implementing-attack-surface-management

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 1.00). The prompt shows and encourages embedding API keys and secrets directly in code and CLI arguments (e.g., Shodan key and Censys ID/Secret), which would require the LLM to handle or emit secret values verbatim and thus is insecure.

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (high risk: 0.85). The skill’s runtime workflow ingests outsider-authored free text via public reconnaissance results—e.g., it queries Shodan/Censys and runs ProjectDiscovery tools (subfinder/httpx/nuclei) whose fetched page/content and scan outputs (titles, service banners, vulnerability findings) are then read into the agent’s LLM context through the generated JSON/text reports (e.g., httpx_results.json, nuclei_results.txt, asm_report.json).

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.70). The skill instructs installing and running binaries fetched from GitHub (go install github.com/projectdiscovery/subfinder/v2/cmd/subfinder@latest, github.com/projectdiscovery/httpx/cmd/httpx@latest, github.com/projectdiscovery/nuclei/v3/cmd/nuclei@latest) and instructs updating nuclei templates at runtime (nuclei -ut which pulls from the ProjectDiscovery templates repo, e.g., https://github.com/projectdiscovery/nuclei-templates), all of which fetch remote code/templates that are executed and directly control scanning behavior required by the pipeline.