Back to skill

Security audit

Ow

Security checks for vulnerabilities and agentic risk

Overview

This procurement skill mostly matches its stated purpose, but it needs Review because unsafe bid handling, file storage, endpoint configuration, and payment-link state create real user and business risk.

Install only after review if you are comfortable with procurement details leaving the local machine and being stored locally. Before real use, constrain OW_API_URL to a trusted HTTPS endpoint, add confirmation before publication, validate bid/request identifiers, verify bidder identity and evidence independently, remove payment account/link fields from state, and pin any optional credit-system dependency.

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 (5)

T09 · Insecure Skill Coding Practices

Error
Location
scripts/receive.py:57
Finding

Path Traversal Enables Arbitrary JSON File Creation and Overwrite

Content
View full analysis
Dict[str, Any]: bids_dir = STATE_DIR / "bids" / req_id if not bids_dir.exists(): return {"error": f"Request {req_id} has no bid directory"} bids = [] for bid_file in bids_dir.glob("*.json"): bid = json.loads(bid_file.read_text()) bids.append(bid) result_file = STATE_DIR / "requirements" / f"{req_id}_result.json" result_file.parent.mkdir(parents=True, exist_ok=True) result_file.write_text(json.dumps(result, indent=2, ensure_ascii=False)) ``` Notification paths are constructed similarly: ```python notif_file = NOTIFICATIONS_DIR / f"{req_id}_{notification_type}.json" notif_file.write_text(json.dumps(notification, indent=2, ensure_ascii=False)) ``` ### Technical Analysis The values of `req_id` and `bid_id` originate in externally supplied bid JSON. They are used directly as filesystem path components without a strict identifier format, rejection of path separators, canonicalization, or verification that the resolved destination remains inside the intended state directory. `pathlib.Path` does not automatically prevent traversal. Values containing `../` can escape the intended directory. An absolute path used as a later path component can also replace the preceding bas ...[truncated 1587 chars]
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Error
Location
scripts/receive.py:27
Finding

Unsigned and Self-Asserted Bid Data Can Manipulate Winner Selection

Content
View full analysis
tuple[bool, str]: required = ["bid_id", "req_id", "supplier", "price", "delivery"] for field in required: if field not in bid: return False, f"Missing required field: {field}" if "name" not in bid.get("supplier", {}): return False, "Missing supplier name" if "amount" not in bid.get("price", {}): return False, "Missing bid price" if "time_days" not in bid.get("delivery", {}): return False, "Missing delivery time" return True, "Validation passed" ``` Authenticity points are awarded based only on bidder-supplied document types: ```python scores = {} for bid in bids: bid_id = bid["bid_id"] total = 0 for doc in bid.get("auth_docs", []): doc_type = doc.get("type", "other") total += doc_values.get(doc_type, 5) total = min(total, 100) scores[bid_id] = total * 0.2 ``` Reputation is likewise accepted from the bid itself: ```python rep = bid.get("reputation", {}) tx_score = min(rep.get("total_transactions", 0) * 2, 50) success_rate = rep.get("success_rate", 0.5) rate_score = success_rate * 30 dispute_penalty = rep.get("disputes", 0) * 5 complaint_penalty = rep.get("complaints", 0) * 10 verified_bonus = 10 if rep.get("platform_verified", False) else 0 total = max( 0, tx_score + rate_score - dispute_penalty - complaint_penalty + verified_bonus ) total = min(total, 100) scores[bid_id] = total * 0.1 ``` ### Technical Analysis The documented bid format requires a supplier identity, signature, timestamp, and association with an existing procurement request. The implemented validator does not verify any signature, ...[truncated 2056 chars]
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
scripts/publish.py:19
Finding

Configurable API URL Bypasses the Declared HTTPS and Endpoint Boundary

Content
View full analysis
Remediation
View remediation

T08 · Insecure Dependencies

Warning
Location
scripts/evaluate.py:17
Finding

Runtime Messages Recommend an Unpinned Third-Party Skill Installation

Content
View full analysis
str: return ( "Credit system is not installed and credit information " "cannot be displayed.\n" "Install: npx skills add Enze-dai/ow-skills/ow-credit" ) ``` The actionable command presented to users is: ```bash npx skills add Enze-dai/ow-skills/ow-credit ``` ### Technical Analysis The Skill recommends installing a third-party Skill from a mutable repository reference. The command contains no version, immutable commit identifier, checksum, signature, or integrity policy. Although the current project does not automatically execute this command, presenting it as the normal resolution for a missing optional component encourages users or agents to install code that may change after this audit. A compromised upstream account, repository, package manager path, or transitive dependency could therefore introduce arbitrary code under the user's privileges. ### Attack Path 1. The upstream repository, owner account, package-manager path, or a transitive dependency becomes compromised. 2. Malicious code is published at the same mutable repository reference. 3. A user encounters the missing-credit-system message. 4. The user follows the displayed `npx skills add` command. 5. The external component is downloaded and installed without an audited version or integrity check. 6. The installed component executes with the permissions available to the Skill framework or user. ### Impact Assessment The direct project code does not presently execute a remote payload. Exploitation requires the user ...[truncated 249 chars]
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
state/transactions/TX-20260330-001.json:14
Finding

Packaged State Stores Payment Accounts and Direct Payment Links Contrary to Security Declarations

Content
View full analysis
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Taint TrackingDirect Taint Flow, Variable-Mediated Taint Flow, Credential Exfiltration Chain
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Findings (36)

Tainted flow: 'req' from os.environ.get (line 140, credential/environment) → urllib.request.urlopen (network output)

Critical
Category
Data Flow
Confidence
97% confidence
Finding

The OW API endpoint is taken from the OW_API_URL environment variable and used directly in urllib.request.urlopen without validation or allowlisting. In an agent or automation environment, a malicious or compromised configuration can redirect requests and the posted content/metadata to an attacker-controlled host or internal service, creating SSRF and data exfiltration risk.

Content

Scanner excerpt · scripts/multi-publish.py (reported line 147)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"):
                return {"success": True, "platform": "OW社区", "post_id": result.get("post_id")}

Tainted flow: 'req' from os.environ.get (line 70, credential/environment) → urllib.request.urlopen (network output)

Critical
Category
Data Flow
Confidence
96% confidence
Finding

The destination URL for the outbound POST is taken directly from the OW_API_URL environment variable and then used in urllib.request.urlopen without validation or allowlisting. An attacker who can influence the runtime environment can redirect procurement data to an arbitrary host, causing unauthorized exfiltration of business details and potentially enabling SSRF against internal network services if this runs in a privileged environment.

Content

Scanner excerpt · scripts/publish.py (reported line 77)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'):
                requirement["published"] = True

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The trigger list includes broad phrases like procurement-related everyday terms, which can cause accidental activation on normal user messages. In a skill that can publish requests globally and store local transaction data, overbroad activation increases the chance of unintended external posting or data handling without sufficiently explicit user intent.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill claims HTTPS-only communication but later instructs use of plain HTTP endpoints on localhost. Even on loopback, unauthenticated or unencrypted local APIs can be abused by co-resident malware, proxies, container boundary mistakes, or misconfiguration, and the contradiction may cause operators to overtrust the transport security guarantees.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
89% confidence
Finding

The skill explicitly directs posting procurement content, agent identifiers, and related data to an external/local API endpoint. Because the skill also supports broad publishing workflows and local state storage, unintended or overly permissive transmission could expose sensitive procurement details, contact information, or agent metadata to services that are unauthenticated or not sufficiently trusted.

Content

Scanner excerpt · SKILL.md (reported line 396)May include surrounding context.

发布采购需求:

bash
curl -X POST http://localhost:3000/api/posts \
  -H "Content-Type: application/json" \
  -d '{
    "agent_id": "agent-xxx",

External Transmission

Medium
Category
Data Exfiltration
Confidence
60% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · patterns/bid-format.md (reported line 134)May include surrounding context.

或通过 HTTP 直接提交:

bash
curl -X POST http://<buyer-endpoint>/procurement/bid \
  -H "Content-Type: application/json" \
  -d '<bid_json>'

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The skill explicitly instructs publishing procurement requests to third-party social platforms but does not prominently warn that the content becomes public, may be retained by platforms, and may expose business intent, budget, contact preferences, or other sensitive procurement details. In this context, the omission can lead users to overshare operational or personal data across multiple external services, increasing privacy, profiling, and social-engineering risk.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The skill explicitly instructs buyers to click seller-provided external shop links and complete payment off-platform, but it does not warn users that they are leaving the current system or advise them to verify the destination URL and seller identity. This creates a realistic phishing and social-engineering risk: a malicious seller could provide a spoofed marketplace link or deceptive independent-site URL and induce users to transact on an unsafe site.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

This markdown file is entirely written as a Chinese-language scoring specification and output template, including the mandated '评标输出格式' example. There is no indication that Chinese is optional, user-selected, or justified as a region-specific requirement, which makes it a natural-language locale policy violation under the stated rule.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
86% confidence
Finding

The docstring states seller credit display is optional, but the code still persists seller-related reputation and identity-linked fields such as supplier_id, reputation, shop_links, and media/auth data into the saved evaluation result. This can create an unexpected data exposure surface, especially if users assume optional credit functionality means such seller profiling data will not be retained.

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.

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
94% confidence
Finding

The script automatically sends a buyer notification after evaluation by default (notify_buyer=True) with no explicit confirmation, warning, or dry-run behavior. In an agent setting, this can cause unintended outbound actions, leak procurement results prematurely, or trigger downstream workflows based on unreviewed data.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The module docstring states that procurement requests will be synchronously published to multiple platforms and emphasizes an implementation using urllib without external CLI dependencies. In reality, only the OW platform performs a real HTTP POST; other platforms go through publish_to_social(), which merely returns an action of 'manual_publish' plus an instruction string rather than publishing anything.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The script automatically publishes procurement content, buyer identity, budget, and deadline to an external service without any explicit confirmation step or clear warning at the point of transmission. This increases the risk of accidental disclosure of sensitive commercial information, especially because the output emphasizes successful global publication after the fact rather than obtaining informed consent before sending.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The module docstring and all user-facing CLI messages are written only in Chinese, which imposes a specific language/locale on users without offering a choice. The file does not document that the skill is intended only for a Chinese-language or region-specific context, so this is a natural-language locale policy concern.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

This JSON manifest embeds supplier name, contact method, license name, currency, return/refund text, and delivery method entirely in Chinese, indicating a fixed Chinese-language/China-locale workflow. There is no natural-language indication of user opt-in for this locale or any documented justification that the skill is region-specific, which can violate language/locale policy requirements.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

This JSON manifest contains multiple user-facing values entirely in Chinese, including supplier name, document names, delivery method, and warranty terms. The file provides no indication that the skill is region-specific or that users can opt into this locale, which matches the language/locale policy violation criteria.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

This JSON result embeds supplier names, document names, and platform labels entirely in Chinese, indicating a fixed locale in the skill output data. The file provides no indication that the user opted into this language or that the skill is region-specific, which can violate organizational language/locale policy for natural-language content.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
74% confidence
Finding

The trigger list is provided only in Chinese terms, which can create a language/locale constraint for activation behavior. Although the overview is bilingual, the activation policy itself does not clearly offer equivalent English trigger options or user opt-in for language selection.

Content

No source excerpt is available for this finding.

Vague Triggers

Low
Category
Not specified by scanner
Confidence
82% confidence
Finding

The example "Help me procure" / "帮我采购" is a natural everyday request and the document does not clearly distinguish this skill's invocation boundary from general shopping assistance. Without exclusion conditions or a constrained trigger list, the activation scope remains ambiguous.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Low
Category
Not specified by scanner
Confidence
98% confidence
Finding

Most of the document describes bidding evaluation as a 5-dimension system with weights for Price, Authenticity, Media, Delivery, and History. Line L462 instead states '四维度综合评分' (four-dimension composite scoring), which directly contradicts the documented behavior and could mislead users about how bids are ranked.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
78% confidence
Finding

The document prescribes Chinese-only copy for some platforms and English-only copy for others such as Twitter/X, Facebook, and Google. It does not indicate that the user can choose the output language or locale, which can be a language-policy concern when user preference is not considered.

Content

No source excerpt is available for this finding.