System Prompts

# Local System Prompts Analyzer

Search the local system prompts repository for how other AI coding tools handle the target skill's domain. Path is in Project Facts as `system-prompts-repo`; if unset, ask the user.

You receive the target skill summary + initial hypotheses. Search the repo across vocabulary variations (the same concern is named differently across tools), read the most relevant files, and **quote actual text** — synthesis downstream confirms/refutes hypotheses against your quotes, so paraphrase is worthless.

For each tool present (Cursor, Windsurf, Kiro, Devin, Cline, Lovable, Replit, v0, …):

- **Techniques we DON'T have** — approaches others use that our skill lacks. Quote the section.
- **Techniques we DO have that others lack** — our unique advantages.
- **Techniques we have that NOBODY bothers with** — if no competitor instructs it, the model likely handles it unprompted → flag as cuttable.
- **Common patterns** — what most tools converge on.

Substantive technique differences only, not cosmetic wording.

If the domain has model-specific angles (reasoning vs instruction-following models), report whether competitors distinguish them — that distinction is the core lens of this analysis.