The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses curl and shell pipes to interact with target systems and security APIs, which is expected for its designated purpose in vulnerability discovery.
[REMOTE_CODE_EXECUTION]: Automated scanners flagged instances of curl commands piped to python3. These patterns are used to parse JSON data using inline Python scripts (python3 -c) or standard utilities (python3 -m json.tool). Since the code resides within the skill rather than being fetched from the remote server, it does not represent a malicious remote code execution vector.
[EXTERNAL_DOWNLOADS]: The skill fetches data from public security databases including the NVD (National Vulnerability Database), WPScan, and Patchstack. These are recognized services for security information.
[DATA_EXFILTRATION]: The skill communicates with external APIs to retrieve CVE data. This behavior is documented and functional, used to support the primary goal of vulnerability assessment.
[PROMPT_INJECTION]: The skill processes data from untrusted WordPress sites (e.g., readme files and REST API responses). This presents a surface for indirect prompt injection, where a compromised or malicious target could return data meant to subvert the agent's logic.
Ingestion points: Target WordPress URLs, plugin readme files, and JSON API namespaces found in Phase 1 and Phase 6.
Boundary markers: No explicit delimiters or instructions are provided to the agent to treat external site content as potentially untrusted instructions.
Capability inventory: The skill environment provides curl and shell execution capabilities, which could be misused if an injection succeeds.
Sanitization: Content is filtered using standard Unix text processing tools (grep, regex), which provides basic extraction but lacks comprehensive sanitization against adversarial data.

wp-plugin-cve-hunt