Main#252
Open
godlockin wants to merge 10 commits intoericc-ch:masterfrom
Open
Conversation
- src/lib/models.ts: resolveModel/resolveModelId with snapshot suffix stripping
(e.g. claude-opus-4-7-{snapshot} -> claude-opus-4.7) and family alias matching
- src/lib/claude-settings.ts: read env from ~/.claude/settings*.json for
COPILOT_REASONING_EFFORT overrides
- src/services/copilot/responses.ts: GitHub Copilot /responses endpoint adapter
for gpt-5.3-codex / gpt-5.4-mini (uses reasoning.effort, not budget_tokens)
- routes: thread resolveModel through chat-completions and count-tokens handlers
- anthropic-types: extend thinking.type with "adaptive" and add reasoning_effort
- main.ts: install global undici Agent with 10min headers/body timeout, 2min keep-alive, 30s connect — default 300s headersTimeout was killing >5min Copilot upstream responses (DOMException TIMEOUT_ERR) - start.ts: pass bun idleTimeout=255 (Bun max) to srvx + reusePort=true so keep-alive connections aren't closed under 10s default, which surfaces as "socket connection was closed unexpectedly" on the Claude Code side
Upstream Copilot opus-4.7 only accepts {type: "enabled"} (no "adaptive",
no "disabled"), and rejects effort levels above "medium". Coerce both
"enabled" and "adaptive" to "enabled" with optional budget_tokens, and
downgrade higher efforts to "medium" with a warning so legacy clients
keep working.
Co-Authored-By: Claude Opus 4 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4 <noreply@anthropic.com>
Probed upstream Copilot directly: opus-4.7 now accepts all effort levels (low/medium/high/xhigh/max) without 400, and reasoning_effort shows real gradient in completion tokens (low~214 → max~374) on a fixed prompt with thinking enabled. Lift the medium cap so high-effort requests reach the model instead of being silently downgraded. Tests updated to reflect the new pass-through behavior. Co-Authored-By: Claude Opus 4 <noreply@anthropic.com>
- New /usage/view route serving a self-contained dark-mode dashboard rendering quota_snapshots (chat/completions/premium_interactions) with progress bars, plan/SKU/reset metadata, and refresh controls. - /usage now content-negotiates: text/html → viewer, JSON otherwise, preserving API compatibility. - Auto-refresh defaults to 30s, with on/off toggle and 10s/30s/60s/5m selector; pauses on tab hidden, persists prefs to localStorage, honors ?endpoint, ?auto, ?refresh URL params. - claude-copilot.sh: add --daemon-start/--daemon-stop subcommands and print log path + usage URL after server is ready. Co-Authored-By: Claude Opus 4 <noreply@anthropic.com>
…vements Brings in the installer/docs/launcher work from godlockin/master (aa4bdaf) while keeping main's adaptive→enabled thinking translation and opened-up opus-4.7 effort cap (low/medium/high/xhigh/max) — master had reverted non-stream-translation.ts which silently drops the thinking field and caused upstream 400s for clients sending thinking.adaptive. - setup.sh: 5-step installer (deps, build, auth, shell config, launchd plist) - claude-copilot.sh: --daemon shows local dashboard URL (usage/view) - README: quick start, model tier table, effort support docs - .gitignore: add copilot-data/ docker volume exclusion - bun.lock: dependency updates from master Co-Authored-By: Claude Opus 4 <noreply@anthropic.com>
…fort
Even after coercing thinking.adaptive→enabled, upstream Copilot still
returned 400 "Input tag 'adaptive' found using 'type'" — likely because
the validator scans the overall request body (including conversation
content) for thinking-shaped strings, which is fragile when chat history
mentions the word "adaptive".
Effort already conveys reasoning intent (low→max); thinking is redundant
for this provider. Stop forwarding thinking unconditionally.
Tests: 3 thinking-translation cases updated to expect undefined.
Probe: thinking.{adaptive,enabled+budget=31999} + reasoning_effort
{low,medium,high,xhigh,max} all → 200.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
No description provided.