The Agent Skills Directory

[COMMAND_EXECUTION]: The workflow scripts utilize curl to interact with external web servers. User-defined inputs such as TARGET_URL, AUTH_COOKIE, and ATTACKER_DOMAIN are interpolated directly into shell commands within SKILL.md to perform the vulnerability assessment.
[DATA_EXFILTRATION]: To identify web cache deception, the skill transmits a user-provided session cookie (AUTH_COOKIE) to the TARGET_URL. This is an intended function of the diagnostic test to check if authenticated data is incorrectly cached by the server.
[PROMPT_INJECTION]: The skill ingests untrusted data from the TARGET_URL response in Step 5, which could be exploited for indirect prompt injection if the target server is malicious.
Ingestion points: The content of the HTTP response is retrieved via curl and assigned to the resp variable in SKILL.md for inspection.
Boundary markers: There are no boundary markers or instructions to the agent to ignore potentially malicious embedded content within the server response.
Capability inventory: The skill possesses networking capabilities (curl) and text processing tools (grep, sed, cut).
Sanitization: The response content is not sanitized or escaped before being echoed and processed by the agent's logic.

web-vuln-cache-poisoning