Distinguish between "provider rejects probe" (yellow warning) and
"genuinely broken" (red error) in health check results.
Backend: add AppError::HttpStatus variant to carry structured HTTP
status codes, populate http_status on error results, classify codes
into short labels (e.g. "Auth rejected (401)"), and truncate overly
long response bodies.
Frontend: route 401/403/400/429/5xx to toast.warning with localized
hints explaining the error may not indicate actual unusability; route
404/402/connection errors to toast.error. Add i18n keys for all three
locales (zh/en/ja).
Also deduplicate check_once by reusing build_stream_check_result.
When a repo itself is a single skill (SKILL.md at repo root), the
discovery phase sets directory to the repo name, but after ZIP
extraction (which strips the root folder), no matching subdirectory
exists. Add a fallback to check if SKILL.md exists directly in the
extracted temp directory before reporting SKILL_DIR_NOT_FOUND.
Fixes installation of repos like zlbigger/Google-SEOs.skill.
The per-provider proxy configuration (meta.proxyConfig) is removed
because its scope is too narrow and covered by global proxy settings
and proxy takeover mode. Users can achieve the same result via the
global proxy panel.
Changes:
- Remove ProviderProxyConfig type (frontend TS + backend Rust)
- Remove ProviderAdvancedConfig proxy UI block, keep testConfig/pricingConfig
- Simplify http_client: delete build_proxy_url_from_config,
build_client_for_provider, get_for_provider
- Simplify forwarder/stream_check/model_fetch to use global client
- Remove i18n keys (en/zh/ja)
- Fix pre-existing test bug in transform.rs (extra None arg)
- Fix request classification: treat messages containing tool_result as
agent continuation instead of user-initiated, preventing false premium
charges on every tool call
- Add subagent detection via __SUBAGENT_MARKER__ and metadata._agent_
fallback, setting x-interaction-type=conversation-subagent
- Add deterministic x-interaction-id derived from session ID to group
requests into a single billing interaction
- Add orphan tool_result sanitization to prevent upstream API errors
that could cause retries and duplicate billing
- Reorder pipeline: classify (on original body) → sanitize → merge →
warmup, ensuring classification sees raw tool_result semantics
- Enable warmup downgrade by default with gpt-5-mini model
- Enhance session ID extraction priority chain for Copilot cache keys
- Detect infinite whitespace bug in streaming tool call arguments
SSRF protection (private IP blocking, suspicious hostname detection) was
originally added for web-server threat models but is unnecessary for a
local desktop app where the user already has full network access. This
removal unblocks legitimate use cases like enterprise intranet APIs,
Docker container addresses, and self-hosted services.
Retained: HTTPS enforcement and same-origin checks which still provide
meaningful security (protecting API keys in transit and preventing
scripts from leaking keys to unrelated domains).
OpenClaw gateway injects `[message_id: UUID]` metadata at the end of
every message, wasting display space. Strip this suffix from both title
and summary fields.
Also change session title display from single-line truncate to
line-clamp-2, so longer titles (e.g. OpenClaw's timestamp-prefixed
messages) can show more meaningful content across two lines.
Previously Codex and OpenClaw sessions only showed the working directory
basename as the title, making it hard to distinguish sessions in the same
project. Now both providers extract the first real user message as the
session title, matching the existing Claude Code behavior.
- Codex: first user message → dir basename (skips AGENTS.md injection)
- OpenClaw: displayName (sessions.json) → first user message → dir basename
- Move TITLE_MAX_CHARS constant to shared utils.rs
- Use Option<&HashMap> for OpenClaw parse_session to avoid leaky abstraction
Extract message_id from Claude API responses (msg_xxx) and use it to
generate a shared request_id format (session:{msg_xxx}) between the
proxy logger and session log sync. When session sync encounters the
same request_id via INSERT OR IGNORE, it skips the duplicate.
- Add message_id field to TokenUsage, extracted from Claude responses
- Add TokenUsage::dedup_request_id() to generate shared request IDs
- Define SESSION_REQUEST_ID_PREFIX constant to eliminate magic strings
- Change proxy logger to INSERT OR REPLACE for richer-data-wins semantics
Add pricing data for 4 new providers (Qwen, xAI Grok, Mistral, Cohere)
and supplement existing providers (MiniMax M2.5/M2.7, GLM-5/5.1,
Doubao Seed 2.0, MiMo V2 Pro, OpenAI o1/o3/codex-mini/gpt-5-mini/nano).
Fix outdated prices for deepseek-chat, deepseek-reasoner, and kimi-k2.5.
Fix display_name casing "Mimo" → "MiMo" for consistency.
Use prepared statement in seed_model_pricing() to avoid recompiling SQL
on each of ~130 INSERT iterations.
Schema migration v8→v9: DELETE + re-seed model_pricing for existing users.
Prevent users from switching to official providers (Anthropic/OpenAI/Google)
when proxy takeover is active, as using a proxy with official APIs may cause
account bans.
Defense-in-depth across 4 layers:
- Backend: ProviderService::switch(), hot_switch_provider(), switch_proxy_provider command
- Frontend: useProviderActions soft guard with error toast
- UI: ProviderActions button disabled with ShieldAlert icon
- Tray menu: official provider items disabled with ⛔ indicator
Also warns when enabling proxy takeover while current provider is official.
* Preserve cache hints when collapsing system prompts
Strict OpenAI-compatible chat backends still need fragmented Claude\nsystem prompts collapsed into one leading system message, but that\nnormalization should not silently drop stable cache hints. Preserve\nmessage-level cache_control when the merged system fragments agree,\nand fall back to omitting it when the fragments conflict.\n\nConstraint: Must keep single-system normalization for Nvidia/Qwen-style chat backends\nRejected: Always copy the first cache_control | could misrepresent conflicting cache boundaries\nConfidence: high\nScope-risk: narrow\nReversibility: clean\nDirective: If system prompt merging changes again, preserve cache_control whenever the merged metadata is unambiguous\nTested: cargo test proxy::providers::transform --manifest-path src-tauri/Cargo.toml\nNot-tested: End-to-end prompt caching behavior against cache-aware OpenAI-compatible upstreams\nRelated: #1881
* Tighten cache hint inheritance for merged system prompts
The follow-up cache hint fix still treated mixed present/absent\ncache_control across fragmented system prompts as inheritable, which\nexpanded the cache scope after prompt collapse. Treat that mix as\nambiguous and only preserve cache_control when every merged fragment\nexplicitly agrees on the same value.\n\nConstraint: Must preserve strict-backend system prompt normalization from #1942\nRejected: Inherit first present cache_control | widens cache scope when later fragments were intentionally uncached\nConfidence: high\nScope-risk: narrow\nReversibility: clean\nDirective: Any future merged-system cache hint logic should treat missing cache_control as semantically significant\nTested: cargo test proxy::providers::transform --manifest-path src-tauri/Cargo.toml\nNot-tested: End-to-end upstream caching behavior against cache-aware relays\nRelated: #1881\nRelated: #1946
* Keep cache-control merge regressions easy to review
Reflow the two long cache-control regression assertions in transform.rs so the neighboring merge cases stay rustfmt-aligned and easier to scan.
This keeps the preserved code change separate from the untracked Markdown design notes the user did not want committed.
Constraint: Exclude Markdown design files from the commit while preserving the local code change
Rejected: Include docs in the same commit | user explicitly asked to leave Markdown files out
Confidence: high
Scope-risk: narrow
Reversibility: clean
Directive: Treat this as a readability-only test change; do not infer runtime behavior changes from it
Tested: cargo test --manifest-path src-tauri/Cargo.toml test_anthropic_to_openai_drops_ --lib
Tested: cargo check --manifest-path src-tauri/Cargo.toml --tests
Tested: pnpm format:check
Tested: pnpm typecheck
Not-tested: Full application integration and manual flows
Responses conversions still use promptCacheKey, but chat completions now stay a pure shape transform. This keeps Claude -> chat requests aligned with providers that do not understand the field and keeps stream checks consistent with production behavior.
Constraint: Issue #1919 requires removing prompt_cache_key from Claude -> OpenAI Chat requests
Rejected: Add a runtime toggle for chat injection | requested behavior is unconditional removal
Confidence: high
Scope-risk: narrow
Reversibility: clean
Directive: Keep promptCacheKey limited to Claude -> Responses conversions unless a provider-specific contract is proven
Tested: cargo test anthropic_to_openai
Tested: cargo test anthropic_to_responses_with_cache_key
Tested: cargo test transform_claude_request_for_api_format_responses
Not-tested: Full src-tauri test suite
Related: #1919
* feat(window): add app-level window controls with settings toggle
Add a persistent settings toggle to enable app-level minimize/maximize/close controls and hide system decorations when enabled, providing a Wayland-friendly fallback for broken native titlebar interactions.
Co-authored-by: Cursor <cursoragent@cursor.com>
* fix(window): restrict app-level window controls to Linux only and fix startup flicker
- Guard useAppWindowControls with isLinux() in App.tsx so it's always
false on macOS/Windows even if persisted as true
- Wrap set_decorations call in lib.rs with #[cfg(target_os = "linux")]
- Only show the toggle in WindowSettings on Linux
- Skip setDecorations effect while settingsData is still loading to
prevent the Rust-side decoration state from being overridden by the
undefined->false fallback, which caused a brief title bar flicker
---------
Co-authored-by: wzk <wx13571681304@outlook.com>
Co-authored-by: Cursor <cursoragent@cursor.com>
Co-authored-by: Jason <farion1231@gmail.com>
When thinking.type is "adaptive" (Claude's maximum thinking mode) and
output_config.effort is absent, resolve_reasoning_effort() incorrectly
mapped it to "high" instead of "xhigh" in OpenAI format conversions.
Kaku is a WezTerm-derived macOS terminal, so reusing the existing WezTerm-compatible launch path keeps the change small while making it selectable in settings and session resume flows.
Constraint: Kaku support should stay macOS-only and avoid introducing a separate launcher model
Rejected: Treat Kaku as a silent WezTerm fallback | users could not explicitly choose it in settings
Confidence: high
Scope-risk: narrow
Reversibility: clean
Directive: Keep Kaku on the shared WezTerm-compatible launch path unless upstream drops the start-compatible CLI
Tested: pnpm typecheck; pnpm format:check; cargo check --manifest-path src-tauri/Cargo.toml; cargo fmt --manifest-path src-tauri/Cargo.toml --check; cargo test --manifest-path src-tauri/Cargo.toml --lib session_manager::terminal::tests
Not-tested: End-to-end launch against a locally installed Kaku.app
Related: #1954
- Make migrate_v6_to_v7 check skills table existence before ALTER
- Make migrate_v7_to_v8 check model_pricing table existence before UPDATE
- Fix SessionManagerPage test: use getByRole heading instead of getAllByText
which breaks when highlightText splits text across <mark> elements
- Add content_hash and updated_at fields to 4 InstalledSkill literals in skill_sync.rs
- Add useCheckSkillUpdates and useUpdateSkill to UnifiedSkillsPanel test mock
- Suppress unused import warning in auto_launch.rs test module
Instead of showing directory basenames for all Claude sessions, extract
titles from JSONL content with a priority chain:
1. custom-title metadata (set via /rename in Claude Code)
2. First real user message (skipping /clear, /compact caveats)
3. Directory basename (fallback)
Display a one-time informational dialog explaining the Common Config
Snippet feature when users first open the add/edit provider form.
Uses a derived isOpen state from settings to avoid race conditions.
Adds commonConfigConfirmed flag to both TS and Rust settings types.
Introduce a one-time welcome dialog that explains CC Switch's workflow
to new users: how their existing config is preserved as a "default"
provider and how the bundled "Official" preset enables one-click revert.
Upgrade users are excluded by checking is_providers_empty() at startup
and never see the dialog.
Persistence follows the existing *_confirmed convention in AppSettings
(proxy/usage/stream_check/failover), stored in settings.json. The field
is only written when the user explicitly clicks the confirm button,
keeping its semantics strictly about user acknowledgement.
Also adds two reusable DAO helpers:
- Database::is_providers_empty for fresh-install detection, using
EXISTS(SELECT 1) for a short-circuit query.
- Database::get_bool_flag accepting "true" | "1", with
init_default_official_providers migrated to use it.
Dialog copy in zh/en/ja uses conditional phrasing so it stays
accurate whether or not existing live config was found.
Drops the friction of clicking the manual "Import current config" button
for OpenCode and OpenClaw — they now match the auto-import behavior the
previous commit added for Claude/Codex/Gemini.
- New "1.6." startup block in lib.rs runs both
import_opencode_providers_from_live and import_openclaw_providers_from_live
on every launch. The functions are id-keyed and idempotent, so re-running
just picks up new providers added externally to the live JSON files.
- Both functions now use a new Database::get_provider_ids() helper
(HashSet<String> from a single SELECT id-only query) instead of
get_all_providers(), avoiding the N+1 endpoint sub-queries that would
otherwise hit the startup hot path on every launch.
New and existing users now see a built-in "Claude Official" / "OpenAI
Official" / "Google Official" entry in their provider list, so switching
back to the official endpoint is one click away instead of buried in the
README.
- New providers_seed.rs holds the three seeds (id, name, settings_config,
icon) keyed by AppType, with a single is_official_seed_id() helper that
scans OFFICIAL_SEEDS so the id list has one source of truth.
- Database::init_default_official_providers() runs once per database
(gated by an official_providers_seeded setting flag), appends each seed
to the end of the sort order, and never touches is_current.
- Startup also auto-imports the live config (settings.json / auth.json /
.env) as a "default" provider before seeding, so users with an existing
manual config don't lose it when they click the official preset.
- Database::has_non_official_seed_provider() replaces the get_all_providers
call in import_default_config's gating check with an id-only scan,
dropping the N+1 endpoint sub-queries from every startup.
Post-merge cleanup from a simplify review pass on the phase 1-4
OpenCode/OpenClaw changes.
- Rename check_once_opencode_like → check_once_without_adapter. The
new name directly expresses the intent (bypass get_adapter) instead
of suggesting the function is somehow "like" OpenCode.
- Drop two "Phase 4 会美化错误消息" phase-history markers from
docstrings; git history is the right place for them.
- Document in resolve_opencode_base_url why its default endpoints
cannot be merged with ProviderType::default_endpoint(): the former
encode AI SDK package defaults (e.g. @ai-sdk/openai ships with the
/v1 suffix) while the latter encode proxy upstream hosts. They
happen to overlap but are two independent truth sources.
Phase 4: polish the four remaining edge cases uncovered by Phase 1-3.
Custom headers passthrough
- check_claude_stream and check_gemini_stream now accept an optional
extra_headers map which is appended after all built-in headers so it
can override defaults (e.g. a custom User-Agent).
- OpenClaw reads from settings_config.headers.
- OpenCode reads from settings_config.options.headers.
- All pre-existing Claude/Codex/Gemini call sites pass None.
OpenClaw custom auth header (Longcat-style)
- When settings_config.authHeader is true, the provider expects a
custom auth header whose name is only known to the OpenClaw gateway
itself. Return a dedicated openclaw_auth_header_not_supported error
so the user sees a meaningful explanation instead of a 401.
Bedrock error polish
- The bedrock-converse-stream (OpenClaw) and @ai-sdk/amazon-bedrock
(OpenCode) branches now explain why (SigV4 signing) and point to
the official consoles as an alternative test path.
OpenCode baseURL fallback
- resolve_opencode_base_url: when options.baseURL is empty and the
npm package has a canonical default endpoint (@ai-sdk/openai,
@ai-sdk/anthropic, @ai-sdk/google), fall back to that endpoint.
@ai-sdk/openai-compatible still requires an explicit baseURL
because its whole purpose is to point at a custom OpenAI clone.
Tests
- 8 new unit tests covering authHeader detection, baseURL resolution
(explicit / fallback / error), and header map extraction on both
apps. Total stream_check tests: 18 → 26.
Phase 3: implement stream check for OpenCode providers by mapping the
`settings_config.npm` (AI SDK package name) to the corresponding API
protocol and delegating to the existing stream checkers.
Package mapping:
- @ai-sdk/openai-compatible → openai_chat
- @ai-sdk/openai → openai_responses
- @ai-sdk/anthropic → anthropic (ClaudeAuth strategy)
- @ai-sdk/google → gemini (Google strategy)
- @ai-sdk/amazon-bedrock → not supported (phase 4 message polish)
Note: OpenCode nests baseURL/apiKey under `settings_config.options`
(different from OpenClaw's root-level fields) and uses `baseURL` with
a capital L. Three new extractors (base_url / api_key / npm) encode
these shape differences so check_opencode_stream stays symmetric with
check_openclaw_stream.
Frontend: drop the remaining `appId !== "opencode"` filter in
ProviderList.tsx — both apps can now test providers.
Phase 2: extend check_openclaw_stream to cover the full non-Bedrock
protocol set declared by openclawApiProtocols.
- openai-responses → check_claude_stream(api_format="openai_responses")
- anthropic-messages → check_claude_stream(api_format="anthropic"),
using AuthStrategy::ClaudeAuth (Bearer-only) so Claude relay
services that reject a simultaneous x-api-key still work. Official
Anthropic also accepts pure Bearer on /v1/messages.
- google-generative-ai → check_gemini_stream with AuthStrategy::Google.
bedrock-converse-stream still errors out but with a dedicated
openclaw_bedrock_not_supported key; its user-facing message will be
polished in phase 4.
Each protocol now builds its own AuthInfo inside the match arm because
the auth strategy is protocol-specific.
Phase 1 of extending stream health check to OpenCode/OpenClaw apps.
- Add early-dispatch path for OpenCode/OpenClaw in check_once so they
bypass the adapter layer (which only knows Claude/Codex/Gemini
settings_config shapes).
- Introduce check_openclaw_stream dispatcher that reads the `api` field
from settings_config and routes to the existing check_claude_stream
with api_format="openai_chat" for "openai-completions". Other
protocols return localized errors to be lit up in phases 2 and 4.
- Extract build_stream_check_result helper to avoid duplicating the
StreamCheckResult construction logic between the two code paths.
- Unblock the test button for OpenClaw providers in ProviderList.tsx.
OpenCode still returns the "not yet supported" error; it will be
enabled in phase 3.
Linux users reported the window UI (including native title bar buttons)
couldn't receive clicks until manually maximizing and restoring the
window. Root causes: (1) Tauri webview did not acquire focus on startup
so first clicks were consumed by X11/Wayland click-to-activate
(Tauri #10746, wry #637); (2) GTK surface input region failed to
renegotiate on the visible:false + show() path under some
WebKitGTK/compositor combinations.
- Add linux_fix::nudge_main_window helper that performs set_focus plus
a ±1px no-op resize after window show, with a 500ms reconciliation
readback to compensate for dropped resize requests on slow
compositors.
- Wire the helper into every window re-show path: normal startup,
deeplink, single_instance, tray show_main, and lightweight exit.
- Set WEBKIT_DISABLE_COMPOSITING_MODE=1 at startup to avoid resize
crashes and Wayland surface negotiation issues.
- Remove data-tauri-drag-region on Linux from App.tsx header and the
shared FullScreenPanel (used by all provider/MCP/workspace forms)
to avoid Tauri #13440 in Wayland sessions. Extract drag-region
constants to src/lib/platform.ts for reuse.
All Rust changes are gated by #[cfg(target_os = "linux")]; frontend
changes preserve macOS/Windows behavior via runtime isLinux() checks.
Known limitation: tiling Wayland compositors ignore set_size, so
GDK_BACKEND=x11 remains the user-side workaround.
Codex OAuth (ChatGPT Plus/Pro) providers previously fell through to the
default UsageFooter branch and showed no quota at all, while Copilot and
official Codex providers already had a wham/usage-backed quota footer.
This wires up the same five-hour / seven-day tier badges for codex_oauth
provider cards by reusing the existing query_codex_quota function and
SubscriptionQuotaFooter rendering, parameterized to keep both the CLI
credential path ("codex") and the cc-switch managed OAuth path
("codex_oauth") working from a single source of truth.
- Parameterize services::subscription::query_codex_quota with tool_label
and expired_message; promote SubscriptionQuota constructors to
pub(crate). The CLI path keeps its existing "codex" label and the
"re-login with Codex CLI" message; the new path passes "codex_oauth"
and a cc-switch-specific re-login hint.
- Add a new get_codex_oauth_quota Tauri command in commands/codex_oauth.rs
that resolves the ChatGPT account (explicit binding > default account
> not_found), pulls a valid access_token from CodexOAuthManager
(auto-refresh handled), and delegates to query_codex_quota.
- Extract SubscriptionQuotaFooter's render body into a pure
SubscriptionQuotaView component (props: quota / loading / refetch /
appIdForExpiredHint / inline). The existing SubscriptionQuotaFooter
becomes a thin wrapper with identical props and behavior, so
CopilotQuotaFooter and the official Claude/Codex/Gemini paths are
untouched. This avoids duplicating ~280 lines of five-state rendering.
- Add CodexOauthQuotaFooter, a 38-line wrapper that calls the new
useCodexOauthQuota hook and forwards to SubscriptionQuotaView.
- ProviderCard inserts an isCodexOauth branch between isCopilot and
isOfficial, keyed off PROVIDER_TYPES.CODEX_OAUTH (newly added to
config/constants.ts to centralize the previously scattered string).
- Frontend hook caches per (codex_oauth, accountId) so multiple cards
bound to the same ChatGPT account share one fetch via react-query
dedup; cards bound to different accounts get independent fetches.
- No new i18n keys: existing subscription.fiveHour / sevenDay / expired /
refresh / queryFailed / expiredHint are reused.
Adds a new managed OAuth provider that lets Claude Code route requests
through a user's ChatGPT Plus/Pro subscription via the chatgpt.com
backend-api/codex endpoint.
- CodexOAuthManager: OpenAI Device Code flow with multi-account support,
JWT-based account identification, and automatic access_token refresh.
- Reuses the generic managed-auth command surface (auth_start_login,
auth_poll_for_account, etc.) via provider dispatch in commands/auth.rs.
- ClaudeAdapter detects codex_oauth providers, forces the base URL to
the ChatGPT backend, pins api_format to openai_responses, and emits
Authorization + originator headers; the forwarder injects the dynamic
access_token and ChatGPT-Account-Id per request.
- transform_responses gains an is_codex_oauth path that aligns the body
with OpenAI's codex-rs ResponsesApiRequest contract: sets store:false,
appends reasoning.encrypted_content to include, strips max_output_tokens
/ temperature / top_p, injects default instructions/tools/parallel_tool_calls,
and forces stream:true. Covered by 9 new unit tests plus regression
guards for the non-Codex path.
- Stream check reuses the same transform flag so detection matches the
production request shape.
- Frontend adds CodexOAuthSection + useCodexOauth hook, integrates it
into ClaudeFormFields / ProviderForm / AuthCenterPanel, ships a new
"Codex (ChatGPT Plus/Pro)" preset, and adds zh/en/ja i18n strings.
Session logs use placeholder provider_ids (_session, _codex_session,
_gemini_session) that don't exist in the providers table, causing LEFT
JOIN to return NULL and display "Unknown". Add COALESCE fallback in all
4 usage queries to show meaningful names like "Claude (Session)".
Add dashboard-level app type filter to usage statistics, replacing the
DataSourceBar with a more useful segmented control. All components
(summary cards, trend chart, provider stats, model stats, request logs)
now respond to the selected app filter.
Backend: add optional app_type parameter to get_usage_summary,
get_daily_trends, get_provider_stats, and get_model_stats queries.
Frontend: new AppTypeFilter type, updated query keys with appType
dimension for proper cache separation, and RequestLogTable local
filter auto-locks when dashboard filter is active.
- Use UPSERT with WHERE guard instead of INSERT OR IGNORE, so updated
token values on existing messages are properly synced without
unnecessary rewrites of unchanged rows
- Include cached tokens in the skip-zero filter to stop silently
discarding pure cache-hit records
- Restrict file collection to session-*.json to match documented scope
and prevent ingesting non-session JSON files
Parse ~/.gemini/tmp/*/chats/session-*.json for precise per-message
token data (input/output/cached/thoughts). Integrates with existing
background sync and manual sync button alongside Claude and Codex.
Normalize model names from JSONL session logs before storage and pricing
lookup: lowercase, strip provider prefix (openai/), strip date suffixes
(-YYYY-MM-DD, -YYYYMMDD). Also clamp cached tokens to not exceed input.
- Fix 13 Chinese model prices that were stored as CNY values in USD
fields (DeepSeek, Kimi, MiniMax, GLM, Doubao, Mimo)
- Add 12 new models: GPT-5.4/mini/nano, o3, o4-mini, GPT-4.1/mini/nano,
Gemini 3.1 Pro/Flash Lite, Gemini 2.5 Flash Lite, Gemini 2.0 Flash,
DeepSeek Chat/Reasoner, Kimi K2.5
- Merge pricing migration into existing v7→v8 to avoid extra version bump
Parse Claude Code JSONL session files (~/.claude/projects/) and Codex
SQLite database (~/.codex/state_5.sqlite) to track API usage without
requiring proxy interception. This enables usage statistics for users
who don't use the proxy feature.
Key changes:
- Add session_usage.rs: incremental JSONL parser with message.id dedup
- Add session_usage_codex.rs: import thread-level token data from Codex
- Add data_source column to proxy_request_logs (proxy/session_log/codex_db)
- Add session_log_sync table for tracking parse offsets
- Background sync every 60s + manual sync via DataSourceBar UI
- Schema migration v7→v8
- i18n support for zh/en/ja
- Hide "暂无描述" text when skill has no description (skills.sh API
doesn't return descriptions), show empty spacer instead
- Change skills.sh result link from guessed subdirectory path to repo
root URL, since skillId doesn't reflect the actual nested path
Add skills.sh API integration allowing users to search and install from
a catalog of 91K+ agent skills directly within CC Switch. The search
results are converted to DiscoverableSkill objects and reuse the existing
install pipeline. Includes fallback directory search for repos where
skills are nested in subdirectories, and filters out non-GitHub sources.
Allow users to choose between storing skills in CC Switch's managed
directory (~/.cc-switch/skills/) or the Agent Skills open standard
directory (~/.agents/skills/). Includes migration logic that safely
moves files before updating settings, with confirmation dialog for
non-empty installations.
- Add content_hash and updated_at fields to skills table (DB migration v6→v7)
- Compute directory content hash on install/import/restore for version tracking
- Add check_updates command: downloads repos, compares hashes, returns update list
- Add update_skill command: backs up old files, re-downloads and replaces SSOT
- Backfill content_hash for existing skills on first update check
- Add "Check Updates" button and per-skill update badge/button in UnifiedSkillsPanel
- Add i18n keys for zh/en/ja
Add a new "Official" (官方) template type in the usage query panel that
queries account balance via each provider's native API endpoint.
Follows the same zero-script pattern as Token Plan — Rust handles the
HTTP call, frontend auto-detects the provider from base URL.
Supported providers and endpoints:
- DeepSeek: GET /user/balance
- StepFun: GET /v1/accounts
- SiliconFlow: GET /v1/user/info (cn + com)
- OpenRouter: GET /api/v1/credits
- Novita AI: GET /v3/user/balance
* fix: handle UTF-8 multi-byte characters split across stream chunk boundaries
Replace String::from_utf8_lossy with append_utf8_safe in all four SSE
streaming paths. When a multi-byte UTF-8 character (e.g. Chinese, emoji)
is split across TCP chunk boundaries, from_utf8_lossy silently replaces
the incomplete halves with U+FFFD (�). This caused intermittent garbled
output in Claude Code when using the Copilot reverse proxy, because the
format conversion streams reconstruct SSE events from the corrupted buffer.
The new append_utf8_safe function preserves incomplete trailing bytes in
a remainder buffer and merges them with the next chunk before decoding,
ensuring characters are never split during UTF-8 conversion.
Fixes: intermittent U+FFFD replacement characters in Claude Code output
via Copilot proxy (not reproducible with direct Copilot connections like
opencode because they pass through raw bytes without format conversion).
* style: fix cargo fmt formatting in UTF-8 boundary tests
---------
Co-authored-by: Cod1ng <codingts@gmail.com>
Co-authored-by: encodets <encodets@gmail.com>
Some OpenAI-compatible chat providers reject requests when Claude-side\nsystem fragments arrive as multiple system messages. Normalize the\nconverted OpenAI chat payload so system content becomes a single\nleading system message while leaving the rest of the message stream\nunchanged.\n\nConstraint: Nvidia/Qwen-style chat completions require a single leading system prompt\nRejected: Reorder system messages only | still leaves fragmented system prompts for strict backends\nConfidence: high\nScope-risk: narrow\nReversibility: clean\nDirective: Keep OpenAI chat system prompts normalized unless a provider explicitly requires fragmented system messages\nTested: cargo test proxy::providers::transform --manifest-path src-tauri/Cargo.toml\nNot-tested: Full end-to-end proxy capture against Nvidia upstream in this session\nRelated: #1881
Jason
zerone0x
Dex Miller
v2v
Cod1ng