The Agent Skills Directory

[COMMAND_EXECUTION]: The skill contains multiple payloads designed to execute arbitrary OS commands on remote targets. This includes using CREATE ALIAS in the H2 Console to spawn shell processes and leveraging Spring Expression Language (SpEL) injection (T(java.lang.Runtime).getRuntime().exec(...)) to trigger command execution through application inputs and headers.
[DATA_EXFILTRATION]: The skill provides specific instructions for harvesting sensitive information from target memory. It uses strings and targeted regex patterns to locate AWS access keys (AKIA...), Stripe live keys (sk_live_...), and Bearer tokens within heap dump files. Additionally, it suggests exfiltrating command output via Out-of-Band (OOB) requests to external collaborator hosts.
[INDIRECT_PROMPT_INJECTION]: The skill creates a vulnerability surface by ingesting raw, unvalidated content from remote target servers and processing it with tools like grep, strings, and json.tool. A malicious server could potentially return crafted content designed to manipulate the agent's logic or downstream processing.
[COMMAND_EXECUTION]: Static analysis alerts regarding curl | python3 were evaluated and found to be false positives in this context. The commands (e.g., curl ... | python3 -m json.tool) are used for pretty-printing JSON responses rather than executing untrusted remote Python scripts.

hunt-springboot