මේ skill ඒක හැදුවෙ මම 
Prompt එක ලිව්වෙ මම
The whole thing landed as one "Add files via upload" commit with no history, no iteration, no typos, and nine reference files of near-identical length (5–8 KB each) with the same voice and the same checkbox format throughout. Humans writing 60 KB of checklists over time leave fingerprints: inconsistent formatting, a section that's twice as long because it's their specialty, stale references, the odd personal opinion. This has none of that. Every framework gets equal, encyclopaedic, evenhanded coverage, which is what a model produces when asked to "write a security skill covering X, Y, Z."
The grep block is another giveway: it looks like a regex list but contains prose fragments like "for tokens/passwords" and "(hardcoded values)" mixed in. A human who actually runs those greps would have noticed they don't work. A model writing what a grep list
should look like produces exactly that.
The SKILL.md frontmatter and the "Use / Layout / Customise" README structure follow Anthropic's skill-authoring conventions very closely, which suggests it was produced with Claude (or Claude Code) following the skill-creator guidance. The "Claude-specific tips" section at the end of ai-llm.md points the same way.
