html-stable-publish

Warn

Audited by Gen Agent Trust Hub on Sep 13, 2026

Risk Level: MEDIUMREMOTE_CODE_EXECUTIONINDIRECT_PROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
  • [REMOTE_CODE_EXECUTION]: The skill uses npx postplan in SKILL.md and references/postplan.md, which downloads and executes the postplan package from the npm registry at runtime. As the source is not a verified or well-known organization, this poses a supply chain risk.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied HTML and static assets, which serves as an ingestion point for untrusted data that could contain malicious instructions.
  • Ingestion points: Files and directories passed to scripts/preflight_static_site.py and npx postplan upload.
  • Boundary markers: None explicitly defined to separate data from instructions during processing.
  • Capability inventory: File system reading via preflight_static_site.py, network exfiltration via npx postplan upload.
  • Sanitization: The Python preflight script uses regex to block known sensitive filenames (e.g., .env, id_rsa) and local path patterns, but does not sanitize the HTML content itself for prompt injection.
  • [DATA_EXFILTRATION]: The skill's primary function is to upload local content to https://postplan.dev. While the preflight script attempts to prevent the upload of credentials and system files, the workflow inherently moves local data to an external third-party domain.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Sep 13, 2026, 05:37 AM
Security Audit — agent-trust-hub — html-stable-publish