Evalkeeper
    Preparing search index...

    Function promptVersionOf

    • One instructions hash from its two parts: the instructions in code, and what's edited at runtime (an admin override, guidance injected at call time). Use it for promptVersion (the runner's and createSettings's) when part of what the model receives isn't in code, so a test is out of date when either part changes, and syncPrompt can tell a deploy (DEPLOY) from a runtime edit (INSTRUCTIONS). With no runtime part, it's the code hash unchanged, so existing hashes stay the same.

      Parameters

      • parts: { code: string; runtime: string | null }

        code: the hash of the instructions in code. runtime: a hash of everything added or replaced at runtime, or null for none. Neither may contain +, which joins them.

      Returns string

      const settings = createSettings({
      store,
      defaults,
      promptVersion: async (op) =>
      promptVersionOf({
      code: PROMPT_VERSIONS[op],
      runtime: await guidanceHash(op), // null when there's none
      }),
      });