Back to skill

Security audit

Agent Commerce Discovery

Security checks for vulnerabilities and agentic risk

Overview

This is a non-executing commerce-discovery guide, but it asks for sensitive credentials beyond what its examples need and includes copyable production patterns with credential and file-write risks.

Review before installing. Treat this as a guide, not a ready-to-run integration. Provide only narrowly scoped credentials you actually need, avoid supplying Stripe or agent signing keys unless a specific operation requires them, pin the GreenHelix API origin to trusted hosts, and validate webhook service identifiers before copying the file-write or marketplace-update examples into production.

Vulnerability Patterns
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
  • 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 (3)

T09 · Insecure Skill Coding Practices

Error
Location
SKILL.md:122
Finding

Configurable API Origin Can Exfiltrate the GreenHelix Bearer Credential

Content
View full analysis
dict: """Execute a single tool on the gateway.""" resp = self.session.post( f"{GATEWAY_URL}/v1", json={"tool": tool, "input": input_data}, timeout=30, ) ``` ### Technical Analysis The destination of every authenticated request is controlled by the `GREENHELIX_API_URL` environment variable. The client places the GreenHelix API key in the session-wide `Authorization` header and then sends that header to the configured origin without validating its scheme or hostname. Environment variables may be modified by deployment configuration, parent processes, compromised build pipelines, container configuration, or another component sharing the runtime environment. If `GREENHELIX_API_URL` points to an attacker-controlled origin, the example transmits the bearer credential directly to that origin. A session-wide authorization header also creates redirect-related risk. Unless redirects are explicitly disabled or constrained, an unexpected redirect could cause credentials or authenticated request context to reach an unintended endpoint, depending on the HTTP client’s redirect behavior and destination relationship. This outbound authentication is necessary for authenticated GreenHelix operations, but allowing an a ...[truncated 1163 chars]
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Error
Location
SKILL.md:1473
Finding

Webhook-Supplied Service Identifier Reaches a Filesystem Path Without Validation

Content
View full analysis
None: """Write updated JSON-LD to file (or key-value store in production).""" target = next( (j for j in jsonld if service_id in j.get("@id", "")), None, ) if target: with open(f"/var/data/schemas/{service_id}.jsonld", "w") as f: json.dump(target, f, indent=2) ``` The identifier can originate from a network webhook: ```python @app.post("/webhooks/catalog") async def receive_catalog_webhook(request: Request): """Handle incoming catalog update webhooks from GreenHelix.""" body = await request.body() signature = request.headers.get("X-GreenHelix-Signature", "") if not sync_manager.verify_webhook_signature(body, signature): return JSONResponse(status_code=401, content={"error": "Invalid signature"}) payload = json.loads(body) event_type = payload.get("event_type", "") service_id = payload.get("service_id", "") changes = payload.get("changes", {}) # Update our local graph with the external service's changes pipeline._on_catalog_update(event_type, service_id, changes) return {"status": "accepted"} ``` ### Technical Analysis The webhook handler extracts `service_id` from JSON and passes it into the update pipeline. The pipeline eventually interpolates this value directly into an absolute filesystem path: ```python f"/var/data/schemas/{service_id}.jsonld" ``` There is no allowlist, character restriction, basename conversion, canonical-path check, or containment verification. A value containing path separators and traversal components can therefore resolve outside `/var/data/schemas`. Webhook HMAC validation confirms that a party possessing the webhook secret created the payload, bu ...[truncated 2137 chars]
Remediation
View remediation

T05 · Unauthorized Access and Privilege Escalation

Warning
Location
SKILL.md:14
Finding

Skill Metadata Requests Unused Signing and Payment Credentials

Content
View full analysis
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (10)

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The guide states that no API key is required to get started, yet the primary client example immediately initializes bearer-token authentication and repeatedly uses authenticated gateway calls. This mismatch can mislead users into supplying real credentials earlier than intended or trusting that examples are safe to run in a low-risk sandbox when they may trigger authenticated external requests.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
50% 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 · SKILL.md (reported line 39)May include surrounding context.

md
Forty percent of digital commerce services are invisible to AI agents. Not because the services are bad, but because they are structured for humans -- HTML pages, marketing copy, PDF brochures -- and agents cannot parse any of it. The services exist, the demand exists, and the agents have budgets to spend. But the transaction never happens because the agent never finds the service. This is the discovery gap, and it is the single largest source of lost revenue in the agentic economy.
Between January 2025 and March 2026, AI-referred traffic to commerce services grew 805% year-over-year. Google Shopping Graph now indexes over 45 billion product listings with structured attributes. ChatGPT's product search handles 2.3 million commerce queries per day. Perplexity Shopping launched with 30+ retail partners and AI-native product cards. The agents are shopping. The question is whether they can find you.
This guide is the practitioner's manual for making your services discoverable to AI agents. It covers the full stack: knowledge graphs that model your service catalog as machine-readable entities, protocol endpoints that announce your capabilities through UCP, MCP, A2A, and ACP, structured data that agents can parse and compare, real-time catalog synchronization that prevents stale listings, trust signals that help agents rank you above competitors, and analytics that measure whether discovery is actually converting into transactions. Every chapter contains production Python code against the GreenHelix A2A Commerce Gateway -- 128 tools accessible at `https://api.greenhelix.net/v1` via a single the REST API (`POST /v1/{tool}`) endpoint.

## What You'll Learn
- Chapter 1: The Discovery Problem

External Transmission

Medium
Category
Data Exfiltration
Confidence
50% 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 · SKILL.md (reported line 60)May include surrounding context.

md
Forty percent of digital commerce services are invisible to AI agents. Not because the services are bad, but because they are structured for humans -- HTML pages, marketing copy, PDF brochures -- and agents cannot parse any of it. The services exist, the demand exists, and the agents have budgets to spend. But the transaction never happens because the agent never finds the service. This is the discovery gap, and it is the single largest source of lost revenue in the agentic economy.
Between January 2025 and March 2026, AI-referred traffic to commerce services grew 805% year-over-year. Google Shopping Graph now indexes over 45 billion product listings with structured attributes. ChatGPT's product search handles 2.3 million commerce queries per day. Perplexity Shopping launched with 30+ retail partners and AI-native product cards. The agents are shopping. The question is whether they can find you.
This guide is the practitioner's manual for making your services discoverable to AI agents. It covers the full stack: knowledge graphs that model your service catalog as machine-readable entities, protocol endpoints that announce your capabilities through UCP, MCP, A2A, and ACP, structured data that agents can parse and compare, real-time catalog synchronization that prevents stale listings, trust signals that help agents rank you above competitors, and analytics that measure whether discovery is actually converting into transactions. Every chapter contains production Python code against the GreenHelix A2A Commerce Gateway -- 128 tools accessible at `https://api.greenhelix.net/v1` via a single the REST API (`POST /v1/{tool}`) endpoint.

## What You'll Learn
- Chapter 1: The Discovery Problem

External Transmission

Medium
Category
Data Exfiltration
Confidence
50% 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 · SKILL.md (reported line 2273)May include surrounding context.

md
Forty percent of digital commerce services are invisible to AI agents. Not because the services are bad, but because they are structured for humans -- HTML pages, marketing copy, PDF brochures -- and agents cannot parse any of it. The services exist, the demand exists, and the agents have budgets to spend. But the transaction never happens because the agent never finds the service. This is the discovery gap, and it is the single largest source of lost revenue in the agentic economy.
Between January 2025 and March 2026, AI-referred traffic to commerce services grew 805% year-over-year. Google Shopping Graph now indexes over 45 billion product listings with structured attributes. ChatGPT's product search handles 2.3 million commerce queries per day. Perplexity Shopping launched with 30+ retail partners and AI-native product cards. The agents are shopping. The question is whether they can find you.
This guide is the practitioner's manual for making your services discoverable to AI agents. It covers the full stack: knowledge graphs that model your service catalog as machine-readable entities, protocol endpoints that announce your capabilities through UCP, MCP, A2A, and ACP, structured data that agents can parse and compare, real-time catalog synchronization that prevents stale listings, trust signals that help agents rank you above competitors, and analytics that measure whether discovery is actually converting into transactions. Every chapter contains production Python code against the GreenHelix A2A Commerce Gateway -- 128 tools accessible at `https://api.greenhelix.net/v1` via a single the REST API (`POST /v1/{tool}`) endpoint.

## What You'll Learn
- Chapter 1: The Discovery Problem

External Transmission

Medium
Category
Data Exfiltration
Confidence
50% 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 · SKILL.md (reported line 638)May include surrounding context.

md
The `.well-known` URI pattern (RFC 8615) provides a standardized location for machine-readable metadata. Agents crawling for services check these paths before attempting any other discovery method. If your service is at `https://api.example.com`, agents will look for:

- `https://api.example.com/.well-known/ucp` -- UCP service profile
- `https://api.example.com/.well-known/agent.json` -- A2A agent card
- `https://api.example.com/.well-known/ai-plugin.json` -- OpenAI/ChatGPT plugin manifest
- `https://api.example.com/.well-known/openapi.yaml` -- OpenAPI specification

External Transmission

Medium
Category
Data Exfiltration
Confidence
50% 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 · SKILL.md (reported line 639)May include surrounding context.

md
The `.well-known` URI pattern (RFC 8615) provides a standardized location for machine-readable metadata. Agents crawling for services check these paths before attempting any other discovery method. If your service is at `https://api.example.com`, agents will look for:

- `https://api.example.com/.well-known/ucp` -- UCP service profile
- `https://api.example.com/.well-known/agent.json` -- A2A agent card
- `https://api.example.com/.well-known/ai-plugin.json` -- OpenAI/ChatGPT plugin manifest
- `https://api.example.com/.well-known/openapi.yaml` -- OpenAPI specification

External Transmission

Medium
Category
Data Exfiltration
Confidence
50% 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 · SKILL.md (reported line 640)May include surrounding context.

md
The `.well-known` URI pattern (RFC 8615) provides a standardized location for machine-readable metadata. Agents crawling for services check these paths before attempting any other discovery method. If your service is at `https://api.example.com`, agents will look for:

- `https://api.example.com/.well-known/ucp` -- UCP service profile
- `https://api.example.com/.well-known/agent.json` -- A2A agent card
- `https://api.example.com/.well-known/ai-plugin.json` -- OpenAI/ChatGPT plugin manifest
- `https://api.example.com/.well-known/openapi.yaml` -- OpenAPI specification

External Transmission

Medium
Category
Data Exfiltration
Confidence
50% 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 · SKILL.md (reported line 641)May include surrounding context.

md
The `.well-known` URI pattern (RFC 8615) provides a standardized location for machine-readable metadata. Agents crawling for services check these paths before attempting any other discovery method. If your service is at `https://api.example.com`, agents will look for:

- `https://api.example.com/.well-known/ucp` -- UCP service profile
- `https://api.example.com/.well-known/agent.json` -- A2A agent card
- `https://api.example.com/.well-known/ai-plugin.json` -- OpenAI/ChatGPT plugin manifest
- `https://api.example.com/.well-known/openapi.yaml` -- OpenAPI specification

External Transmission

Medium
Category
Data Exfiltration
Confidence
50% 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 · SKILL.md (reported line 1010)May include surrounding context.

md
The `.well-known` URI pattern (RFC 8615) provides a standardized location for machine-readable metadata. Agents crawling for services check these paths before attempting any other discovery method. If your service is at `https://api.example.com`, agents will look for:

- `https://api.example.com/.well-known/ucp` -- UCP service profile
- `https://api.example.com/.well-known/agent.json` -- A2A agent card
- `https://api.example.com/.well-known/ai-plugin.json` -- OpenAI/ChatGPT plugin manifest
- `https://api.example.com/.well-known/openapi.yaml` -- OpenAPI specification

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The JSON-LD example sets "availableLanguage": ["en"], which imposes a language/locale constraint in natural-language-adjacent configuration content. The guide does not present this as optional, user-selectable, or region-specific, so it conflicts with the policy against forcing a specific language without user opt-in.

Content

No source excerpt is available for this finding.