Skip to content

v6.23.13

Choose a tag to compare

@github-actions github-actions released this 15 Nov 17:33
· 6 commits to master since this release
573beae

Auto-released from merge of PR #1897 (trigger: push).

  • A seat with no five-hour window reads 5h none, not 5h 0%. A verified Ultra account has no five-hour limit, and its responses carry the weekly rate-limit headers and no 5h-* header at all (read on a live pool on 2026-11-15; a Max seat beside it carries both). dario read the missing header as 0, which routes correctly, and printed 5h 0%, which looks like a window nobody has used. dario accounts list --live, dario doctor, the dashboard, the terminal UI and the 429 log lines now say none; GET /accounts carries has5hWindow and GET /admin/accounts has_5h_window. Routing is unchanged, and a response with no rate-limit headers does not un-learn it. The dashboard sample has such a seat, its plan chips read Ultra and Max, and the README screenshots are rendered from it (#1893).
  • The proxy says when a seat is sent a usage-limit signal it does not act on. Claude Code 2.1.325 can carry on at lower priority, or finish a turn inside a grace zone, when a limit is reached; Anthropic announces both in anthropic-ratelimit-unified-slow-* and -grace-* response headers. dario uses neither. It now writes one line per seat and kind of signal when a response carries one, so you can see that the account is being offered it.
  • The test file count is 309 wherever it is stated. The README (two places) and the landing page said 289, and the release checklist said 77.
  • --pool-strategy=headroom-by-plan counts an Ultra seat by its week. Anthropic renamed Max 20x to Ultra on 2026-10-30: a verified Ultra account has no five-hour limit and a week 10 times Pro's, beside Max at 5 times Pro every five hours and every week. The seat's tier id is still default_claude_max_20x, and dario counted it 20, the old five-hour multiple, so an Ultra seat was sent four new conversations for each one a Max seat got with only twice the week to spend. It now counts 10 (Max 5, Pro 1). The other strategies read each seat's own windows and are unchanged: a seat that reports no five-hour window is measured by its week alone (#1893).
  • The docs name the plans as Anthropic does now. Pro, Max and Ultra in the README, the landing page, the FAQ, and the pool and two-plans pages; docs/multi-account-pool.md says how a pool of one Max and one Ultra seat behaves under each strategy.
  • An advisor_20260301 tool is sent with the beta it needs. dario has forwarded the entry as declared since 6.22.2, but the API accepts it only with advisor-tool-2026-03-01, and the template has that flag only when the Claude Code capture it came from did: the bundled template does, a live capture from an install without the advisor does not. On such an instance the request failed with tools.0: Input tag 'advisor_20260301' found using 'type' does not match any of the expected tags unless the client sent the flag itself or DARIO_PASSTHROUGH_BETAS named it. dario now adds the flag to a request it builds from the template whenever the outbound tools carry an advisor entry, in the position the bundled template has it. A request with no advisor entry is unchanged, and --passthrough and the credits rung still send only the client's own betas (#1812).

Built + tested + npm-published inline by cc-drift-auto-release.yml on this run — tokenless via OIDC trusted publishing, provenance-attested. See the workflow log for the publish output.