Add a "Grok Official" preset and seed (grokbuild-official) whose empty
config represents the official login state: no custom [model.*] tables
are written, so Grok CLI falls back to its built-in xAI OAuth login and
cc-switch never touches those credentials.
Backend:
- Seed entry in OFFICIAL_SEEDS plus ensure_grokbuild_official_provider
command for on-demand repair (the one-shot master seeding flag is
already set for existing databases).
- Split validation into syntax-only (empty allowed) for live reads,
writes and official snapshots, keeping the full custom-model shape
check for non-official provider writes and imports. Backup/restore
can now round-trip an official-state live file.
- Manual import (command layer only) recognizes an official-state live
config and imports it as the official entry set as current, matching
the Codex official-login import outcome. Startup auto-import keeps
rejecting official-state live so a deleted official entry is never
resurrected on launch: startup import only captures real user data
as "default" and never manufactures official entries.
- Manual import also ensures the official entry before importing
(claude-desktop precedent) and after a successful custom import, so
first-time users end up with default + official like other apps.
- Proxy takeover guards skip or reject official-state live configs in
all three takeover paths, consistent with the official-provider
takeover ban.
Frontend:
- Grok Official preset entry in the GrokBuild form: official category
hides connection fields and passes the raw config through untouched.
- Filter managed-OAuth presets out of the GrokBuild preset list; they
were never wired for this app and produced keyless broken configs.
Tests cover seed presence, official round-trip, ensure-after-deletion,
and the four import scenarios including startup non-resurrection.
Add a managed "xAI (Grok) OAuth" Codex provider that routes through the
local proxy to api.x.ai via the shared Grok CLI OAuth identity, plus the
native-Responses compatibility layer that makes Codex 0.142+ traffic work
against xAI's strict upstream serde parser.
Provider:
- codex.rs: recognize the xai_oauth placeholder in extract_auth, hard-pin
the base URL to api.x.ai and the tool profile to native Responses
- forwarder.rs: treat xAI OAuth auth failures as non-retryable
- presets + ProviderForm/CodexFormFields: managed OAuth preset that hides
the api key/endpoint fields and derives the provider type across apps
Native Responses compatibility (gated on is_xai_oauth, so no other
provider is affected):
- transform_codex_responses_namespace: flatten Codex's private
namespace/plugin tool declarations into top-level function tools on the
request; restore the flat function_call names back to {name, namespace}
on the response (streaming and non-streaming) so the client matches its
own namespaced tool registry
- transform_codex_responses_xai_sanitize: strip the OpenAI-backend-private
fields xAI rejects (external_web_access, prompt_cache_retention,
safety_identifier, the additional_tools carrier, tool_search, ...) with
deterministic removals that keep the prompt-cache prefix stable
- wire both into the native passthrough after the request transform;
response restore runs in a dedicated handler so the generic passthrough
hot path is untouched
Ports the proven approach of sub2api's Grok Responses gateway. Verified
with a 4-round codex -> xAI OAuth workload: all tasks green, zero upstream
errors.
Treat Grok like Claude/OpenCode: install via the official xAI installer
(POSIX shell / Windows PowerShell), keep npm as fallback, discover
~/.grok/bin, and anchor updates to `grok update` only for native installs.
Schema v16 wipes codex_session detail rows, _codex_session rollups and
Codex rollout cursors inside the migration savepoint (cursor deletion
uses pure shape matching so CODEX_HOME drift cannot orphan cursors);
the next session sync re-imports history from source JSONL under the
corrected importer. Fresh installs traverse the same branch as a no-op.
Add a manual "Rebuild Codex usage" maintenance action (single-flight,
hard-fail backup before reset, unconditional refresh notification even
when reimport is empty or fails after reset) with a destructive confirm
dialog, result toast and four-locale strings. Historical proxy-side
duplicate rows are intentionally left untouched; history whose source
JSONL was already deleted cannot be reconstructed.
Derive request ids from upstream envelope ids (Codex/OpenAI top-level
id, Gemini responseId, Claude message id with non-empty filtering)
scoped as session:{app_type}:{provider_id}:{id} for non-Claude sources;
Claude keeps bare session:{id} to preserve session-log convergence.
The logger now queries and conditionally writes under a single
connection guard: identical semantic replays return without writing or
notifying, session_log rows may be upgraded by proxy, and same-id
different-semantic responses land on a deterministic SHA-256 collision
fallback key instead of being overwritten. Fixes the random-UUID
INSERT OR REPLACE duplication behind #5496.
Rewrite the Codex session importer: identify forks by forked_from_id or
subagent thread_spawn.parent_thread_id (conflicts defer), anchor thread
identity to the rollout filename UUID, and align child token_count
signatures against the parent rollout as a strict prefix so replayed
parent history only establishes the cumulative baseline and is never
imported as child usage. Ambiguous alignments defer without advancing
the cursor and recover once the parent appears; parent signatures are
cached per (path, cutoff).
Fixes the 3x-20x double counting behind #5335/#5433 while keeping
reference-type subagents fully counted (#5381).
Detect codex_session rows sharing the same model/token fingerprint
within the dedup window under a different request_id. Predicates reuse
the COALESCE(data_source,'proxy') shape so the existing expression
index serves the lookup. Wired up by the Codex importer rewrite.
Replace per-insert usage notifications with a single notification per
sync pass, guard all sync entry points (startup backfill, 60s loop,
manual sync) behind a process-wide single-flight tokio mutex, and move
blocking file/DB work onto spawn_blocking with MissedTickBehavior::Skip.
Extend SessionSyncResult with deferredFiles/suspectedDuplicates
observability fields (saturating aggregation) and add a test seam that
counts notifications even without an injected AppHandle.
grok-4.5 is the default model of the xAI presets but had no pricing
row, so usage dashboards recorded zero cost for it. Official list
price: $2 input / $6 output / $0.50 cached read per 1M tokens.
Seed-only change: the startup INSERT OR IGNORE backfills existing
databases; no repair entry needed for a missing row.
Add an xAI OAuth manager using the OAuth 2.0 Device Authorization Grant
with endpoints resolved from xAI's OIDC discovery document. All HTTP goes
through the app-managed proxy client.
- Managed provider kind xai_oauth: forced openai_responses wire format,
pinned api.x.ai base URL, bearer injection gated to the xAI origin,
tokens registered for log redaction, single-auth-key takeover policy.
- Token cache cannot bypass account state: cache hits re-validate account
usability, refresh commits run under the mutation lock with a
refresh-token CAS check, and pending logins are re-checked before an
account is persisted.
- Refresh classification: 401/403 with any body and 400 with a non-JSON
body mark the account for re-auth; 429/5xx stay transient.
- Shared auth_* commands dispatch to xAI with guard types mirroring the
Copilot/Codex branches.
Codex >= 0.144.5 requires supports_reasoning_summaries in every external
model catalog entry and rejects the whole file at startup when it is
missing, which bricks both the CLI and the desktop app.
The ProxyChat catalog template is cloned from dynamic sources first
(models_cache.json, then `codex debug models --bundled`). models_cache.json
is shared by every Codex install on the machine and each version serializes
its own ModelInfo shape, so the cache's field set follows whichever process
wrote it last and cannot be assumed to satisfy the current external-catalog
schema. Deleting the generated catalog never helped: any provider save or
switch regenerated it from the same stale template.
Backfill parser-required fields (explicit whitelist, currently only
supports_reasoning_summaries) from the bundled static template when the
dynamic template lacks them. Optional capability fields are deliberately
NOT backfilled so their missing-means-parser-default semantics survive,
and existing values always win.
The Kimi For Coding subscription reports its K3 model under the short
id `k3` (no `kimi-` prefix), whereas the regular Moonshot API still uses
the full `kimi-k3`. Seed a `k3` alias row right after `kimi-k3` at the
same official list price ($3/M in, $15/M out, $0.30/M cache read) so
coding-plan usage is priced correctly. Mirrors the existing
hunyuan-hy3/hy3 bare-alias precedent.
Pricing is matched by exact model id, so this does not touch the
`kimi-for-coding` endpoint alias (intentionally unpriced, subscription
semantics) nor risk prefix mismatches. Seed-only, no schema bump.
Prefix anchored npm invocations with their sibling bin directory so env-based Node shebangs work under a minimal GUI PATH. Cover Codex repair commands and add a regression test.
Seed kimi-k3 at the official list price ($3/M input, $15/M output,
$0.30/M cache read; released 2026-07-16 with a 1M context window).
Cross-checked against the Moonshot pricing docs, OpenRouter, and
models.dev. Seed-only change: ensure_model_pricing_seeded picks it up
on next launch, no schema version bump needed.
Backend half of the logging overhaul.
Retention:
- Keep the last 4 rotated files at 20MB each and stop deleting logs on
startup, so a crash's prior-run logs survive a restart.
- Apply the persisted log level right after plugin registration instead
of at the end of setup.
- Bound crash.log with size-based rotation (5MB x 2).
Secret redaction on every log path:
- Proxy: strip userinfo/query from upstream URLs (keep path for
diagnostics), exact-match redact known auth.api_key/access_token,
classify request/response bodies instead of logging them, header
allowlist.
- Redact the Gemini `?key=` in cache-trace endpoints.
- Omit MCP custom-field values (headers may carry tokens).
- Redact deeplink and model-fetch URLs before logging.
Docs:
- FAQ points users to the persistent crash.log for support workflows.
Switching providers or toggling proxy takeover on Windows flashed a
transient cmd window and froze the UI for up to a couple of seconds.
Three fixes:
- Spawn `codex debug models --bundled` with CREATE_NO_WINDOW so the
console child of the GUI-subsystem app (npm's codex.cmd runs via
cmd.exe) no longer opens a visible window.
- Cache the ProxyChat model catalog template in a process-wide OnceCell:
only successful loads are cached, failures stay retryable, so the
Codex CLI starts at most once per app run instead of on every switch.
- Run switch_provider via spawn_blocking: sync Tauri commands execute on
the main thread, so the blocking catalog generation (and the
block_on'd per-app switch lock) froze the UI. Concurrency is already
serialized by ProxyService::lock_switch_for_app.
Normalize function tool parameters on the Codex Responses -> Chat Completions bridge: default null/missing parameters (direct and nested function forms) to {"type":"object","properties":{}}, coerce explicit type:null, and add a root type for top-level oneOf schemas so strict OpenAI-compatible upstreams (DeepSeek etc.) no longer reject built-in Codex tools like codex_app__automation_update. Also hardens the Codex -> Anthropic tool path with the same object-type guarantee and adds regression coverage for all reported shapes.
The tray menu used a hardcoded zh (Simplified Chinese) fallback whenever
settings.language was unset (i.e. first install). On systems whose UI language
resolves to Traditional Chinese / Japanese / English via the frontend's
navigator-based detection, the tray therefore showed Simplified Chinese until
the user manually switched language once.
Detect the OS locale via the sys-locale crate and map it to a supported tray
language, mirroring the frontend getInitialLanguage precedence
(zh-TW/HK/MO/Hant -> zh-TW, other zh -> zh, ja -> ja, en -> en, otherwise zh).
This keeps the tray consistent with the UI from the first launch with no timing
window. An explicitly stored settings.language still takes precedence, so user
choices are never overridden. Adds unit tests for the locale mapping.
Co-authored-by: LaiYueTing <LaiYueTing@users.noreply.github.com>
Co-authored-by: Jason <farion1231@gmail.com>
* fix(ci): run backend checks on Windows/macOS and repair platform-gated tests
The backend CI job ran only on ubuntu-22.04, so every #[cfg(target_os =
"windows")] / macOS path — tests and clippy lints alike — was never
compiled or run. As a result main shipped 8 failing Windows unit tests and
a hard `cargo clippy -- -D warnings` failure on Windows, all invisible to
CI because it only builds and tests Linux.
Changes:
- ci: run the backend job on a {ubuntu-22.04, windows-latest, macos-latest}
matrix (fail-fast: false); gate the apt install step on runner.os ==
'Linux' and use `shell: bash` for the dist placeholder so it works on all
three runners.
- misc.rs: fix 6 stale `anchored_upgrade_windows` tests. Commit 5092fe51
removed codex from `prefers_official_update`, so codex now anchors to the
package manager with no `codex update ||` prefix; update the five
package-manager expectations accordingly, and retarget the no-sibling
"official-update-without-fallback" test to `claude` (still in the list) so
that branch stays covered instead of being silently dropped.
- codex_state_db.rs / codex_history_migration.rs: the two sqlite_home tests
built TOML basic strings from Windows paths, whose backslashes are invalid
escape sequences and made the override fail to parse; switch to TOML
literal (single-quoted) strings so the path is taken verbatim.
- clippy (13 Windows-only warnings): move `use std::io::Write` into the
#[cfg(unix)] block that actually uses it; convert 3 unneeded `return`s in
Windows-gated tail positions to expressions; mark 9 POSIX-shell helpers
#[cfg_attr(windows, allow(dead_code))] since they are used only by the
macOS/Linux terminal launchers.
Verified locally (Windows 11, Rust 1.95.0):
cargo test --lib → 1748 passed; 0 failed (was 1740/8)
cargo clippy -- -D warnings → clean (was 13 errors)
cargo fmt --check → clean
Co-Authored-By: Claude <noreply@anthropic.com>
* fix(skills): resolve home via get_home_dir so tests isolate on Windows
services/skill.rs called dirs::home_dir() directly in five places. On
Windows dirs::home_dir() resolves through the Known Folder API and
ignores HOME/USERPROFILE, so the CC_SWITCH_TEST_HOME override set by the
test harness never applied: the skill_sync integration tests scanned the
runner's real user profile, found no skills, and failed (the first panic
then poisoned the shared test mutex, cascading to all seven). On Unix
the harness also sets HOME, which dirs::home_dir() honors, so the suite
passed there by accident.
Route all five call sites through crate::config::get_home_dir(), the
crate-wide home resolver that prefers CC_SWITCH_TEST_HOME. The three
now-unreachable GET_HOME_DIR_FAILED error branches are dropped:
get_home_dir() is infallible, matching every other config-dir resolver.
Production behavior is unchanged (CC_SWITCH_TEST_HOME is unset outside
tests, so get_home_dir falls through to dirs::home_dir).
Add a regression test that discriminates on every platform by pointing
CC_SWITCH_TEST_HOME at a temp dir different from $HOME; it is marked
#[serial_test::serial] to stay mutually exclusive with the other tests
mutating the same process-global variable. Verified the test fails
against the old code on macOS with the same failure mode as Windows CI.
---------
Co-authored-by: Claude <noreply@anthropic.com>
Co-authored-by: Jason <farion1231@gmail.com>
Keep non-empty call IDs across continuation deltas and release parallel tool calls in Chat index order when identity fields arrive late. Preserve valid sparse and later calls during finalization.
* fix: normalize function parameters type to "object" for strict OpenAI-compatible providers
Some Responses tools carry parameters with `type: null` (e.g.
codex_app__automation_update), causing HTTP 400 from strict
OpenAI-compatible providers like DeepSeek that require
`{"type": "object", "properties": {...}}`.
This adds normalize_function_parameters() to ensure the type
field is always "object" in both branches of
responses_function_tool_to_chat_tool.
Closes#4705
* style: fix cargo fmt issues
* fix: handle null function parameters
The forced 1-hour cache_control TTL (schema v14) was a mistake. Injected
breakpoints return to Anthropic's standard 5-minute TTL, caller-owned
markers are preserved verbatim instead of having their TTLs rewritten,
and the 5m/1h cache-write buckets are removed from usage parsing and
pricing (back to the single aggregate cache-creation rate). The cache
TTL selector is removed from the rectifier settings panel along with its
i18n keys in all four locales.
SCHEMA_VERSION returns to 13: the unreleased cache_creation_1h_tokens
column and the v13->v14 migration are removed. The feature never shipped
in a release and the introducing commit was never pushed, so no
databases were stamped v14 outside this machine (local DB verified at
user_version 13).
Mapped GPT models were rejected by Codex clients with "model does not
support image inputs". Two root causes:
- Catalog entries for native-Responses/Anthropic providers cloned a
template whose input_modalities defaulted to ["text"], so every mapped
model was advertised text-only. model_catalog_json replaces Codex's
built-in model table wholesale, and both the TUI and the IDE extension
block images pre-send when the current model is found without "image".
- Editing the current Codex provider during proxy takeover only refreshed
the DB backup, so removing the mapping left a stale model_catalog_json
pointer (and its text-only catalog file) active in live config.
Changes:
- New shared model_capabilities module: explicit row declaration first,
then a confirmed text-only registry (exact tail match only — prefix
matching removed, variants enumerated so future -vl/-vision models fail
open), everything else unknown.
- Catalog generation writes input_modalities from that inference for all
tool profiles: unknown models fail open to ["text","image"]; only
confirmed text-only models are advertised as ["text"], giving users a
clear client-side prompt instead of silent image stripping.
- Live catalog reverse-import collapses modalities that match current
inference, so registry corrections are not frozen into hidden row
overrides and the rectifier's heuristic opt-out keeps working.
- Saving the current Codex provider while takeover owns live now
re-projects the live config (mirrors the hot-switch path), so mapping
edits and removals take effect immediately.
- Media rectifier delegates to the shared module; its preflight toggle is
documented (4 locales) as proxy-request-only, never affecting catalog
capability declarations.
ChatGPT's Codex backend routes model availability by the originator and version header pair. Requests identifying as cc-switch without a version were assigned to a cohort where gpt-5.6-luna resolved to an unavailable internal engine, resulting in a misleading 404 Model not found response.
Identify takeover requests as codex_cli_rs and send version 0.144.1, satisfying luna's minimal_client_version requirement of 0.144.0. A direct HTTP A/B test confirmed the existing headers returned 404 while the aligned identity completed successfully, so no WebSocket transport workaround is required.
Claude Code caps unknown model ids at 200K, so the preset's standalone
CLAUDE_CODE_AUTO_COMPACT_WINDOW=262144 was clamped by min(window, value)
and never took effect. Pair it with CLAUDE_CODE_MAX_CONTEXT_TOKENS and
route the endpoint's kimi-for-coding alias on every tier, since both
context envs are ignored for claude-* prefixed model ids.
Saved providers get the same context defaults injected into effective
live settings at switch time, with a mirror-inverse strip on backfill so
injected values never harden into provider config. Explicit user values
always win over the injected defaults.
- Add the bare gpt-5.6 row (official Sol alias) plus effort-suffix variants
mirroring the gpt-5.5 accounting shape, priced at Sol rates
- The GPT-5.6 family charges cache writes at 1.25x the uncached input rate
(new for OpenAI; earlier GPT models keep their correct zero): set
6.25 / 3.125 / 1.25 for Sol / Terra / Luna across seed rows
- Repair earlier zero-value gpt-5.6 tier seeds in existing databases,
matching only untouched rows so user-customized prices survive
- Update Codex OAuth presets to the gpt-5.6 family (haiku -> gpt-5.6-luna)
and bump the custom Codex template default model
- Inject CLAUDE_CODE_MAX_CONTEXT_TOKENS / CLAUDE_CODE_AUTO_COMPACT_WINDOW
(372000, the ChatGPT Codex catalog window with a ~353K effective budget,
openai/codex#31860) into effective live settings so Claude Code stops
assuming a 200K window and compacts before the upstream rejects the prompt
- Gate the injected defaults on every configured model being gpt-5.6*:
gpt-5.5's upstream catalog oscillates between 272K and 372K and must not
inherit them; explicit user values always win
- Strip the injected values on switch-away backfill (mirror-inverse of the
injection conditions) so program defaults never harden into per-provider
explicit values, and keep both keys out of the shared common-config snippet
Let users switch between the built-in OpenAI provider and third-party Codex providers directly from the provider panel while takeover mode remains active.
Centralize the frontend capability predicate so only the fixed codex-official seed receives native-login routing support, and keep copied UUID-based official entries clearly marked as unsupported.
Add an idempotent backend command that recreates the deleted official seed, wire it into the add-provider flow, refresh localized guidance, and add mutation and provider-action regression coverage.
Do not derive unauthenticated health-check targets from runtime adapter defaults for official providers. Batch checks now skip official entries, and direct base-URL resolution fails explicitly instead of probing first-party endpoints without credentials.
Cover the Codex official provider in the stream-check regression tests so future adapter changes cannot silently reintroduce these probes.
Allow the built-in Codex official provider to participate in takeover mode while preserving Codex's native OAuth or API-key credentials instead of persisting them into provider records.
Project official routing into a dedicated TOML provider, normalize inline tables, clean stale managed placeholders, and fail closed when the live configuration cannot be transformed safely.
Validate forwarded authorization, make official 401/403 responses non-retryable, avoid circuit-breaker pollution, and share the first-party ChatGPT endpoint across the Codex and Claude adapters.
Parse and retain Anthropic's ephemeral 5-minute and 1-hour cache-creation token buckets while preserving the existing aggregate cache-write metric for compatibility.
Price 1-hour writes at the documented premium relative to the configured 5-minute write rate, clamp inconsistent provider details safely, and include TTL buckets in usage diagnostics.
Persist 1-hour cache-write tokens with schema version 14 so zero-cost backfills and later pricing updates retain the original TTL semantics. Keep session import paths compatible through zero-valued detail fields.
Honor both the optimizer master switch and the cache-injection sub-switch before mutating native optimizer requests.
Use the available four-breakpoint budget across tools, system content, the latest cacheable message, and an older user anchor for long tool-heavy conversations.
Preserve caller-owned breakpoint limits, avoid thinking blocks as cache targets, normalize configured TTLs, and add regression coverage for disabled optimization and long histories.
Fail closed on HTTP 2xx failure envelopes and pre-output SSE failures so semantic upstream errors can trigger failover instead of becoming empty successful replies.
Finalize incomplete and truncated streams explicitly, handle clean EOF and whole JSON responses, and keep tool-call stop reasons and terminal event ordering consistent.
Preserve structured tool results, URL images, documents, system roles, and signed thinking across both conversion directions. Drop incomplete historical tool calls safely and classify malformed completed arguments as non-retryable client requests.
Keep Codex-to-Anthropic prompt caching enabled by default while honoring the dedicated cache-injection switch.
Codex always sends prompt_cache_key in its Responses requests, but the
Responses -> Chat Completions conversion dropped it, breaking session
cache affinity on upstreams that route by key (e.g. Kimi Coding).
- Re-inject prompt_cache_key after conversion in the forwarder: an
explicit client key wins, otherwise a client-provided session ID;
generated per-request UUIDs are never sent upstream.
- Provider-aware gating: "auto" enables only known-compatible upstreams
(api.openai.com, api.kimi.com/coding) because strict gateways reject
unknown fields with HTTP 400 (e.g. Fireworks); an advanced
Auto/Enabled/Disabled override is available on the Codex form in all
four locales.
- Kimi For Coding preset opts in explicitly.
Warn when caller-provided cache breakpoints already exceed the supported total of four while preserving the original markers and upgrading their TTLs.
Clarify that automatic injection is governed by the remaining breakpoint budget, and add regression coverage proving excess caller markers are never deleted or reordered.
Round-trip encrypted Responses reasoning items through bridge-owned Anthropic thinking blocks, discard orphaned reasoning-only history, and consume the official reasoning text event vocabulary.
Track concurrent streaming items by stable IDs and output indexes, reuse a dedicated fallback block for keyless legacy reasoning, recover tool arguments from done events, and close blocks in protocol order.
Normalize empty or incomplete non-streaming tool arguments, reject malformed completed calls, and persist upstream usage before returning terminal conversion errors.
Parse cache_write_tokens from OpenAI usage details and preserve cache creation data across Chat, Responses, and Anthropic conversion paths.
Add explicit input-token semantics to request logs and rollups so legacy rows subtract cache reads only while new total-inclusive rows subtract both cache reads and writes. Migrate v12 databases, normalize rollups to fresh input, and cover historical backfill behavior with regression tests.
Add a show/hide toggle for the header project profile switcher under
the Homepage Display section in settings. Defaults to visible so
existing users keep the current behavior.
Add Hunyuan Hy3 (released 2026-07-06, 256K context) to seed_model_pricing
under both the hunyuan-hy3 and hy3 ids, since the upstream billing id isn't
yet confirmed and one of the two should match logged usage.
Prices come from Tencent's launch-day list price (CNY 1 / 4 / 0.25 per 1M
input / output / cache-hit), converted at 1 USD ~= 7.14: 0.14 / 0.56 / 0.035.
cache_write is "0" (no published write rate). Hy3 actually uses input-length
tiered billing (<16K / 16-32K / >=32K); the single-price table holds the
lowest tier, so long-context requests are under-billed until this is revisited
against the official billing page. New model, so seed only.
Add the three GPT-5.6 tiers to seed_model_pricing, cross-checked against
OpenAI's official pricing page and OpenRouter:
gpt-5.6-sol 5 / 30 / 0.50 (cache read)
gpt-5.6-terra 2.50 / 15 / 0.25
gpt-5.6-luna 1 / 6 / 0.10
cache_write is "0": OpenAI publishes only a cached-input (read) rate and no
separate cache-write price, matching the gpt-5.x family convention. Base ids
only for now — effort-suffix variants are deferred until the upstream logging
suffix set for 5.6 is confirmed. New model, so seed only (no repair branch,
no SCHEMA_VERSION bump).
Switching the Claude provider from a third-party (DeepSeek/MiMo/...) to a
managed Codex provider wrote both ANTHROPIC_API_KEY and
ANTHROPIC_AUTH_TOKEN as PROXY_MANAGED into ~/.claude/settings.json,
making Claude Code warn "Both ANTHROPIC_AUTH_TOKEN and ANTHROPIC_API_KEY
set - auth may not work as expected" on every run.
The double key is an accretion artifact: the ManagedAccount takeover
policy originally injected only ANTHROPIC_API_KEY (Copilot, #1049), and
the Codex fix that preserved ANTHROPIC_AUTH_TOKEN to avoid the login
prompt (#3784 / PR #3789) added the token on top without removing the
API key insert.
Inject exactly one placeholder instead: AUTH_TOKEN for Codex-managed
providers (keeps the #3784 fix), API_KEY for Copilot (keeps the #1049
behavior). Local-proxy auth is unaffected - the forwarder resolves the
PROXY_MANAGED placeholder from either header.
Updates the three Codex takeover tests to assert API_KEY is absent and
adds a regression test for the exact third-party -> Codex switch from
the report.
Fixes#4919
* Add test for Fable model env key exclusion
Add regression test to ensure Fable model env keys are excluded from common config.
* Add Fable model to CLAUDE_MODEL_OVERRIDE_ENV_KEYS
* Add files via upload
* Delete .github/workflows/build-windows-unsigned.yml
* feat(codex): support native Anthropic Messages protocol as upstream
Allow gateways that only expose the native Anthropic Messages protocol
(/v1/messages) to be used by Codex: the local proxy performs bidirectional
request/response/streaming conversion between Responses and Anthropic.
Backend:
- Add two conversion modules: transform_codex_anthropic / streaming_codex_anthropic
- codex.rs: add routing detection and auth: ANTHROPIC_AUTH_TOKEN→Bearer (default),
ANTHROPIC_API_KEY→x-api-key, mutually exclusive
- handlers.rs: add handle_codex_anthropic_to_responses_transform
- forwarder.rs: support optional Claude Code client fingerprint impersonation
(User-Agent / anthropic-beta / x-app / system prompt first-line injection)
and /responses→/v1/messages rewriting
- codex_config: the anthropic format reuses the NativeResponses profile to strip
custom tools
- ProviderMeta: add impersonateClaudeCode
Frontend:
- CodexApiFormat: add "anthropic"; the form adds auth field selection and an
impersonation toggle
- Add en/ja/zh/zh-TW copy
Robustness:
- Downgrade when tool history / forced tool_choice conflicts with extended
thinking, avoiding upstream 400s
- Emit cache_creation_input_tokens in usage and use saturating_add to guard
against overflow
- Append a unique suffix to non-streaming/streaming output-item ids to avoid
multi-segment text/thinking overwriting each other
* fix(codex): harden Anthropic bridge against empty text blocks and truncated streams
- Drop empty/whitespace-only text content blocks when rebuilding Anthropic
messages from Responses history; Anthropic 400s on empty text blocks (e.g. an
empty assistant text emitted alongside a tool_use), which broke follow-up and
tool-result requests. Also drop messages left without content.
- Do not report a truncated Anthropic stream as completed: when the SSE
connection ends before message_stop with no stop_reason, emit an incomplete
response if partial output exists, or a failed (stream_truncated) response
otherwise, mirroring the chat converter's EOF handling.
* fix(codex): resolve Anthropic-bridge review findings
Blocking:
1. Defer stripping the [1m] long-context marker until after catalog matching and model write-back, re-stripping it on the final Codex→Anthropic body and setting a flag to emit the context-1m-2025-08-07 beta header, so the marker is no longer lost or overridden by the default model.
2. Gate Anthropic thinking on the trailing turn only (via trailing_turn_allows_thinking) instead of scanning full history, so a Codex session that resends history each round no longer permanently loses thinking after the first tool call.
3. Inject 5m ephemeral cache_control on the Codex→Anthropic body by reusing cache_injector (handling the system string→array conversion), so system/tools/history are cached instead of re-sent at full price every round.
4. Add a shared base_url_is_full_endpoint helper (normalizing whitespace/query/fragment/trailing slash) used by both the Anthropic and Chat paths, so a base URL already ending in /v1/messages is treated as a full endpoint instead of double-appending to /v1/messages/v1/messages.
5. Align the catalog tool-profile predicate with the routing predicate so resolve_codex_catalog_tool_profile returns the Anthropic profile whenever the request converts to Anthropic, preventing freeform tools like apply_patch from being silently filtered by a ProxyChat catalog.
6. Explicitly disable native web_search for the Anthropic profile (including the no-catalog branch) via set_codex_native_web_search_field, so Codex no longer treats it as available while the transform silently drops it.
7. Only forward tool_choice when tools survive filtering, dropping it otherwise, to avoid a non-retryable 400 from upstream when tool_choice is sent with no tools.
8. Lower the fallback default to max_tokens=8192 (only when max_output_tokens is omitted) and clamp the thinking budget to max_tokens/2, disabling thinking below the 1024 floor, to avoid hard 400s on low-output-ceiling models/gateways.
Minor:
9. Centralize the Codex/OpenAI fingerprint-header denylist in is_codex_client_fingerprint_header so impersonating Claude Code uniformly drops originator/session_id/conversation_id/chatgpt-account-id/x-client-request-id/openai-* and the x-stainless-*/x-codex-* prefixes.
10. Retain content_block_start.input as start_input and fall back to it at block close when no input_json_delta arrived, so a gateway that carries the full tool input on the start event no longer yields empty tool arguments.
11. Extract a shared codex_responses_sse module as the single Responses SSE envelope builder that both the chat and anthropic streaming emitters delegate to, with byte-for-byte-unchanged wire output, so future event-format fixes touch one place instead of two.
* fix(codex): add per-provider max_output_tokens override for Codex→Anthropic path
Codex does not forward model_max_output_tokens in the request body,
causing the proxy to fall back to a conservative 8192 default. This
truncates long or thinking-heavy responses (stop_reason=max_tokens).
- Add maxOutputTokens field to ProviderMeta (Rust + TypeScript)
- Inject the value into the Anthropic request body before transform,
taking precedence over request-supplied and default values
- Add numeric input in Codex form fields (Anthropic format only)
- Add i18n entries for label, placeholder, and hint (en/zh/zh-TW/ja)
- Include roundtrip and omission unit tests for the new field
* fix(codex): harden Anthropic protocol bridge
* fix(codex): address Anthropic bridge review
* fix(codex): preserve flattened Anthropic inputs
* fix(codex): harden Anthropic recovery paths
* fix(ci): satisfy Rust clippy
---------
Co-authored-by: Jason <farion1231@gmail.com>
Jason