feat: add Kimi Code provider with OAuth device flow authentication by soilSpoon · Pull Request #8466 · aaif-goose/goose

soilSpoon · 2026-04-10T10:15:57Z

Summary

Add a Kimi Code provider with OAuth 2.0 Device Authorization Grant (RFC 8628) authentication.

The existing kimi.json declarative provider was broken — it pointed to api.moonshot.cn (Moonshot's direct API) with a MOONSHOT_API_KEY, rather than the Kimi Code platform (api.kimi.com/coding) which uses OAuth. This PR replaces it with a hand-coded provider that works correctly.

How authentication works:

First use: goose configure → select "Kimi Code" → browser opens + user code copied to clipboard → user approves on auth.kimi.com → token saved to ~/.config/goose/kimicode/token.json (mode 0o600)
Subsequent requests: access token reused from memory/disk cache, auto-refreshed 5 min before expiry via refresh_token
Token fully expired: device flow restarts automatically

Follows the same patterns as GithubCopilotProvider (device code flow) and AnthropicProvider (Anthropic-compatible SSE streaming).

Key implementation details:

OAuth host: https://auth.kimi.com, Client ID: 17e5f671-d194-4dfb-9706-5516cb48c098 (public client ID, same as kimi-cli and oh-my-pi)
RFC 8628 slow_down error handled — polling interval increases by 5 s per server request
Supports kimi-k2.5 and kimi-k2-thinking models
Reuses the provider's reqwest::Client for API calls (no per-request client construction)

Testing

Manual testing plan:

goose configure → select "Kimi Code" → complete OAuth device flow in browser
Verify ~/.config/goose/kimicode/token.json exists with 0o600 permissions
Run a session with kimi-k2.5 and confirm streaming responses work
Verify token refresh: manually set expires_at to a past time in the token file and confirm auto-refresh triggers

Related Issues

Relates to #7150
Discussion: (none)

Replace the declarative kimi.json (which incorrectly pointed to api.moonshot.cn with a Moonshot API key) with a hand-coded provider that authenticates via OAuth 2.0 Device Authorization Grant (RFC 8628) against auth.kimi.com, and calls the Anthropic-compatible API at api.kimi.com/coding. - Device flow login with browser open + clipboard copy of user code - Automatic token refresh (5 min before expiry) - Token persisted to ~/.config/goose/kimicode/token.json (0o600) - Persistent device ID in ~/.config/goose/kimicode/device_id - RFC 8628 slow_down handling (interval backoff) - Supports kimi-k2.5 and kimi-k2-thinking models Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: ccdfbfdb94

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

chatgpt-codex-connector · 2026-04-10T10:20:48Z

+                .error_for_status()
+                .context("token poll request failed")?


Parse device-code pending errors before status validation

poll_for_token calls error_for_status() before deserializing the OAuth payload, so any non-2xx token response exits early and never reaches the authorization_pending/slow_down handling below. In OAuth device flow, pending authorization is commonly returned as an error payload, so this can abort goose configure before the user finishes browser approval instead of continuing to poll.

Useful? React with 👍 / 👎.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> Signed-off-by: DaeHee Lee <lee111dae11@proton.me>

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 08ce154123

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

chatgpt-codex-connector · 2026-04-12T09:21:24Z

        registry.register::<GeminiOAuthProvider>(true);
        registry.register::<GithubCopilotProvider>(false);
        registry.register::<GoogleProvider>(true);
+        registry.register::<KimiCodeProvider>(true);


Preserve backward-compatible alias for kimi provider

This registers only kimi_code, while this same commit removes the prior declarative provider (declarative/kimi.json had name: "kimi"), so existing configs that still have GOOSE_PROVIDER: kimi now fail with Unknown provider when sessions are created. Please add a compatibility alias or migration path so upgrades do not break previously working user configurations.

Useful? React with 👍 / 👎.

chatgpt-codex-connector · 2026-04-12T09:21:24Z

+            if let Ok(refreshed) = self.do_refresh_token(&token.refresh_token.clone()).await {
+                self.token_cache.save(&refreshed).await?;
+                let access = refreshed.access_token.clone();
+                *guard = Some(refreshed);
+                return Ok(access);
+            }


Fall back to current token when refresh fails

When a cached token is within the refresh threshold but still unexpired, any refresh error immediately falls through to full device-code login instead of using the still-valid access token for the current request. That means transient token-endpoint outages can trigger unnecessary interactive auth (or outright failure in headless environments) even though the existing token would still work.

Useful? React with 👍 / 👎.

- Remove error_for_status() before parsing OAuth device flow poll response, so authorization_pending/slow_down errors are handled by the polling logic instead of aborting early - Fall back to still-valid token when refresh fails instead of triggering unnecessary device flow re-authentication Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> Signed-off-by: DaeHee Lee <lee111dae11@proton.me>

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: a56ae0fcf5

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

chatgpt-codex-connector · 2026-04-12T09:47:29Z

+        if let Some(token) = self.token_cache.load().await {
+            if let Ok(refreshed) = self.do_refresh_token(&token.refresh_token).await {
+                self.token_cache.save(&refreshed).await.map_err(|e| {
+                    ProviderError::ExecutionError(format!("Failed to save token: {}", e))
+                })?;


Fall back to existing token when refresh fails in configure

When configure_oauth finds a cached token, it only handles the refresh-success path and otherwise immediately drops into full device-code login. If the refresh endpoint is temporarily unavailable (or returns an error) but the cached access token is still unexpired, goose configure unnecessarily forces interactive re-auth and can fail in headless/offline contexts even though requests would still work with the current token. This should mirror get_access_token and reuse the still-valid token before starting device flow.

Useful? React with 👍 / 👎.

RFC 8628 device authorization grant requires application/x-www-form-urlencoded, not JSON. Fixes device flow login failing with 400 error. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> Signed-off-by: DaeHee Lee <lee111dae11@proton.me>

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 4a69f6f7e1

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

chatgpt-codex-connector · 2026-04-12T10:13:28Z

+        self.token_cache
+            .save(&token)
+            .await
+            .map_err(|e| ProviderError::ExecutionError(format!("Failed to save token: {}", e)))?;
+        *self.cached_token.lock().await = Some(token);


Record OAuth configured state after successful Kimi login

configure_oauth only updates the local token cache and in-memory token, then returns success, but it never persists any config-visible signal (e.g., KIMI_CODE_TOKEN or a kimi_code_configured marker). In the server config path, check_provider_configured treats OAuth providers as configured only when that marker (or required key) exists, so Kimi can still be reported as unconfigured and /config/providers/{name}/models can reject it after a successful login done outside the server OAuth endpoint. Persisting configured state here would keep CLI/server behavior consistent.

Useful? React with 👍 / 👎.

chatgpt-codex-connector · 2026-04-12T10:13:28Z

+        let mut effective_interval = interval_secs;
+        for attempt in 0..MAX_POLL_ATTEMPTS {
+            tokio::time::sleep(tokio::time::Duration::from_secs(effective_interval)).await;


Poll device flow until server expiry instead of fixed attempts

The polling loop is capped by a hardcoded attempt count (MAX_POLL_ATTEMPTS), which limits auth to roughly 5 minutes at the default interval and can terminate early even when the device code is still valid. Device-code flows are intended to run until the server-declared expiration window closes, so users who need longer approval time (SSO/MFA/device switching) can be forced into unnecessary re-login cycles. Use the device authorization expires_in to compute an absolute deadline and poll until that deadline instead of a fixed attempt cap.

Useful? React with 👍 / 👎.

- configure_oauth: fall back to still-valid token when refresh fails instead of triggering unnecessary device flow re-auth - poll_for_token: use server-provided expires_in deadline instead of fixed MAX_POLL_ATTEMPTS, giving users more time for SSO/MFA approval Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> Signed-off-by: DaeHee Lee <lee111dae11@proton.me>

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 072be3650e

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

chatgpt-codex-connector · 2026-04-12T13:39:18Z

+        let mut headers = reqwest::header::HeaderMap::new();
+        headers.insert("X-Msh-Platform", KIMI_MSH_PLATFORM.parse().unwrap());
+        headers.insert("X-Msh-Version", KIMI_MSH_VERSION.parse().unwrap());
+        headers.insert("X-Msh-Device-Id", self.device_id.parse().unwrap());


Validate device ID header value before unwrapping

get_or_create_device_id accepts any non-empty contents from ~/.config/goose/kimicode/device_id, but kimi_headers later does self.device_id.parse().unwrap(). If that file is corrupted or manually edited to include an invalid header value (e.g., control/non-ASCII characters), provider initialization succeeds and every auth/API call panics at runtime instead of returning a recoverable error. Please validate/regenerate invalid IDs or propagate a normal error when building headers.

Useful? React with 👍 / 👎.

jh-block · 2026-04-15T10:06:14Z

Re-assigned to myself since the associated issue is assigned to me

jh-block

just a couple of minor issues. there's also quite a bit of duplication between get_access_token and configure_oauth. maybe configure_oauth can call get_access_token, or a shared helper can be extracted?

eventually we might want to extract a common oauth device flow helper that multiple providers can use, but no need to do that as part of this PR.

jh-block · 2026-04-15T10:10:08Z

+                })
+                .send()
+                .await
+                .context("failed to poll for token")?


need a raise_for_status() here to handle possible HTTP errors

Addressed in commit a5113f5, but not as a literal error_for_status() — the OAuth device flow (RFC 8628) returns authorization_pending and slow_down as 4xx responses with a JSON body, so raising on any non-2xx would break polling entirely (that was the original Codex P1 on the first commit).

The new code in poll_for_token reads the body first, attempts to parse it as PollResp, and only surfaces the HTTP status when parsing fails and the status is non-success. That preserves RFC 8628 semantics while still giving full status + body context on real transport errors.

Covered by:

poll_for_token_handles_authorization_pending_then_success (400 + pending body)

poll_for_token_surfaces_http_error_on_unparseable_body (502 + unparseable body)

jh-block · 2026-04-15T10:11:10Z

+        let known: Vec<String> = KIMI_CODE_KNOWN_MODELS
+            .iter()
+            .map(|s| s.to_string())
+            .collect();


this test isn't actually testing anything; it's just testing that a constant contains the values that it contains

Agreed — removed in commit a5113f5 and replaced with wiremock-backed tests that exercise real refresh/poll behavior:

use_or_refresh_returns_fresh_token_without_calling_endpoint

use_or_refresh_falls_back_to_existing_token_when_refresh_fails

use_or_refresh_returns_new_token_on_successful_refresh

use_or_refresh_preserves_refresh_token_when_server_omits_it

poll_for_token_handles_authorization_pending_then_success

poll_for_token_accepts_response_without_refresh_token

poll_for_token_surfaces_http_error_on_unparseable_body

kimi_headers_skips_invalid_device_id_without_panic

validates_device_id_rejects_invalid_bytes

… tests Address PR aaif-goose#8466 review feedback: - Extract `ensure_token` + `use_or_refresh` so `get_access_token` and `configure_oauth` share the cache → refresh → device-flow ladder. - Validate device_id against `HeaderValue::from_str` and regenerate when the on-disk file is corrupted; `kimi_headers` no longer panics. - Persist `kimi_code_configured = true` on successful OAuth so the server's `check_provider_configured` reports the provider as ready. - In `poll_for_token`, parse the body before status checks (preserves RFC 8628 `authorization_pending` / `slow_down` semantics) but surface the HTTP status when the body is unparseable. - Drop the trivial constants test and replace with wiremock-based coverage for refresh fallback, refresh success, pending-then-success polling, invalid device_id headers, and HTTP error surfacing. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> Signed-off-by: DaeHee Lee <lee111dae11@proton.me>

Follow-ups to the OAuth refactor (aaif-goose#8466): - Name the previously hardcoded fallbacks: DEFAULT_TOKEN_LIFETIME_SECS, DEFAULT_DEVICE_CODE_LIFETIME_SECS, DEFAULT_POLL_INTERVAL_SECS, SLOW_DOWN_BACKOFF_SECS. - Warn when the cached token.json fails to deserialize instead of silently dropping through to device flow. - Move the interactive "Please visit <url>" prompt to stderr so that stdout-parsing CLI workflows are not disturbed. - Emit tracing::info for a new OAuth device-flow login, and tracing::debug for refresh success / failure / fallback — but only for outcome-bearing events. The procedural "trying cache", "token is fresh" lines were noise and have been dropped. - Document the KIMI_CODE_TOKEN ConfigKey as a marker (the actual token is cached on disk), and clarify the kimi_headers fallback comment. Signed-off-by: DaeHee Lee <lee111dae11@proton.me>

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: b155ae3df6

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

chatgpt-codex-connector · 2026-04-16T09:10:26Z

+            if let (Some(access_token), Some(refresh_token)) =
+                (resp.access_token, resp.refresh_token)
+            {


Accept device-flow tokens without refresh_token

poll_for_token only returns success when both access_token and refresh_token are present, but OAuth token responses may legitimately omit refresh_token; in that case this code keeps polling until timeout even though authentication already succeeded. This makes goose configure fail for providers/environments that issue access-only token responses.

Useful? React with 👍 / 👎.

Fixed in commit 3b04aa8. poll_for_token now succeeds on just access_token and defaults refresh_token to empty when the server omits it, per RFC 6749. Covered by new test poll_for_token_accepts_response_without_refresh_token.

chatgpt-codex-connector · 2026-04-16T09:10:26Z

+        struct RefreshResp {
+            access_token: String,
+            refresh_token: String,
+            expires_in: Option<i64>,


Keep prior refresh token when refresh response omits it

do_refresh_token requires refresh_token in the refresh response payload, but OAuth refresh responses are allowed to return only a new access_token. When that happens deserialization fails, use_or_refresh treats refresh as failed, and once the current token expires users are forced back through interactive device auth instead of continuing with the existing refresh token.

Useful? React with 👍 / 👎.

Fixed in commit 3b04aa8. RefreshResp.refresh_token is now Option<String>, and do_refresh_token falls back to reusing the prior refresh_token when the server omits one, per RFC 6749 §6. Covered by new test use_or_refresh_preserves_refresh_token_when_server_omits_it.

Codex P2 follow-ups on aaif-goose#8466: - `poll_for_token` now succeeds on just `access_token`, defaulting `refresh_token` to empty if the server omits it. - `do_refresh_token` now deserializes `refresh_token` as Option<String> and falls back to reusing the prior refresh_token when the server omits it per RFC 6749 §6. - Added wiremock coverage for both paths. Signed-off-by: DaeHee Lee <lee111dae11@proton.me>

soilSpoon · 2026-04-16T09:32:38Z

@jh-block Thanks for the review. Quick status update — all inline feedback is addressed:

Duplication between get_access_token and configure_oauth — extracted ensure_token (the cache → refresh → device-flow ladder) and use_or_refresh (per-token freshness/refresh/fallback) in commit a5113f5. Both entry points now go through ensure_token, so the duplicated branching is gone.
raise_for_status() on the poll request — replied inline. Short version: a plain error_for_status() would break RFC 8628, so I read the body first and only surface the HTTP status when parsing fails AND status is non-2xx. Covered by poll_for_token_surfaces_http_error_on_unparseable_body and poll_for_token_handles_authorization_pending_then_success.
Trivial constants-only test — removed and replaced with 9 wiremock-backed tests.

Also addressed during this round:

Device-code panic on corrupted device_id file → validated at load and regenerated.
Missing kimi_code_configured marker → persisted on successful configure_oauth so the server's check_provider_configured reports the provider as ready.
Refresh-failure fallback → reuses the still-unexpired token instead of forcing device flow.
Hardcoded attempt cap → replaced with a deadline derived from expires_in.
poll_for_token no longer requires refresh_token in the response, and do_refresh_token reuses the prior refresh_token when the server omits one (both per RFC 6749; 3b04aa8).
Named constants for fallback timeouts, warn on corrupted token.json, eprintln! instead of println!, and outcome-focused tracing on refresh/fallback/device-flow events.

Follow-ups intentionally out of scope (happy to split into separate PRs):

Shared RFC 8628 device-flow helper for kimicode + githubcopilot — matches your suggestion. Only these two currently use device flow (chatgpt_codex and gemini_oauth use authorization-code variants).
Typed request struct for the stream field (currently payload.as_object_mut().unwrap().insert(...)) — touches shared formats::anthropic::create_request.
Optional process-level single-flight protection against concurrent device-flow logins (file lock on token.json).

All tests pass, clippy clean. Ready for another look.

Follow-up to aaif-goose#8466. Hitting the actual Kimi Code API after merge revealed that the `/v1/models` endpoint exposes a single model, `kimi-for-coding`, and silently routes any other model name to it. The previously hardcoded `kimi-k2.5` / `kimi-k2-thinking` entries never existed as distinct models and only misled the configure picker. - Replace hardcoded constants with `kimi-for-coding`. - Add `fetch_live_models` that queries `{api_base}/v1/models` and `fetch_supported_models` now uses the live catalogue, falling back to `KIMI_CODE_KNOWN_MODELS` when the endpoint is unreachable or returns an empty list. - New wiremock tests cover the happy path, 5xx fallback, and empty-list fallback. Existing users with `GOOSE_MODEL: kimi-k2.5` in their config keep working because the server's silent fallback routes them to `kimi-for-coding` — no user-facing migration required. Signed-off-by: DaeHee Lee <lee111dae11@proton.me>

Follow-up to aaif-goose#8466. Hitting the actual Kimi Code API after merge revealed that `/v1/models` exposes a single model, `kimi-for-coding`, and the inference endpoint silently accepts any `model` field and routes every request to it. The previously hardcoded `kimi-k2.5` / `kimi-k2-thinking` entries never existed as distinct models and only misled the configure picker. - Replace hardcoded defaults with `kimi-for-coding`. - `fetch_supported_models` now queries `{api_base}/v1/models` directly and propagates errors, mirroring the `openai_compatible` provider. No fallback: by the time this runs OAuth has completed, and a failure is better surfaced than hidden behind a stale static list. - New wiremock tests cover the server catalogue response and the 5xx error-propagation path. Existing users with `GOOSE_MODEL: kimi-k2.5` in their config keep working because the server silently routes them to `kimi-for-coding`. Signed-off-by: DaeHee Lee <lee111dae11@proton.me>

Both `kimicode` and `githubcopilot` implemented RFC 8628 device-code polling independently. Their two copies diverged enough to hide bugs: - `githubcopilot::poll_for_access_token` called `error_for_status()` before parsing the JSON body, so RFC 8628 error responses like `authorization_pending` (returned with 4xx status per §3.5) could surface as a plain HTTP error instead of polling continuation. - It also ignored `slow_down` (§3.5: MUST increase interval by 5s) and used a hardcoded 36 × 5s poll window instead of the server's `expires_in`. - `kimicode` handled both correctly — now its implementation is the shared one. Add `providers::oauth_device_flow` with `run_device_flow` + `refresh_device_flow_token` entry points. Providers configure via a `DeviceFlowConfig` that carries endpoint URLs, client_id, scopes, extra headers, and request encoding (`Form` for kimi, `Json` for github). `DeviceFlowTokens` exposes `refresh_token` and `expires_at` as `Option`s so callers can apply their own fallback when the server omits them (RFC 6749 §5.1 / §6). - kimicode: `device_flow_login`, `poll_for_token`, and `do_refresh_token` become thin wrappers over the helper, normalizing the on-disk `KimiToken` shape via `tokens_to_kimi`. - githubcopilot: `login` → `run_device_flow`; `get_device_code` and `poll_for_access_token` removed. The RFC 8628 compliance issues above are fixed as a side effect. Net change: +250 in helper (+ tests), −370 across the two providers (−120 for a net reduction of ~315 lines). Testing: - 9 new wiremock tests in `oauth_device_flow` cover pending → success, `slow_down` backoff, expires_in timeout, HTTP error body, server-side `access_denied`, missing refresh_token, and refresh happy-path. - kimicode tests now focus on Kimi-specific integration (token cache, refresh fallback); duplicated polling tests removed — the helper tests cover equivalent behavior. Follow-up to aaif-goose#8466. Signed-off-by: DaeHee Lee <lee111dae11@proton.me>

* main: (37 commits) polish: refine sidebar activity indicators, add placeholder token, and tidy search field (#8606) feat: add /edit command to cli for on-demand prompt editing (#8566) docs(mcp): add Rendex MCP Server extension tutorial (#8541) Lifei/delete tauri backend acp (#8582) chore: set goose binaries as executable in package.json (#8589) feat: add Novita AI as declarative provider (#8432) feat: add Kimi Code provider with OAuth device flow authentication (#8466) fix: chat loading-state model placeholder (#8431) fix: expand tool calls by default when Response Style is Detailed (#8478) fix: create logs dir before writing llm request log (#8522) fix: enable token usage tracking and configurable stream timeout for Ollama provider (#8493) fix tauri-plugin-dialog version constraint to match other plugins (#8542) call goose serve from tauri frontend via goose-acp client (#8549) failed the script when bundle:default fails and cleanup "alpha" (#8580) pass globally unique conversation identifier as sessionId in databricks api call (#8576) fix: use sqlx chrono decode for thread timestamps instead of manual parsing (#8575) docs: remove stale gemini-acp references (#8572) show individual untracked files in git changes widget (#8574) fix: update publishing flow to include new sdk dir (#8573) fix: remove double border on content in chat (#8545) ...

Both `kimicode` and `githubcopilot` implemented RFC 8628 device-code polling independently. Their two copies diverged enough to hide bugs: - `githubcopilot::poll_for_access_token` called `error_for_status()` before parsing the JSON body, so RFC 8628 error responses like `authorization_pending` (returned with 4xx status per §3.5) could surface as a plain HTTP error instead of polling continuation. - It also ignored `slow_down` (§3.5: MUST increase interval by 5s) and used a hardcoded 36 × 5s poll window instead of the server's `expires_in`. - `kimicode` handled both correctly — now its implementation is the shared one. Add `providers::oauth_device_flow` with `run_device_flow` + `refresh_device_flow_token` entry points. Providers configure via a `DeviceFlowConfig` that carries endpoint URLs, client_id, scopes, extra headers, and request encoding (`Form` for kimi, `Json` for github). `DeviceFlowTokens` exposes `refresh_token` and `expires_at` as `Option`s so callers can apply their own fallback when the server omits them (RFC 6749 §5.1 / §6). - kimicode: `device_flow_login`, `poll_for_token`, and `do_refresh_token` become thin wrappers over the helper, normalizing the on-disk `KimiToken` shape via `tokens_to_kimi`. - githubcopilot: `login` → `run_device_flow`; `get_device_code` and `poll_for_access_token` removed. The RFC 8628 compliance issues above are fixed as a side effect. Net change: +250 in helper (+ tests), −370 across the two providers (−120 for a net reduction of ~315 lines). Testing: - 9 new wiremock tests in `oauth_device_flow` cover pending → success, `slow_down` backoff, expires_in timeout, HTTP error body, server-side `access_denied`, missing refresh_token, and refresh happy-path. - kimicode tests now focus on Kimi-specific integration (token cache, refresh fallback); duplicated polling tests removed — the helper tests cover equivalent behavior. Follow-up to aaif-goose#8466. Signed-off-by: DaeHee Lee <lee111dae11@proton.me>

chatgpt-codex-connector Bot reviewed Apr 10, 2026

View reviewed changes

fix: apply cargo fmt formatting to kimicode provider

08ce154

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> Signed-off-by: DaeHee Lee <lee111dae11@proton.me>

chatgpt-codex-connector Bot reviewed Apr 12, 2026

View reviewed changes

jamadeo assigned DOsinga Apr 13, 2026

jh-block assigned jh-block and unassigned DOsinga Apr 15, 2026

jh-block reviewed Apr 15, 2026

View reviewed changes

soilSpoon and others added 2 commits April 16, 2026 17:01

chatgpt-codex-connector Bot reviewed Apr 16, 2026

View reviewed changes

jh-block approved these changes Apr 16, 2026

View reviewed changes

jh-block added this pull request to the merge queue Apr 16, 2026

Merged via the queue into aaif-goose:main with commit 93e6f8d Apr 16, 2026
20 checks passed

This was referenced Apr 16, 2026

Kimi Code Support #7150

Closed

Built-in Kimi provider appears misconfigured for Kimi For Coding / Coding Plan #8584

Closed

soilSpoon mentioned this pull request Apr 16, 2026

fix(kimicode): align model metadata with live Kimi Code API #8588

Merged

4 tasks

This was referenced Apr 17, 2026

refactor(providers): extract shared OAuth device-flow helper #8619

Merged

refactor(providers): extract http_status module and rename handle_status_openai_compat #8620

Merged

github-actions Bot mentioned this pull request Apr 23, 2026

chore(release): release version 1.32.0 #8764

Closed

BrewTestBot mentioned this pull request Apr 23, 2026

block-goose-cli 1.32.0 Homebrew/homebrew-core#279029

Merged

Conversation

soilSpoon commented Apr 10, 2026

Summary

Testing

Related Issues

Uh oh!

chatgpt-codex-connector Bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

chatgpt-codex-connector Bot Apr 10, 2026

Choose a reason for hiding this comment

Uh oh!

chatgpt-codex-connector Bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

chatgpt-codex-connector Bot Apr 12, 2026

Choose a reason for hiding this comment

Uh oh!

chatgpt-codex-connector Bot Apr 12, 2026

Choose a reason for hiding this comment

Uh oh!

chatgpt-codex-connector Bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

chatgpt-codex-connector Bot Apr 12, 2026

Choose a reason for hiding this comment

Uh oh!

chatgpt-codex-connector Bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

chatgpt-codex-connector Bot Apr 12, 2026

Choose a reason for hiding this comment

Uh oh!

chatgpt-codex-connector Bot Apr 12, 2026

Choose a reason for hiding this comment

Uh oh!

chatgpt-codex-connector Bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

chatgpt-codex-connector Bot Apr 12, 2026

Choose a reason for hiding this comment

Uh oh!

jh-block commented Apr 15, 2026

Uh oh!

jh-block left a comment

Choose a reason for hiding this comment

Uh oh!

jh-block Apr 15, 2026

Choose a reason for hiding this comment

Uh oh!

soilSpoon Apr 16, 2026

Choose a reason for hiding this comment

Uh oh!

jh-block Apr 15, 2026

Choose a reason for hiding this comment

Uh oh!

soilSpoon Apr 16, 2026

Choose a reason for hiding this comment

Uh oh!

chatgpt-codex-connector Bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

chatgpt-codex-connector Bot Apr 16, 2026

Choose a reason for hiding this comment

Uh oh!

soilSpoon Apr 16, 2026

Choose a reason for hiding this comment

Uh oh!

chatgpt-codex-connector Bot Apr 16, 2026

Choose a reason for hiding this comment

Uh oh!

soilSpoon Apr 16, 2026

Choose a reason for hiding this comment

Uh oh!