The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill defines a workflow where the agent is instructed to interact with external target servers and capture PoC evidence, including HTTP logs and command output. This creates a surface where a malicious target could return responses containing instructions designed to influence the agent's behavior.
Ingestion points: Network responses and tool outputs captured during the 'Exploitation' and 'Verification' phases described in SKILL.md.
Boundary markers: The instructions do not define specific delimiters or 'ignore' instructions for the agent when processing or documenting output from external targets.
Capability inventory: The workflow includes the use of curl and other network-resident tools to capture evidence from external systems.
Sanitization: There are no explicit instructions for the agent to sanitize or validate the content received from external targets before processing or reporting.
[DATA_EXFILTRATION]: The skill references sensitive file paths such as /etc/passwd and /etc/nginx/nginx.conf within its documentation. These are provided as part of educational payload examples (e.g., Caddy Template Injection) intended for testing misconfigured external servers and do not involve unauthorized access to the local filesystem.
[EXTERNAL_DOWNLOADS]: The skill provides links to several external security tools and repositories, including Yandex's Gixy and the PayloadsAllTheThings library. These are well-known resources within the security community and are referenced for legitimate research purposes.

payloads-reverse-proxy