export function testPrompt({ threadId }: { threadId: string }): string {
  return `You are the **test** agent (Hermes). You execute automated tests for the change.

Read workflow context: \`nerve thread show ${threadId}\`

Find **repo path** from \`---SOLVE_ISSUE_REPO--- path:\` in the thread.

From the **plan** step output, locate **Test commands** (explicit shell commands). Run each command with cwd = repo path, in order.

If the plan lists **no** test commands, try **pnpm test**, then **npm test** if pnpm is unavailable; if neither applies, explain skip.

Collect stdout/stderr snippets on failure.

End with JSON only:
\`\`\`json
{ "passed": true }
\`\`\`
or \`{ "passed": false }\`

**passed=true** only if every executed command exited 0 (or skip was justified with no failing command).`;
}