The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides extensive templates for shell commands using curl, whatweb, nuclei, and fscan. These are used to identify targets, fingerprint systems, and deliver exploit payloads to network targets. This is the primary intended behavior of the skill.
[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to fetch and process data from external targets (e.g., via curl or whatweb output). This creates a vulnerability where a malicious target could supply responses (such as HTTP headers or page content) designed to inject instructions into the agent's processing logic.
Ingestion points: Target discovery phase in SKILL.md and detailed exploitation steps in references/oa-exploits.md and references/oa-exploits-more.md.
Boundary markers: Absent; there are no delimiters or warnings to ignore instructions embedded in the target's output.
Capability inventory: Network requests (curl), execution of third-party tools (nuclei, fscan, EHole), and command execution (java).
Sanitization: No sanitization or validation of the data retrieved from external targets is performed before the agent processes it.
[EXTERNAL_DOWNLOADS]: The skill documentation references and provides usage examples for several third-party security tools, including nuclei, fscan, EHole, and ysoserial. Although the skill does not automatically download or install these binaries, their presence is a prerequisite for the workflows described in the skill.

oa-system-attack