The Agent Skills Directory

[EXTERNAL_DOWNLOADS] (MEDIUM): The skill installs the Nuclei security scanner directly from a remote GitHub repository using go install github.com/projectdiscovery/nuclei/v3/cmd/nuclei@latest. Because this uses the @latest tag rather than a pinned version/hash and targets a repository not in the specific trusted organization list, it presents a supply chain risk where the tool could be modified at the source.
[REMOTE_CODE_EXECUTION] (MEDIUM): The skill executes remote code by pulling and running Docker images from ghcr.io/zaproxy/zaproxy:stable. Although ZAP is a reputable tool, running opaque containers with broad network/filesystem access represents a significant security surface.
[COMMAND_EXECUTION] (LOW): The skill makes extensive use of subprocess calls to run system commands like docker, apt-get, and nikto. This is consistent with the primary purpose of a DAST skill but increases the potential impact of any successful injection.
[INDIRECT_PROMPT_INJECTION] (LOW): The skill ingests untrusted data from external web applications (headers, HTML, API responses) through the reporting mechanisms of ZAP, Nuclei, and Burp Suite.
Ingestion points: zap-baseline.py -r baseline-report.html, nuclei -o results.json, and scanner.get_issues(scan_id).
Boundary markers: Absent; reports are processed as raw data without explicit instruction delimiters.
Capability inventory: Subprocess execution (docker, apt-get), file system writes (mkdir, -o results.json), and network operations (requests.post).
Sanitization: No evidence of sanitization or validation of the tool outputs before they are processed by the agent.

dast-scanning