Add system-wide guardrails against fabricated results
This commit is contained in:
@@ -33,10 +33,12 @@ test("bundled prompts and skills do not contain blocked promotional product cont
|
||||
|
||||
test("draft workflow explicitly forbids fabricated results and unproven figures", () => {
|
||||
const draftPrompt = readFileSync(join(repoRoot, "prompts", "draft.md"), "utf8");
|
||||
const systemPrompt = readFileSync(join(repoRoot, ".feynman", "SYSTEM.md"), "utf8");
|
||||
const writerPrompt = readFileSync(join(repoRoot, ".feynman", "agents", "writer.md"), "utf8");
|
||||
const verifierPrompt = readFileSync(join(repoRoot, ".feynman", "agents", "verifier.md"), "utf8");
|
||||
|
||||
for (const [label, content] of [
|
||||
["system prompt", systemPrompt],
|
||||
["draft prompt", draftPrompt],
|
||||
["writer prompt", writerPrompt],
|
||||
["verifier prompt", verifierPrompt],
|
||||
|
||||
Reference in New Issue
Block a user