several entries are fixed to Chinese labels such as 阿里云百炼, 百度千帆, 火山方舟, 腾讯混元, and 智谱AI开放平台. Because this file applies those locale-specific names unconditionally and contains no indication of user choice or justification, it may violate a language/locale policy requiring opt-in or documented regional constraints.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
82% confidence
Finding

Line L20 uses a Chinese plan name variant ('免费版') within an otherwise largely English-language catalog for the global OpenAI entry. This can create an implicit locale inconsistency or force a specific language presentation without any documented user choice or regional justification in the file.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
78% confidence
Finding

This file includes vendor_name values in Chinese script for some entries while other entries are in English, but the file provides no natural-language indication that locale selection is intentional or user-configurable. Under the language/locale policy rule, forcing or embedding a specific language without opt-in can create inconsistent user-facing behavior.

Content

No source excerpt is available for this finding.

Unpinned Dependencies

Low
Category
Supply Chain
Confidence
90% confidence
Finding

The dependency specification uses a lower-bound only constraint (playwright>=1.40.0), which allows installation of any newer major or minor release. This can introduce supply-chain and integrity risk because future versions may contain breaking changes or newly introduced malicious/compromised code, reducing build reproducibility and making security review harder.

Content

Scanner excerpt · requirements.txt (reported line 4)May include surrounding context.

text
# Model Pricing Sync - Python Dependencies
# Minimum Python version: 3.10+

playwright>=1.40.0

# Installation steps:
# 1. pip install -r requirements.txt