Put the response status and headers on the error the backoff reads

[unolife]

calculateDelay in core/llm/utils/retry.ts looks for

error.headers["retry-after"] ||
error.headers["x-ratelimit-reset"] ||
error.headers["ratelimit-reset"] || ...

so a retry waits exactly as long as the provider asked rather than guessing.

Nothing ever put them there. parseError builds its error from the response and keeps only the text — no status, no headers — so that branch could not fire and every rate limit fell through to exponential backoff.

What changed

parseError now annotates whatever it built with the response's status and a lower-cased plain-object copy of its headers.

• Lower-cased because that is how Headers yields them, and how the reader spells the names it checks first.
• A plain object because the reader uses bracket access, which returns undefined on a Headers instance regardless of casing.
• Left unset when the response carried none — an empty object reads as "the provider answered with no limits"; absent says it never answered.

No behaviour changes on its own. This only makes the existing header path reachable, and puts the limit within reach of anything else that wants it.

Testing

jest llm/index.test.ts — 189 passed, including three new cases covering the Retry-After round trip, the lower-casing, and the absent-headers case.

[github-actions]

Thank you for your submission, we really appreciate it. Like many open-source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution. You can sign the CLA by just posting a Pull Request Comment same as the below format.

---

I have read the CLA Document and I hereby sign the CLA

---

_{You can retrigger this bot by commenting recheck in this Pull Request. Posted by the CLA Assistant Lite bot.}