humanize

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 1.00). The skill requires reading and then outputting or writing back full file contents while explicitly preserving YAML frontmatter, code fences and other non‑prose sections verbatim, so any secrets in those parts (or provided inline) would be included in the LLM's generated output and can be exfiltrated.

MEDIUM W013: Attempt to modify system services in skill instructions.

• Attempt to modify system services in skill instructions detected (high risk: 0.90). The skill explicitly instructs the agent to read a file by path and write the humanized content back in-place, allowing arbitrary filesystem modifications (including privileged system files) and thus risks changing the machine's state.