Prompt file imported from Antonio-Redondo/staywise-ai (.github/prompts/new-skill.prompt.md). Fill in {{skillName}}, {{purpose}}, {{inputType}}, {{outputType}}, {{edgeCases}} before use. Copyright stays with the author.

/new-skill

Create a new pure-function skill following the "skills versus nodes" rule in .github/instructions/langgraph.instructions.md.

Before generating, verify all three conditions hold for this skill:

1. It is a pure function (no I/O, no LLM, no env vars, no Date.now or Math.random unless seeded).
2. It is unit-testable with fixture inputs and deterministic outputs.
3. It would still make sense if the agent layer did not exist.

If any condition fails, stop and tell me this should be a node or a client, not a skill.

Generate two files:

1. src/skills/{{skillName}}.ts

   • Export a named function. No default export.
   • Explicit type signature on the public surface.
   • Early returns over nested conditionals.
   • JSDoc comment above the function explaining the why, not the what.

2. tests/skills/{{skillName}}.test.ts

   • Vitest tests covering at least one happy path, one boundary case, and one rejection case.
   • Inline fixtures, no shared test setup.
   • One assertion per test where reasonable.

Purpose: {{purpose}}

Input type: {{inputType}}

Output type: {{outputType}}

Edge cases to cover in tests: {{edgeCases}}

Success criteria: pnpm test passes, pnpm typecheck passes, the skill is pure, and tests cover happy path plus edge cases.