Skip to content

Commit 563a182

Browse files
lidge-junwtfsayo
andauthored
feat(devin): Cognition cloud provider, carried from lidge-jun#4078 and hardened (lidge-jun#4285)
* feat(devin): add experimental Cognition (Devin/Windsurf) adapter Carries the adapter from lidge-jun#4078 onto current dev and places the test in its layout domain (tests/providers/devin-adapter.test.ts) with the layout map and membership fixture updated. Co-authored-by: Sayo <[email protected]> * fix(devin): harden the cloud-direct adapter before it ships Four independent reviews of the carried lidge-jun#4078 adapter found one credential-leak blocker, one abort blocker, and a set of routing and lifecycle defects. Credentials: RegisterUser and GetUserJwt copied raw upstream bodies into Error.message, which reaches CLI output, the adapter error event, and /api/logs. A Connect error can quote the request, and the request holds the sign-in token or the api_key; redactSecretString does not match a bare JWT. Every auth and chat error now reports status, an allowlisted Connect code, and a trace id only. The four credential-bearing POSTs stop following redirects, and the api-server host is checked against a Cognition allowlist before it reaches a URL - including on the way into auth.json, so an EU or FedStart tenant host survives a reload instead of being dropped by the Copilot-only validator. Routing: the adapter posted to the static registry baseUrl, so an EU or FedStart account signed in and then sent every RPC to a server it is not provisioned on. The signed-in account's tenant now decides the host. Cancellation: after headers arrived nothing observed the caller's signal, so a client cancel drained until the idle timer fired and then surfaced as truncated_stream, while the adapter emitted neither done nor error and left the bridge to synthesize adapter_eof. The body is cancelled on abort and the turn reports the cancellation. Also: a natural completion no longer reports stopReason "stop", which was costing every clean turn its final_answer phase; sampling options reach the cloud instead of its 128k/0.7 defaults; thinking stays out of replayed assistant content; usage survives an error; gzip frames are bounded on output as well as input; dotted model ids normalize to the catalog spelling; the session cache is bounded; and logout clears the cached user_jwt whose payload carries the api_key. Co-authored-by: Sayo <[email protected]> * test(devin): assemble the userinfo URL so the privacy scanner does not read it as an email * feat(devin): pin the client version to the shipped release and accept the real token shape Evidence from a live free-tier account and the shipped Devin Desktop 3.9.19 bundle. Details in devlog/_plan/260911_devin_two_providers/003_live_evidence.md. The sign-in token is not a JWT. A real sign-in returns a 47-character ott$<base64url> one-time value and RegisterUser accepts it, so the JWT-shape gate would have rejected every real login. The paste parser now recognises one opaque credential-shaped word rather than a token format. RegisterUser returned api_server_url https://server.self-serve.windsurf.com for an ordinary free account, which is what the tenant-routing fix in the previous commit exists for: the hardcoded server.codeium.com was wrong for this account before anyone reached an enterprise tenant. The api-server allowlist gains the staging and beta hosts the shipped bundle names, and the client version default moves from 2.0.0 - which predates the Devin rebrand - to the 3.9.19 the desktop client reports, overridable through OPENCODEX_DEVIN_CLIENT_VERSION. Co-authored-by: Sayo <[email protected]> * docs(devin): carry the upstream MIT notice on the derived cloud-direct files A similarity check against rsvedant/opencode-windsurf-auth puts wire.ts at 1.000, index.ts at 0.988, chat.ts at 0.912, metadata.ts at 0.863, auth.ts at 0.835 and catalog.ts at 0.753 - same module split, same comments, same field layout. These files are a derivative of that repository's src/cloud-direct/, which is MIT licensed, Copyright (c) 2026 Vedant, and the carry arrived with no notice at all. The full permission notice sits in the module entry point and the other five files carry a short attribution header pointing at it, which is what MIT asks for in a distributed source tree. Co-authored-by: Sayo <[email protected]> * docs(devin): record that the cloud chat path is unverified on a free account * fix(devin): explain the opaque cloud denial instead of guessing at it * docs(devin): restore the cloud provider rows and adapter sections on the new dev devin-cli landed on dev as lidge-jun#4288, so this branch now carries only the cloud provider. The provider rows, the login line, and the adapters reference sections for devin are re-added on top of the current docs, and every locale still leads with the measured result that its chat path is unverified. * fix(devin): restore the invalid_argument matcher the rebuild dropped The rebuild re-applied the last commit's documentation by hand and its chat.ts hunk went with it, so the runtime explanation was back to keying only on permission_denied while the measured free account returns invalid_argument - the one trailer it needed to fire for. Also from the rebuild audit: the conformance test no longer names devin-cli in guards the RUN_TURN_ONLY_WIRES set already skips, the two layout maps list the devin test files alphabetically, and structure/adapters/registry.md records why the cloud devin wire is a direct registry entry alongside devin-cli. * feat(devin): fix the request encoding that made every chat turn fail The cloud provider could not complete a single turn on any account. A paid account settled what it was not: all 229 catalogue models came back enabled and GetChatMessage failed exactly as it had on the free tier, so entitlement was never the cause. Importing the working reference's zero-dependency builder and sending its request through our own transport returned HTTP 200 and a real stream, which put the fault in our encoder rather than the wire. Diffing the two encoded messages field by field left one difference: in CompletionConfiguration, lidge-jun#2 is the output cap and lidge-jun#3 is the context window, and we had them swapped. A caller asking for 32 output tokens wrote 32 into the context-window field, and Cognition answered with an opaque invalid_argument. Fields lidge-jun#6 and lidge-jun#11 are not part of the message at all. A temperature of exactly 0 is refused with that same opaque error. Deterministic output is the ordinary case for a coding client, so it is clamped to the smallest accepted value rather than silently replaced with the service default. Three transport facts had to hold together, which is why testing them one at a time looked fruitless: the credential is the session token doubled and dash-joined in an Authorization: Basic header while the body keeps one copy, the request envelope is uncompressed, and Metadata lidge-jun#31 carries a 732-character fingerprint whose length the service checks. The metadata identity is its own seven-field shape rather than the desktop client's telemetry set, the request carries the verified tag set, and the short-lived user_jwt is now opt-in because the chat path does not need it. Verified live on a paid account: six combinations, two hosts by three models, all returning PONG with a finish reason and usage. A regression test pins the tag map so the swap cannot return silently. Co-authored-by: Sayo <[email protected]> * fix(devin): stop the opaque-denial message asserting a retracted explanation The error string still told the user entitlement was proved and that the request fields had been ruled out. That was the hypothesis this work retracted: the same sentence came back for every turn until the CompletionConfiguration tag map was corrected, and a temperature of exactly 0 still produces it. It now points at the request first and names the test that pins the accepted field layout, and only then at the account's model access. Also from the pre-merge review: the comments claiming the hosted chat path needs the user_jwt, the stale 128k output-default comment, the promptId that is now optional because lidge-jun#22 is omitted on a first turn, and an English docs line that claimed tool calls were verified when the live evidence is chat and usage across three models. --------- Co-authored-by: Sayo <[email protected]>
1 parent 13ad8e7 commit 563a182

49 files changed

Lines changed: 4051 additions & 13 deletions

File tree

Some content is hidden

Large Commits have some content hidden by default. Use the searchbox below for content that may be hidden.
Lines changed: 76 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,76 @@
1+
# 001 — Devin/Cognition as two providers
2+
3+
Objective: opencodex gains two Devin-family providers.
4+
5+
- `devin` — cloud-direct. Connect-RPC to Cognition's `exa.api_server_pb.ApiServerService`,
6+
carried from PR #4078 (author @wtfsayo) onto current `dev` and hardened.
7+
- `devin-cli` — local. Spawns the Devin CLI and speaks Agent Client Protocol
8+
(newline-delimited JSON-RPC on stdio), modeled on the user-supplied working
9+
`server.mjs` proxy and the reference executor in `.tmp/openproxy-ref`.
10+
11+
`.tmp/openproxy-ref` (quangdang46/openproxy) is read-only reference. No code or
12+
license-bearing text from it enters this repository.
13+
14+
## Work phases
15+
16+
| id | outcome |
17+
|---|---|
18+
| wp1 | Carry + harden the cloud-direct `devin` adapter on current `dev` |
19+
| wp2 | Live Cognition evidence (free signup + client download via aside), fold verified constants in |
20+
| wp3 | Second provider `devin-cli` over ACP stdio |
21+
| wp4 | Docs/locale parity, full gates, PR, merge into `dev` |
22+
23+
## wp1 — what changes and why
24+
25+
The carry itself is done: `git merge --squash pr4078` applied cleanly onto
26+
`9ea5759226`, the root-level test moved to its layout domain
27+
(`tests/providers/devin-adapter.test.ts`) with `scripts/test-layout/layout.json`
28+
and `tests/fixtures/test-layout-expected.json` updated, and the focused suites pass
29+
(36/36). Four independent reviewers audited the result. Their findings define wp1's
30+
diff:
31+
32+
### 1. Tenant api-server routing (major, real runtime failure)
33+
34+
`src/oauth/devin.ts` stores RegisterUser's `api_server_url` on the credential, but
35+
`src/adapters/devin.ts` always posts GetUserJwt / GetCascadeModelConfigs /
36+
GetChatMessage to `provider.baseUrl`, which `src/providers/registry.ts` hardcodes to
37+
`https://server.codeium.com`. EU and FedStart tenants return a different host
38+
(`eu.windsurf.com/_route/api_server`, `windsurf.fedstart.com/_route/api_server`), so
39+
those accounts log in and then send every call to the wrong server. GitHub Copilot
40+
already threads `credential.apiBaseUrl` through; Devin must do the same, falling back
41+
to the default host only when RegisterUser returned nothing.
42+
43+
### 2. Portal/register override (major, real runtime failure)
44+
45+
Login always signs in against `DEFAULT_REGION`. `src/oauth/devin/types.ts` documents
46+
a `--portal-url` override that nothing wires, so a non-US tenant never reaches its
47+
matching RegisterUser host. Honor the override and persist it next to the api-server
48+
URL on the credential.
49+
50+
### 3. Model-id normalization (minor, degraded path)
51+
52+
`src/adapters/devin.ts` has no dotted-to-hyphen map. With the live catalog missing we
53+
append `-medium` to the raw id, turning `swe-1.6` into `swe-1.6-medium`, which
54+
Cognition answers with an opaque `permission_denied`. Normalize `.` to `-` before
55+
lookup and suffix only ids that actually carry an effort segment.
56+
57+
### 4. Docs/locale parity (major, deferred to wp4)
58+
59+
English `providers.md` and `reference/adapters.md` gained `devin`; the seven locales
60+
(`ko ja zh-cn zh-tw fr ru tr`) still jump from `cursor` to `github-copilot` and from
61+
`cursor` to `azure-openai`. No test compares them, but AGENTS.md forbids a locale
62+
contradicting the English source. Both providers land in every locale in wp4, once
63+
the final surface is known.
64+
65+
### 5. Auth and streaming findings
66+
67+
Two reviewers (credential handling; streaming terminal/abort semantics) are still
68+
running. Their blockers and majors fold into this same wp1 diff before A closes.
69+
70+
## Boundaries
71+
72+
- No change to `src/router.ts`, `src/server/lifecycle.ts`, or
73+
`src/server/responses/core.ts` reaching `src/lab/`.
74+
- No new CLI command, so `skills/ocx/` and `src/cli/capabilities.ts` stay as they are.
75+
- `devin` keeps `dashboardPreset: false` and stays out of the featured lists.
76+
- Security notes stay in `.tmp/`, never in `devlog/`.
Lines changed: 69 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,69 @@
1+
# 002 — wp1 audit: folded reviewer findings
2+
3+
Four independent reviewers (xai/grok-4.6, high effort) audited the carried commit
4+
`142c095673`. Three returned; the streaming reviewer is still running and its
5+
findings fold into this same cycle if they arrive before C. Verdicts below are mine
6+
after reading the cited code.
7+
8+
## Accepted — blocker
9+
10+
**Raw upstream bodies in auth error messages.** `register-user.ts:96,113` and
11+
`cloud-direct/auth.ts:99,126` copy the response body into `Error.message`. That
12+
message reaches CLI output, the adapter's `emit({ type: "error" })`, and
13+
`/api/logs`. A Connect error that echoes `firebase_id_token`, or a 200 whose
14+
`user_jwt` fails the shape regex, publishes a live credential; `redactSecretString`
15+
does not match a bare `eyJ…` JWT. Confirmed by reading both files. Fix: status plus
16+
allowlisted Connect code plus trace id, never the body.
17+
18+
## Accepted — major
19+
20+
1. **Tenant api-server routing.** `credential.apiBaseUrl` is written at login but
21+
no call site reads it, and `store.ts:461` only persists Copilot origins, so an
22+
EU/FedStart host is dropped on the next load anyway. Thread it through
23+
`mintUserJwt`, the catalog fetch, and `streamChatEvents`, and teach the store to
24+
persist a validated Devin origin.
25+
2. **Redirect following on credential POSTs.** Both credential POSTs use the default
26+
`redirect: "follow"`, so a 307/308 forwards the Firebase token or the protobuf
27+
`api_key` to an attacker-chosen `Location`. Set `redirect: "error"` and validate
28+
the host the same way `validateCopilotApiBaseUrl` does.
29+
3. **Credential shape.** `refresh: ""` makes `detectOAuthWarning` report
30+
`stale_credentials` for every Devin account from the moment of login, and
31+
`refreshDevinToken` extends the expiry without contacting Cognition, so a revoked
32+
key keeps looking valid. Use the durable-key house pattern: `refresh` carries the
33+
key, expiry is effectively unbounded, and refresh throws so a 401 marks
34+
`needsReauth`.
35+
4. **Paste parsing.** `loginDevin` posts the entire pasted string as
36+
`firebase_id_token`. The on-screen value is a token, but a user who pastes the
37+
callback URL instead sends a URL. Parse a fragment/query token out of a URL paste
38+
and reject a paste that contains no token.
39+
5. **`clearCachedUserJwt` is never called.** The cached `user_jwt` (its payload
40+
contains `api_key`) survives logout in process memory. Wire it into the Devin
41+
logout path.
42+
43+
## Accepted — minor
44+
45+
6. `result.name` overwrites the JWT `email` with a display name, so reauth identity
46+
comparison collides. Keep the email; the name is not an identity.
47+
7. `registerUser` does not receive `ctrl.signal`, so cancelling login does not abort
48+
the exchange.
49+
8. No dotted-to-hyphen model-id map, so a degraded-path `swe-1.6` becomes
50+
`swe-1.6-medium` and Cognition answers `permission_denied`.
51+
52+
## Rejected / deferred
53+
54+
- **Copying the reference's gRPC-web framing.** `.tmp/openproxy-ref` talks to
55+
`LanguageServerService` over gRPC-web with a Bearer header; we talk to
56+
`ApiServerService` over Connect-RPC with the key inside `Metadata`. They are two
57+
different products. Adopting the reference's headers or field numbers would break
58+
auth and proto decode. Reference value is the CLI/ACP executor, which is wp3.
59+
- **`defaultRefreshPolicy: "disabled"`.** Correct for a durable key; keep it.
60+
- **Docs/locale parity.** Real and required, but the final surface is not known until
61+
`devin-cli` lands, so it is wp4.
62+
- **Dead plugin types** (`PersistedCredentials`, `syncedViaOpencodeAuth`). Removed
63+
where they are genuinely unreferenced; not a leak either way.
64+
65+
## Verification for this cycle
66+
67+
`bun x tsc --noEmit`, the focused Devin/adapter/layout suites, `bun run privacy:scan`,
68+
plus new regression tests for: error messages that must not contain a token, redirect
69+
refusal, host allowlist rejection, tenant host threading, and the dotted model id.
Lines changed: 170 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,170 @@
1+
# 003 — wp2: live Cognition evidence
2+
3+
A free Cognition account was created through the browser on 2026-09-12 and the
4+
shipped desktop client was downloaded. Everything below is measured, not inferred.
5+
6+
## What the account looks like
7+
8+
Devin Desktop 3.9.19 (`Devin-darwin-arm64-3.9.19.dmg`, 337 MB). Windsurf has been
9+
rebranded: `windsurf.com` now redirects to `devin.ai/desktop`, and the bundled
10+
extension still identifies itself as `publisher: codeium`, `name: windsurf`,
11+
`displayName: Devin`. `product.json` reports `windsurfVersion: 3.9.19` and
12+
`codeiumVersion: 1.48.2`.
13+
14+
## Constants confirmed against the shipped client
15+
16+
Read from `Devin.app/Contents/Resources/app/extensions/windsurf/dist/extension.js`:
17+
18+
- Auth0 client id `3GUryQ7ldAeKEuD2obYnppsnmj58eP5u` — present verbatim. The
19+
carried adapter's value is correct.
20+
- Hosts: `server.codeium.com`, `server-staging.codeium.com`,
21+
`server-beta.codeium.com`, `register.windsurf.com`, `eu.windsurf.com/_route/api_server`,
22+
`windsurf.fedstart.com/_route/api_server`, and the tenant template
23+
`your-company.windsurf.com`. The allowlist in `src/oauth/devin/api-base.ts` was
24+
widened to the two staging/beta hosts on this evidence.
25+
- Method names `RegisterUser`, `GetChatMessage` and `GetCascadeModelConfigs` all
26+
appear as string literals.
27+
28+
## What the live calls proved
29+
30+
1. **The sign-in token is not a JWT.** A real sign-in returned a 47-character
31+
`ott$<base64url>` one-time token, and RegisterUser exchanged it successfully.
32+
The JWT-shape gate added during wp1 would have rejected every real login, so
33+
`parseDevinAuthPaste` now checks for one opaque credential-shaped word instead
34+
of a token format. The token is single-use: the second exchange of the same
35+
value fails, which is why the probe needed a fresh sign-in.
36+
37+
2. **The tenant-routing fix is load-bearing, not theoretical.** RegisterUser
38+
returned `api_server_url: https://server.self-serve.windsurf.com` for an
39+
ordinary free account — not `server.codeium.com`, which the registry hardcodes
40+
and the carried adapter always used. Without wp1's change every free-tier
41+
account would have sent its RPCs to a host it is not provisioned on.
42+
43+
3. **The api_key and the catalog work.** `GetCascadeModelConfigs` against that
44+
host returned 227 model uids. Exactly one is enabled on the free tier:
45+
`swe-1-6-slow`. The site advertises "unlimited SWE-2"; the API does not agree,
46+
which is worth knowing before anyone documents a model list.
47+
48+
4. **`GetChatMessage` fails with `invalid_argument`.** Message is the opaque
49+
"an internal error occurred (trace ID: …)". Client version strings `3.9.19`,
50+
`2.0.0` and `1.48.2` in Metadata fields 2 and 7 all fail identically, so the
51+
version pin is not the cause — the comment in `metadata.ts` claiming a version
52+
mismatch produces exactly this error is no longer a sufficient explanation.
53+
The version default was still moved to the shipped `3.9.19` with an
54+
`OPENCODEX_DEVIN_CLIENT_VERSION` override, because `2.0.0` predates the rebrand
55+
and nothing argues for keeping it.
56+
57+
This is the open item. The request encoding is being compared field by field
58+
against the shipped bundle and against the two actively maintained references.
59+
60+
## Ecosystem survey
61+
62+
Twelve independent Windsurf/Cognition proxies were catalogued. The two that
63+
matter here:
64+
65+
- `dwgx/WindsurfAPI` (~2975 stars, updated this week) uses the same
66+
`server.codeium.com` `GetChatMessage` Connect-RPC path we do.
67+
- `rsvedant/opencode-windsurf-auth` (~70 stars) is a direct-cloud Connect-RPC
68+
streaming client for an opencode plugin. Our carried files reference
69+
`opencode auth login`, `syncedViaOpencodeAuth` and an
70+
`opencode-windsurf-auth` CLI in `src/oauth/devin/types.ts`, so #4078 very
71+
likely derives from it. Its license and the derivation are being checked; if
72+
it is derived, attribution is required before this merges.
73+
74+
`quangdang46/openproxy` talks to a different product (gRPC-web
75+
`LanguageServerService`), so it is a secondary reference only.
76+
77+
## wp2 outcome: the cloud chat path stays unverified
78+
79+
Every request-shape hypothesis was tried against the live account and none of
80+
them changed the trailer. In probe order: client version `3.9.19`, `2.0.0`,
81+
`1.48.2`; the Connect request frame sent uncompressed with
82+
`Connect-Content-Encoding` dropped; `Metadata` #31 filled with 732 hex
83+
characters; `GetChatMessageRequest` #2, #15 and #20 added and #22 dropped on the
84+
first turn; `ChatMessagePrompt` #1 `message_id` added; `Authorization: Basic`
85+
in both base64 and raw doubled-key forms; and both hosts. Same
86+
`invalid_argument: an internal error occurred` every time, with a fresh trace id.
87+
88+
The model gate is provably fine. `swe-2-high` and `claude-sonnet-5-medium` are
89+
refused locally as disabled, and a bogus uid is refused as unlisted, so the
90+
failure is specific to `swe-1-6-slow` — the one model a free account has, and a
91+
"slow" lane at that.
92+
93+
**Entitlement now outranks request shape as the explanation.** The site
94+
advertises "Slow Devin Cloud access with limited quotas" for free accounts, and a
95+
slow lane plausibly is not served by this RPC at all. #4078's author reported a
96+
live PONG on 2026-09-09 with the *original* field set, which is the deciding
97+
fact: shipping unverified wire changes would risk regressing an account that
98+
works today in exchange for no measured gain here. The whole experimental delta
99+
was reverted; only the wp1 hardening and the MIT notice remain.
100+
101+
Confirming this needs a paid account or a captured working request. Neither is
102+
available in this session, so the cloud provider is not merge-ready and the
103+
adapter's own model gate is what stops a user hitting this blindly.
104+
105+
## The chat path works. What was actually wrong.
106+
107+
A paid account was obtained on 2026-09-12 and the entitlement hypothesis died
108+
immediately: all 229 catalogue models came back enabled, and `GetChatMessage`
109+
failed exactly as it had on the free account. The failure was never about the
110+
plan.
111+
112+
Isolating it took one decisive move. The most actively maintained reference
113+
(`dwgx/WindsurfAPI`) is zero-dependency ESM, so its request builder can simply be
114+
imported. Building a turn with the reference builder and sending it through our
115+
own transport returned **HTTP 200** and a real Connect stream — which proved the
116+
transport, the headers and the credential were all fine, and put the fault in our
117+
request encoder. Diffing the two encoded messages field by field left exactly one
118+
difference: `CompletionConfiguration` (#8).
119+
120+
reference #1=1 #2=8192 #3=128000 #5=double #7=40 #8=double
121+
ours #1=1 #2=64000 #3=32 #5=double #6=double #7=50 #8=double #11=double
122+
123+
**#2 is the output cap and #3 is the context window; we had them swapped.** A
124+
caller asking for 32 output tokens wrote 32 into the context-window field, and
125+
Cognition answered with an opaque `invalid_argument: an internal error occurred`.
126+
That is why every account failed identically and why no amount of probing the
127+
transport helped. The reference's own comments record the same mis-tagging and
128+
the same re-calibration.
129+
130+
A second, independent trap sat behind it: **a temperature of exactly 0 is
131+
refused** with the same opaque error. Deterministic output is the common case for
132+
coding clients, so it is clamped to the smallest accepted value rather than
133+
silently replaced with the service default.
134+
135+
Three transport facts also had to be right together, and testing them one at a
136+
time is why they looked useless earlier:
137+
138+
- the credential is the session token doubled and dash-joined in
139+
`Authorization: Basic`, while the protobuf body keeps a single copy;
140+
- the request envelope is uncompressed;
141+
- `Metadata` #31 carries 732 hex characters, whose length the service checks and
142+
whose value it does not.
143+
144+
The metadata identity is also its own shape — seven fields, the optional
145+
`user_jwt`, and the fingerprint — not the desktop client's fuller telemetry set.
146+
147+
### Verified
148+
149+
Six combinations, two hosts by three models, all returning `PONG` with a finish
150+
reason and usage:
151+
152+
| host | model | result |
153+
|---|---|---|
154+
| `server.codeium.com` | `swe-2-high` | PONG, stop, 476/36 |
155+
| `server.codeium.com` | `claude-sonnet-5-medium` | PONG, stop, 576/5 |
156+
| `server.codeium.com` | `gpt-5-6-sol-medium` | PONG, 394/6 |
157+
| `server.self-serve.windsurf.com` | `swe-2-high` | PONG, stop, 1/36 |
158+
| `server.self-serve.windsurf.com` | `claude-sonnet-5-medium` | PONG, stop, 576/5 |
159+
| `server.self-serve.windsurf.com` | `gpt-5-6-sol-medium` | PONG, 394/6 |
160+
161+
The tag map is now pinned by a regression test that builds a request and asserts
162+
the field layout, so the swap cannot come back silently.
163+
164+
### What this retracts
165+
166+
The earlier conclusion in this document — that entitlement was the leading
167+
explanation and that the request shape had been ruled out — was wrong. The
168+
request shape was the whole problem; the probing that "ruled it out" changed one
169+
variable at a time against a broken `CompletionConfiguration` that no single
170+
variable could rescue.
Lines changed: 33 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,33 @@
1+
# 004 — wp5: splitting devin-cli out
2+
3+
The wp4 audit recommended splitting, citing MAINTAINERS.md: a new canonical
4+
registry destination is a maintained promise, and when the evidence is incomplete
5+
the repository wants an inert directory row rather than a registry entry. The
6+
cloud `devin` provider cannot complete a turn on the account we can measure.
7+
`devin-cli` does not share that RPC.
8+
9+
## What moved
10+
11+
Branch `codex/260912-devin-cli-provider` from a freshly fetched `origin/dev`
12+
(`29d632ff25`). It carries `src/adapters/devin-cli/` and
13+
`tests/providers/devin-cli-adapter.test.ts` byte-identical, plus only the
14+
`devin-cli` hunks of the adapter registry, the provider registry, the routing
15+
behaviour table, the layout map and the membership fixture. Docs get the English
16+
provider row and adapters section and the provider row in all seven locales.
17+
18+
The tool-conformance skip lists needed care: on the other branch they name both
19+
wires, and here only `devin-cli` exists, so naming a wire that is absent would
20+
have been a silent no-op rather than a skip.
21+
22+
## What stayed
23+
24+
Everything cloud-direct: `src/adapters/devin/`, `src/oauth/devin*`, the `devin`
25+
registry entry and its documentation, the MIT notice for the derived files, and
26+
this plan unit. PR #4285 keeps them.
27+
28+
## Verification
29+
30+
`bun x tsc --noEmit` clean; 76 focused tests pass; `privacy:scan` green. An
31+
independent audit of the split diff (21 files, +925/-5) found no cloud-provider
32+
leakage, agreeing registries, resolving imports, and a PR description that
33+
matches the code. Remote CI on the exact head is the suite gate.

0 commit comments

Comments
 (0)