Creates false positives on pure text-based content

I ran Opus 4.8 on all the skills in my Github repo (https://github.com/francoisdp/skills) it came up with the following report:

All scans are complete. Here are the results before I change any file or push anything. None of the eight skills ships executable code, so every finding is a
pattern match against prose, a LICENSE file, or template comments.

SkillSpector results (the 8 deployed skills)

My assessment of the flagged items

I read every HIGH and CRITICAL line in the source. They are all false positives produced by the scanner pattern-matching documentation and template text:

• deploy-skill — CRITICAL "Harmful Content Injection" at README.md:33. The line is ordinary prose describing the /deploy-skill trigger and that the skill
  refuses to publish another person's work. Nothing harmful.
• marp-deck — 3× HIGH "Hidden Instructions" in templates. These are the Marp HTML comment blocks (<!-- Template: default … Replace [TITLE] -->). Legitimate
  template authoring comments, not hidden directives.
• anti-ai-writing — HIGH "Direct Prompt Extraction" at SKILL.md:916 is a section header about output types. HIGH "Tool Parameter Abuse" was matched inside the
  binary bytes of a PDF example.
• model-router — HIGH "Memory Manipulation" at SKILL.md:81 is the instruction describing what context to pass to a worker subagent, including reading CLAUDE.md.
  Normal dispatch description.
• The LOW at LICENSE:16 appears on all eight. It is the MIT licence text.

In substance, all eight skills are documentation-only, with no executable scripts, no network calls, no credential access, and no persistence mechanism. The tool
over-flags instruction-heavy skills. SkillSpector has no suppression or baseline flag, so I cannot make the raw verdict turn green for the four flagged skills
without editing the skills themselves.

Reproduce the report against the linked francoisdp/skills repository and inspect the cited matches, including README.md:33, SKILL.md:81 and :916, LICENSE:16, and the Marp HTML comments. Compare scanner behavior on prose, templates, license text, and binary example content; the issue is resolved when these documented false positives are handled without hiding genuine findings.