The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The file EXAMPLES.md contains a reference to https://bit.ly/example, which has been flagged by security scanners as a known malicious URL associated with botnet activity. While used as an example for redirect following, its inclusion poses a risk to users who might interact with the link.
[DATA_EXFILTRATION]: The skill provides an interface for Server-Side Request Forgery (SSRF) through the fetch API. Specifically, the allowInsecureSsl: true parameter documented in SKILL.md and REFERENCE.md allows bypassing TLS verification. This can be exploited to access internal, self-signed services (e.g., staging.internal.example.com) or sensitive metadata endpoints, potentially exposing internal data.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it ingests untrusted data from external URLs and returns it to the agent context.
Ingestion points: Web content is retrieved via the fetch tool in SKILL.md and EXAMPLES.md (e.g., response.content).
Boundary markers: No explicit delimiters or instructions are provided to the agent to treat the fetched content as potentially adversarial or to ignore embedded instructions.
Capability inventory: The skill returns the full page body, headers, and metadata. While it does not execute code locally, the content can manipulate the agent's behavior in subsequent steps.
Sanitization: There is no evidence of sanitization or filtering of the content field before it is processed by the agent.

fetch