The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from both user-provided inputs and external network responses, which are then used to drive the agent's logic and shell execution. This creates a vulnerability surface for indirect prompt injection.
Ingestion points: The TARGET_URL and LEGITIMATE_ORIGIN input variables in SKILL.md, as well as the HTTP headers (Access-Control-Allow-Origin, Access-Control-Allow-Credentials) retrieved from external servers during the test workflow.
Boundary markers: None. There are no delimiters or specific instructions to the agent to isolate or ignore potentially malicious content within these inputs.
Capability inventory: The skill uses curl, grep, and bash script interpolation to perform network tests and assess security risks.
Sanitization: No sanitization, validation, or escaping is performed on the input variables or the external HTTP response data before they are processed or used in shell commands.
[COMMAND_EXECUTION]: The skill relies on the execution of shell commands (primarily curl) to perform its core security auditing tasks. While this is the primary purpose of the skill, the lack of input validation for the TARGET_URL and LEGITIMATE_ORIGIN variables could potentially lead to unintended command execution if the inputs contain shell metacharacters.

cors-check