Back to skill

Security audit

Ows

Security checks for vulnerabilities and agentic risk

Overview

This seller automation skill is mostly purpose-aligned, but it can post bids externally and persist remote data with weak confirmation, endpoint, and file-containment controls.

Install only if you are comfortable reviewing each bid before it is sent, keeping auto-bid disabled, constraining OW_API_URL to a trusted HTTPS endpoint, and running the skill with filesystem access limited to its own state directory. Treat buyer messages as untrusted, and do not run the optional npx install command unless you separately verify and pin that dependency.

Vulnerability Patterns
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • 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 (5)

T01 · Skill Instruction Hijacking

Error
Location
scripts/opportunity_notify.py:72
Finding

Untrusted remote buyer content is embedded directly into agent-facing operational notifications

Content
View full analysis
str: score = opp.get('match_score', 0) score_desc = "高度匹配" if score >= 0.8 else "匹配" if score >= 0.5 else "可能匹配" msg = f""" 🎯 发现新商机 | New Business Opportunity ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 📋 需求ID: {opp.get('post_id')} 👤 买家: {opp.get('buyer_name', '未知')} 📍 区域: {opp.get('buyer_region', '未知')} 📦 匹配产品: {opp.get('product')} 📊 匹配度: {score:.0%} ({score_desc}) ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 📝 需求内容: {opp.get('content', '')[:300]}... ✅ 发货状态: {opp.get('ship_status', '可发货')} 🔑 匹配关键词: {', '.join(opp.get('matched_keywords', []))} """ ``` ### Technical Analysis Buyer posts are retrieved from an external API and their `content` property is copied directly into an agent-facing notification. Truncation limits length but does not establish a security boundary. The code does not quote or encode the content as untrusted dat ...[truncated 1740 chars]
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Error
Location
scripts/search.py:85
Finding

Attacker-controlled remote request identifiers allow path traversal and arbitrary JSON state overwrite

Content
View full analysis
Remediation
View remediation
str: if not isinstance(value, str) or not ID_PATTERN.fullmatch(value): raise ValueError("Invalid identifier") return value ``` 2. Generate local UUID filenames instead of trusting platform identifiers. 3. Resolve and verify every output path before use: ```python base = OPPS_DIR.resolve() target = (base / f"{validated_id}.json").resolve() if not target.is_relative_to(base): raise ValueError("Path escapes opportunity directory") ``` 4. Use atomic file replacement to prevent partial state corruption. 5. Keep remote opportunity records separate from trusted configuration files. 6. Validate the JSON schema and expected data types before persistence. 7. Run the Skill with filesystem permissions limited to its own state directory. ]]>

T09 · Insecure Skill Coding Practices

Warning
Location
scripts/media.py:76
Finding

Unsanitized command-line identifiers permit unintended file reads, writes, and media placement

Content
View full analysis
Dict: opp_file = OPPORTUNITIES_DIR / f"{post_id}.json" if not opp_file.exists(): return {"error": f"未找到商机: {post_id}"} opportunity = json.loads(opp_file.read_text()) catalog = json.loads(CATALOG_FILE.read_text()) ``` ```python # scripts/orders.py def load_order(order_id): order_file = ORDERS_DIR / f"{order_id}.json" if order_file.exists(): return json.loads(order_file.read_text()) return None def create_order_from_bid(bid_id, buyer_info): bid_file = BIDS_DIR / f"{bid_id}.json" if not bid_file.exists(): return {"error": f"未找到投标 {bid_id}"} bid = json.loads(bid_file.read_text()) ``` ```python # scripts/media.py bid_image_dir = IMAGES_DIR / bid_id bid_image_dir.mkdir(parents=True, exist_ok=True) for i, img_path in enumerate(images, 1): valid, msg = validate_image(img_path) if valid: ext = pathlib.Path(img_path).suffix dest = bid_image_dir / f"image_{i}{ext}" import shutil shutil.copy2(img_path, dest) if video: valid, msg = validate_video(video) if valid: bid_video_dir = VIDEOS_DIR / bid_id bid_video_dir.mkdir(parents=True, exist_ok=True) ext = pathlib.Path(video).suffix dest = bid_video_dir / f"video{ext}" import shutil shutil.copy2(vide ...[truncated 2115 chars]
Remediation
View remediation

T08 · Insecure Dependencies

Warning
Location
scripts/auto_match.py:41
Finding

The Skill recommends installation of an unpinned third-party component through npx

Content
View full analysis
str: return "📊 信用系统未安装\n 安装:npx skills add Enze-dai/ow-skills/ow-credit" ``` The effective command recommended by the Skill is: ```bash npx skills add Enze-dai/ow-skills/ow-credit ``` ### Technical Analysis The installation reference identifies a repository and subdirectory but does not pin an immutable commit, signed release, package version, or content digest. The command is presented from normal runtime output, which encourages users or agents to install mutable external content after the current Skill has been reviewed. No evidence was found that the present project automatically executes this command. The risk arises when a user or agent follows the recommendation. At that point, the security properties of the environment depend on the current upstream repository and the `npx`-retrieved installation tooling. ### Attack Path 1. The optional credit system is unavailable. 2. The Skill displays the installation command. 3. A user or capable agent executes the command. 4. `npx` obtains tooling and installs content from a mutable external source. 5. If the upstream repository, account, dependency, tag, or distribution path has been compromised, malicious Skill instructions or scripts enter the local environment. 6. The newly installed component executes with the privileges granted to the agent or user. ### Impact Assessment The current repository does not itself retrieve and execute this payload automatically. However, following the recommendation can introduce arbitrary future code or instructions outside the scope of the present audit. The potential impact of a compromised dependency includes: - Execution of malicious installation or runtime ...[truncated 218 chars]
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
scripts/opportunity_notify.py:226
Finding

Declared confirmation and HTTPS guarantees do not match implemented auto-bid and endpoint behavior

Content
View full analysis
Dict: config = catalog.get('auto_match', {}) if not config.get('auto_bid_enabled', False): return {"skipped": True, "reason": "自动投标未开启"} min_score = config.get('auto_bid_min_score', 0.8) if opportunity.get('match_score', 0) < min_score: return {"skipped": True, "reason": f"匹配度不足({opportunity.get('match_score'):.0%}<{min_score:.0%})"} return submit_bid(opportunity, catalog) ``` ```python # scripts/search.py "ow": { "base_url": "http://localhost:3000/api", "auth_required": False, "endpoint": "/posts?type=request" } ``` ```python def search_ow(base_url="http://localhost:3000", limit=50): url = f"{base_url}{PLATFORMS['ow']['endpoint']}&limit={limit}" try: response = requests.get(url, timeout=10) ``` ### Technical Analysis The metadata claims that auto-bidding requires confirmation. The implementation instead performs an immediate submission when `auto_bid_enabled` is true and the score threshold is met. There is no transaction-specific approval check, confirmation token, daily-limit enforcement, or human authorization in `process_auto_bid()`. The metadata also claims HTTPS-only ...[truncated 1843 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
  • Behavioral ASTexec() Call, eval() Call, Dynamic Import
  • Taint TrackingDirect Taint Flow, Variable-Mediated Taint Flow, Credential Exfiltration Chain
Findings (55)

Tainted flow: 'OW_API' from os.environ.get (line 33, credential/environment) → urllib.request.urlopen (network output)

Critical
Category
Data Flow
Confidence
95% confidence
Finding

The request URL is built from the OW_API_URL environment variable and passed directly to urllib.request.urlopen without validation. An attacker who can influence the environment can redirect the script to arbitrary internal or external endpoints, enabling SSRF-style behavior, data exfiltration to attacker-controlled services, or use of insecure schemes if supported by the runtime.

Content

Scanner excerpt · scripts/auto_match.py (reported line 197)May include surrounding context.

python
def search_requests():
    """搜索求购信息"""
    try:
        with urllib.request.urlopen(OW_API, timeout=30) as response:
            data = json.loads(response.read().decode('utf-8'))
        if data.get('success'):
            return data.get('posts', [])

Tainted flow: 'url' from os.environ.get (line 79, credential/environment) → urllib.request.urlopen (network output)

Critical
Category
Data Flow
Confidence
97% confidence
Finding

The script builds the request target from the untrusted OW_API_URL environment variable and immediately fetches it with urllib.request.urlopen. In an agent or CI/runtime environment, a manipulated environment variable can redirect requests to attacker-controlled or internal endpoints, enabling SSRF, data exfiltration, or unintended interaction with privileged network resources.

Content

Scanner excerpt · scripts/multi-platform.py (reported line 80)May include surrounding context.

python
"""搜索OW社区的求购信息(使用 urllib,无需 curl)"""
    try:
        url = os.environ.get("OW_API_URL", "https://www.owshanghai.com/api/posts") + "?type=request&limit=50"
        with urllib.request.urlopen(url, timeout=30) as response:
            data = json.loads(response.read().decode('utf-8'))
        
        if data.get("success"):

Tainted flow: 'req' from os.environ.get (line 207, credential/environment) → urllib.request.urlopen (network output)

Critical
Category
Data Flow
Confidence
98% confidence
Finding

The publish request sends product and agent data to a URL derived from the untrusted OW_API_URL environment variable. If that variable is altered, the script can post sensitive or business data to an attacker-controlled server or internal service, creating SSRF and unintended data disclosure risks.

Content

Scanner excerpt · scripts/multi-platform.py (reported line 214)May include surrounding context.

python
method="POST"
        )
        
        with urllib.request.urlopen(req, timeout=30) as response:
            result = json.loads(response.read().decode('utf-8'))
        
        if result.get("success"):

Tainted flow: 'req' from os.environ.get (line 203, credential/environment) → urllib.request.urlopen (network output)

Critical
Category
Data Flow
Confidence
97% confidence
Finding

The request destination is derived from the OW_API_URL environment variable and then used for a POST via urllib.request.urlopen without validation or allowlisting. An attacker who can influence the runtime environment can redirect bid submissions and seller/business data to an arbitrary host, causing data exfiltration, SSRF-style internal access, or silent submission to a rogue API endpoint.

Content

Scanner excerpt · scripts/opportunity_notify.py (reported line 210)May include surrounding context.

python
method='POST'
        )
        
        with urllib.request.urlopen(req, timeout=30) as response:
            result = json.loads(response.read().decode('utf-8'))
        
        if result.get('success'):

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The trigger list includes very broad commerce terms such as '卖', '出售', '供货', and '订单', which are likely to appear in benign conversations. Overbroad activation can cause unintended invocation of networked search, bid preparation, or publishing behaviors, especially in a skill that interacts with external platforms and stores local state.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The skill describes deriving buyer region from buyer IP addresses but does not present a clear privacy notice, consent model, retention policy, or accuracy limitations. IP-based geolocation is personal data in many contexts and can create privacy, compliance, and profiling risks when used for automated filtering and decision-making.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The skill promotes multi-platform publishing and inclusion of external shop links and seller details, but does not clearly warn that this data may be propagated to multiple third-party services with different privacy and abuse risks. This can expose seller identity, contact details, shop metadata, and bid information more broadly than expected.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill explicitly claims 'HTTPS only' while later documenting use of plain HTTP endpoints for the primary OW API. Even if the endpoint is localhost, this is still a misleading security guarantee and can enable insecure deployment patterns, unsafe proxying, or unauthenticated local service exposure that operators may wrongly assume is protected.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The metadata states that auto-bid requires explicit confirmation, but the later configuration example enables fully automatic bidding without a confirmation step. This inconsistency can lead operators to enable autonomous external actions they believed were gated, causing unintended submissions, spam, or commitments the seller cannot fulfill.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
86% confidence
Finding

The documented curl commands send and retrieve data from an external/local API, including agent identifiers and bid content. External transmission is expected for this skill's purpose, but it still represents a real security boundary because it can disclose business data, create side effects on remote services, and interact with potentially unauthenticated endpoints.

Content

Scanner excerpt · SKILL.md (reported line 935)May include surrounding context.

搜索买家需求:

bash
curl "http://localhost:3000/api/posts?type=request"

提交投标:

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The workflow sends order and shipping data to external parties and APIs, including buyer agent IDs, delivery-related details, tracking numbers, and carrier information, but the skill provides no notice, consent boundary, minimization guidance, or security requirements for that data sharing. This creates a real privacy and compliance risk because operators may implement the flow in a way that exposes personal or transactional data to third-party logistics services or counterparties without adequate user awareness or safeguards.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The payment section describes integrating with payment APIs and callbacks and storing transaction details, but it omits warnings and controls around handling sensitive financial data, validating payment notifications, and protecting transaction records. In practice, this can lead to unsafe implementations such as trusting spoofed callbacks, over-collecting payment data, or exposing financial information, which raises fraud, privacy, and compliance risk.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The file explicitly enables automatic cross-platform searching on a timer across multiple external services, but provides no user-facing warning, consent flow, or disclosure that ongoing network requests and possible query/category data will be transmitted. In a skill context, this can cause unexpected background communication, privacy leakage, and policy/compliance issues even if the behavior is not overtly malicious.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The script collects buyer IP addresses, inferred regions, names, IDs, and message content, prints some of that data, and persists portions of it into the catalog state without minimization or retention controls. In a real deployment this can create privacy and compliance risk, especially if logs or state files are broadly accessible or retained longer than necessary.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The inline comment at L355 frames this section as only creating an opportunity notification for the seller bot. In reality, the block also checks auto_bid_enabled and may call process_auto_bid, which is a materially different action than notification and can trigger downstream operations. That makes the comment actively misleading about the effect of this code path.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The file’s user-facing natural-language description is entirely in Chinese and presents the skill as operating in that language by default. Under the stated policy, forcing a specific language without opt-in can be a locale-policy violation when no alternative or justification is provided.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The file's docstrings, prompts, and user-facing output are entirely in Chinese, which imposes a specific language on all users. The policy allows locale constraints only when they are explicitly justified or when the user is offered a language choice, neither of which appears here.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The module description and all user-facing strings are written exclusively in Chinese, with no indication that other languages are supported or that the user can choose a locale. The policy explicitly calls out language or locale constraints as violations when they are forced without user opt-in.

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/media.py (reported line 62)May include surrounding context.

python
# 检查时长(需要 ffprobe)
    try:
        import subprocess
        result = subprocess.run(
            ['ffprobe', '-v', 'error', '-show_entries', 'format=duration', 
             '-of', 'default=noprint_wrappers=1:nokey=1', str(video_path)],
            capture_output=True, text=True, timeout=10

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The file’s docstring and most user-facing output and instruction strings are written only in Chinese, including operational guidance for search and publish flows. This enforces a specific language experience without any opt-in, fallback, or documented justification for a Chinese-only locale.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The function sends seller-provided product details and agent identifiers to the OW community API via an HTTP POST request. Although the script prints status messages before publishing, it does not provide a confirmation prompt or explicit warning that data will be transmitted to an external service.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
70% confidence
Finding

npx commands without a version suffix (e.g. @1.0.0) create a rug-pull risk if the upstream server is compromised and publishes a malicious update.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
70% confidence
Finding

npx commands without a version suffix (e.g. @1.0.0) create a rug-pull risk if the upstream server is compromised and publishes a malicious update.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The auto-bid path can submit network bids automatically once auto_bid_enabled is set, and the execution path in the 'auto' action performs submissions without an inline confirmation step or fresh warning at the moment of action. Because bid submission transmits seller identity, pricing, logistics, and shop links to external parties, accidental or misconfigured execution can create unauthorized commitments and data disclosure.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The file description and all user-facing strings are written in Chinese, which forces a specific language for users without any opt-in or alternative. Under the policy, locale or language restrictions should either be optional for the user or clearly justified as region-specific.

Content

No source excerpt is available for this finding.