Back to skill

Security audit

Farmos Weather

Security checks for vulnerabilities and agentic risk

Overview

This skill is a straightforward farm weather lookup helper, with a notable caution that it uses an HTTP weather API endpoint.

Install only if you expect the agent to contact the documented farm weather service and you are comfortable with imperial-unit output. The main caution is that the weather API uses HTTP, so farm field references, coordinates, dates, and weather responses could be exposed or modified by someone with access to the network path.

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 (1)

T09 · Insecure Skill Coding Practices

Warning
Location
SKILL.md:23
Finding
Weather API Uses Unencrypted HTTP Transport## Vulnerability Details **File Location**: `SKILL.md`, line 23 **Vulnerability Type**: Plaintext API transport **Risk Level**: Medium **Vulnerable code snippet:** ```markdown ## API Base http://100.102.77.110:8012 ``` ### Technical Analysis The skill configures all weather operations to use an API base URL with plaintext HTTP rather than HTTPS. Application-layer encryption and server authentication are therefore absent. Requests may contain field identifiers, farm coordinates, date ranges, and operational context, while responses may include forecasts and spray-condition evaluations used to make farming decisions. The address belongs to the shared-address range commonly used by private overlay networks. Such an overlay may encrypt traffic at the network layer, but that protection is not established by this project and does not replace authenticated HTTPS at the application layer. A compromised peer, improperly configured route, hostile proxy, or attacker positioned inside the trusted network could potentially observe or modify the traffic. ### Attack Path 1. The agent loads the documented API base URL and constructs a weather request over HTTP. 2. An attacker obtains a network position capable of observing or modifying traffic, such as control of a trusted peer, route, gateway, or proxy. 3. The attacker intercepts farm identifiers, coordinates, dates, or other operational query data. 4. The attacker can modify an HTTP response to report false temperature, precipitation, wind, forecast, or spray-condition information. 5. The agent treats the altered response as service output and may present inaccurate operational guidance to the user. This exploitation path requires access to the relevant network path or trusted overlay environment; the file alone does not grant such access. ### Impact Assessment No local system privileges are directly obtained through this flaw. Its primary effects are: - **Confidentiality ...[truncated 560 chars]
Remediation
## Remediation Suggestions 1. Replace the HTTP endpoint with an HTTPS service hostname: ```markdown ## API Base https://weather.internal.example:8012 ``` 2. Configure a valid certificate issued by a trusted internal or public certificate authority and require normal certificate and hostname validation. 3. Reject TLS certificate errors rather than allowing insecure fallback behavior. 4. Prohibit redirects or automatic downgrade from HTTPS to HTTP. 5. Add authenticated API access where appropriate, using narrowly scoped and securely stored credentials rather than embedding secrets in `SKILL.md`. 6. Apply network access controls so only authorized agents and service hosts can reach the weather API. 7. If a private overlay network is required, retain its encryption but use HTTPS as an additional application-layer control. 8. Document connection timeouts, response validation, and failure behavior so manipulated or malformed responses are not silently treated as trustworthy weather data.
Vulnerability Patterns
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (3)

Vague Triggers

Medium
Confidence
95% confidence
Finding
The trigger phrases include very broad everyday terms such as "forecast" and "what's the weather," which can cause the skill to activate in situations where the user did not specifically intend to invoke this tool. In an agent setting, over-broad activation can lead to unintended data access, unnecessary external API calls, and incorrect workflow branching, especially because the skill also encourages cross-module queries to tasks and observations for some weather-related questions.

Vague Triggers

Medium
Confidence
97% confidence
Finding
Allowing minimum viable input of just "Weather" or any field reference is overly ambiguous and makes accidental invocation likely. This is more dangerous in context because the skill states that any nearby field ID works and all fields are in central Indiana, which lowers precision requirements and increases the chance the agent will fetch weather data for an arbitrary field without clear user intent.

Natural-Language Policy Violations

Low
Confidence
90% confidence
Finding
The instructions require displaying only US imperial units "as-is" and explicitly prohibit conversion or relabeling. This imposes a locale-specific presentation choice without offering the user an option to receive metric units or documenting an explicit user opt-in.