Skip to content
Open
Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
86 commits
Select commit Hold shift + click to select a range
02053f1
fix(windows): exclude foreign users from stop candidates (#8817)
luvs01 Nov 18, 2026
7b1f476
test(server): budget image and search relay fixtures (#8903)
lidge-jun Nov 18, 2026
825bd14
feat(jev): decide effort on every pinned turn with sessionPin.effort …
tachyon6 Nov 18, 2026
cad1df3
fix(update): keep the configured pnpm store during update planning (#…
lidge-jun Nov 18, 2026
6112c07
feat(providers): add LLMTR to the free directory (#7955)
knowhycodata Nov 18, 2026
55dd7b1
fix(pool): report inert on the unified pool-settings read (#8825)
yazzang-homelab Nov 18, 2026
5069c78
fix(oauth): pause a Cloud Code Assist account on its first Terms of S…
yazzang-homelab Nov 18, 2026
9f1537d
test: isolate updater and client fixtures from host installs (#8905)
lidge-jun Nov 18, 2026
d56f721
fix(opencode): select the Connect protocol per model (#8856)
colthreepv Nov 18, 2026
23550ff
fix(adapters): replay a Google 429 only when it carries retry evidenc…
naipi11 Nov 18, 2026
79a85fc
fix(xai): name the Privacy Mode fix when xAI refuses a Legacy (no sto…
lidge-jun Nov 18, 2026
fbdccd8
feat(codex): support Business/Enterprise group-based spend_control qu…
Jinstein Nov 18, 2026
ca381c4
fix(advisor): enforce request tool choice before consultations (#8807)
luvs01 Nov 18, 2026
1f632d2
feat(claude): expose native Claude passthrough as a dashboard, CLI an…
SeongwoongCho Nov 18, 2026
6ee397c
feat(codex): add codexAuthMethod device-only login policy (#8598)
shu-chen-dot-ca Nov 18, 2026
819468c
docs(es): mirror the OpenCode protocol-per-model section (#8913)
lidge-jun Nov 18, 2026
e7317b6
fix(windows): silence helper progress and count the pnpm owner probe …
lidge-jun Nov 18, 2026
f5a0004
test(windows): silence progress before the fixture runs in the checke…
lidge-jun Nov 18, 2026
c38d6ee
test(advisor): close hosted-advisor-choice fixtures before temp-home …
lidge-jun Nov 18, 2026
53257ca
fix(anthropic): withhold hosted advisor under captured spend ceilings…
luvs01 Nov 18, 2026
aa0bb80
fix(link): restore Child file integrations before shutdown (#8803)
luvs01 Nov 18, 2026
64bfd29
feat(anthropic): report monthly usage from extra_usage when no plan w…
Jinstein Nov 18, 2026
191572b
fix(claude): apply per-model compact windows at launch (#8688)
Djaler Nov 18, 2026
4f757e9
feat(reasoning): opt in to Chat raw reasoning on the summary channel …
wewark Nov 18, 2026
d4f2567
test(windows): give PowerShell-backed proofs the spawn budget (#8926)
lidge-jun Nov 19, 2026
9646ad3
fix(link): retry owned reverse forwards and report relay degradation …
lidge-jun Nov 19, 2026
dcf15fc
Revert "fix(link): restore Child file integrations before shutdown (#…
lidge-jun Nov 19, 2026
659d59a
fix(messages): keep fragmented native streams moving (#8883)
0disoft Nov 19, 2026
e4ea3c6
fix(anthropic): opt-in one-shot image-413 recovery (carry of #7308) (…
lidge-jun Nov 19, 2026
688326f
feat(agy-acp): official Antigravity CLI over ACP as a separate provid…
lidge-jun Nov 19, 2026
f59a757
fix(logs): keep proxy-enforced web-search deadlines out of upstream f…
Caledvwlch Nov 19, 2026
270161a
fix(sync): refresh owned DeepSeek Harness catalogs (#8910)
lidge-jun Nov 19, 2026
9dc286e
fix(claude-cli): recognise paid usage and recheck stale plan limits (…
nickdavis Nov 19, 2026
d4a0382
feat(claude): support the Desktop picker on Linux through the user NS…
svc-aibot-363 Nov 19, 2026
fc0816c
fix(windows): recover startup after loopback connect timeouts (#8920)
lidge-jun Nov 19, 2026
da66d3a
fix(responses): attribute local native continuation refusals (#8922)
lidge-jun Nov 19, 2026
1d25776
feat(combos): suggested combos with paid-target consent (carry of #87…
lidge-jun Nov 19, 2026
16723a5
fix(responses): keep display estimates out of Combo spend accounting …
luvs01 Nov 19, 2026
b70f504
feat(server): carry hub quota-window projection from #7959 (#8909)
lidge-jun Nov 19, 2026
0dc315b
fix(codex): write portable sibling model catalog bindings (#8912)
lidge-jun Nov 19, 2026
0252d31
feat(providers): per-provider connectTimeoutMs and stallTimeoutSec ov…
lidge-jun Nov 19, 2026
7252880
feat(update): explain managed release directories in the update dialo…
lidge-jun Nov 19, 2026
f4f04cb
feat(client): carry the authenticated Hub management relay from #8079…
lidge-jun Nov 19, 2026
5f2c151
perf(responses): reduce terminal repair and credential masking work (…
0disoft Nov 19, 2026
780803b
feat(accounts): keep the upstream message and status on an auto-pause…
andrew05060414 Nov 19, 2026
3732baf
feat(integrations): Gemini API inbound and one-click Antigravity app …
andrew05060414 Nov 19, 2026
25deb4d
fix(anthropic): normalize caller instructions in translated OAuth req…
colthreepv Nov 19, 2026
f235042
feat(providers): add M365 Copilot with live account model discovery (…
x3M3x Nov 19, 2026
80e8952
fix(link): restore Child file integrations before shutdown on Windows…
lidge-jun Nov 19, 2026
899797d
feat(claude): opt-in weekly reserve for Claude subscription traffic (…
medking82 Nov 19, 2026
ae7748e
feat(openai): opt-in same-model retries for pre-output ChatGPT capaci…
junhoi-pk Nov 19, 2026
2386261
docs: keep error detail and provider reference within their line caps…
lidge-jun Nov 19, 2026
5aaf8fa
Surface validated Cloud Code Assist appeal links (#8752) (#8936)
lidge-jun Nov 19, 2026
da72b78
feat(desktop): admit proof-verified Codex package builds for compatib…
luvs01 Nov 19, 2026
7c1701e
feat(jev): hide the effort control on a decide-mode JEV Combo (#8861)
tachyon6 Nov 19, 2026
9693a01
fix(combos): reserve output headroom and margin estimates in target a…
omar391 Nov 19, 2026
676baf6
fix(agy-acp): preserve turn policy through native Chat bridges (#8941)
lidge-jun Nov 19, 2026
3f269a3
fix(usage): retain same-account credential recovery attempt outcomes …
lidge-jun Nov 19, 2026
ea24cbc
fix(google): release a superseded retry response before waiting for a…
Caledvwlch Nov 19, 2026
ca96ead
fix(claude): fill the unset Fable tier with claude-fable-5-2 (#8818)
fivetaku Nov 19, 2026
028995f
feat(providers): add AbliteratedLLM API-key preset (#8747)
whiteworthy Nov 19, 2026
35c70e8
Fix gateway client restore after profile or settings edits (#8943)
lidge-jun Nov 19, 2026
24ae9b5
ci: raise the Ubuntu test shard timeout to 20 minutes (#8944)
lidge-jun Nov 19, 2026
9f777c1
feat(usage): cost credit-billed provider rows from providerCredits an…
svc-aibot-363 Nov 19, 2026
faa8b9d
fix(responses): scope keyless loopback replay to its WebSocket (#8816)
luvs01 Nov 19, 2026
2f38a6d
feat(codex): reduce repeated uploads and continuation storage (#7806)
Hylouis233 Nov 19, 2026
e735464
feat(google): CLIProxyAPI parity for Gemini inbound, Interactions API…
lidge-jun Nov 19, 2026
e6432db
feat(anthropic): keep one OAuth account per organization and show its…
SeongwoongCho Nov 19, 2026
7d9de0d
feat(routing): add reset-near-weighted account pool strategy (#8180)
vibewriter-hand Nov 19, 2026
04a1138
fix(combos): cool targets after committed output_limit/stall and hop …
omar391 Nov 19, 2026
5f80219
fix(responses): reject repeated Chat work announcements (#8841)
adaurban Nov 19, 2026
45c512d
feat(accounts): per-account probe and quota table, Antigravity quota-…
andrew05060414 Nov 19, 2026
51775ce
feat(combos): opt-in capabilityPolicy "union" advertises member capab…
andrew05060414 Nov 19, 2026
1b5f65f
feat: preserve native Gemini payloads over Antigravity OAuth (#8757)
Liang-Psych Nov 19, 2026
c520177
fix(responses): report the final terminal-guard leg's usage to the cl…
shu-chen-dot-ca Nov 19, 2026
62ffebb
fix(responses): infer phases from terminal snapshot tool work (#8586)
Reedtrullz Nov 19, 2026
7ebcbb4
test(update): classify the managed-release launch-context read in the…
lidge-jun Nov 19, 2026
87edd82
test(claude): inject NSS stat failures across platforms (#8963)
lidge-jun Nov 19, 2026
ecc9aba
test(server): wait for native-main startup in Daybreak listing (#8966)
lidge-jun Nov 19, 2026
6a4ae9d
fix(windows): keep snapshot owner reads inside the enumeration cap (#…
lidge-jun Nov 19, 2026
bee796f
test(windows): cover the snapshot refresh-read owner cutoff (#8967)
lidge-jun Nov 19, 2026
d61cca1
feat(step-cli): run StepCode as a Step Plan transport (#7267)
Hylouis233 Nov 19, 2026
52184b0
merge(upstream): refresh AR line onto upstream dev d61cca122d
vadymhimself Nov 20, 2026
2251525
feat(anthropic): show Claude usage credits in account quota
vadymhimself Nov 20, 2026
882ac28
fix(refresh): settle post-merge contract tests with upstream dev
vadymhimself Nov 20, 2026
1f4efca
fix(agentradar): keep the hosted Advisor under spend ceilings on the …
vadymhimself Nov 20, 2026
File filter

Filter by extension

Filter by extension


Conversations
Failed to load comments.
Loading
Jump to
The table of contents is too big for display.
Diff view
Diff view
  •  
  •  
  •  
340 changes: 340 additions & 0 deletions .github/scripts/release-evidence-policy.json

Large diffs are not rendered by default.

16 changes: 9 additions & 7 deletions .github/workflows/ci.yml
Original file line number Diff line number Diff line change
Expand Up @@ -626,10 +626,14 @@ jobs:
needs: changes
if: github.event_name != 'pull_request' || (needs.changes.outputs.ci == 'true' && needs.changes.outputs.fast != 'true')
runs-on: ${{ vars.LIDGE_RUNNERS == 'on' && (github.event_name != 'pull_request' || github.event.pull_request.head.repo.full_name == github.repository || contains(fromJSON('["OWNER","MEMBER","COLLABORATOR"]'), github.event.pull_request.author_association)) && fromJSON('["self-hosted","linux","x64","lidge"]') || 'ubuntu-latest' }}
# A fifth of the suite. A shard that needs longer than this is wedged, not
# slow — the old 30-minute ceiling was margin for the Windows leg, which no
# longer runs here (see platform-windows).
timeout-minutes: 15
# A fifth of the suite. 20 rather than 15 is headroom for hosted runners:
# ubuntu-latest shards were cancelled at the 15-minute wall in their last
# batches with no failing tests (#8651 shard 1 at batch 62/65 and shard 5 at
# 66/66; also #8585, #8645, #8841, #8861, #8887), and a cancelled shard is
# neither a pass nor a fail. A shard far beyond 20 minutes is still wedged,
# not slow. The old 30-minute ceiling was margin for the Windows leg, which
# no longer runs here (see platform-windows).
timeout-minutes: 20
strategy:
fail-fast: false
matrix:
Expand Down Expand Up @@ -820,9 +824,7 @@ jobs:
- name: Install dependencies
run: bash scripts/ci/bun-install-frozen.sh . gui

- name: Gateway quota and cache guardrails
run: bun run test:gateway-quota

- { name: Gateway quota and cache guardrails, run: bun run test:gateway-quota }
- name: GUI lint
if: needs.changes.outputs.gui == 'true'
run: |
Expand Down
6 changes: 6 additions & 0 deletions .gitignore
Original file line number Diff line number Diff line change
Expand Up @@ -49,6 +49,12 @@ devlog/**/security-advisory-draft*
.omo/
**/.omo/

# Bun's on-disk build cache. Bun writes it into whichever directory it runs in, and
# `git add -A` has swept two 59 MB copies of it into this history — see
# tests/ci-workflows/repo-hygiene.test.ts, which fails if any path here becomes
# tracked again.
*.bun-build

# Local development worktrees
.worktrees/

Expand Down
2 changes: 1 addition & 1 deletion README.md
Original file line number Diff line number Diff line change
Expand Up @@ -307,7 +307,7 @@ see the [installation docs](https://opencodex.me/getting-started/installation/).

OpenCodex tracks process-retained state in the categories below. Each has a documented bound:

- **14 retained stores** (request log, debug rings, image cache, model cache, vision
- **15 retained stores** (request log, debug rings, image cache, model cache, vision
descriptions, cursor blobs, responses continuation, etc.) are byte-accounted and
evicted by the app-owned memory budget (default 256 MiB), except the native control replay
store, which is pinned and never evicted.
Expand Down
2 changes: 2 additions & 0 deletions bin/ocx.mjs
Original file line number Diff line number Diff line change
Expand Up @@ -1580,6 +1580,8 @@ function createNodeLaunchContext() {
version: 1,
proof: launchProof,
anthropicEnvSlots: preBunAnthropicSlots,
// Running this shim under Bun can already have loaded project dotenv.
managedRelease: process.versions.bun === undefined && process.env.OCX_MANAGED_RELEASE === "1",
codexCliInspectionEnv: codexCliUpdateInspectionContext ? {
codexCliPath: preBunCodexCliPath,
path: preBunPath,
Expand Down
8 changes: 2 additions & 6 deletions bunfig.toml
Original file line number Diff line number Diff line change
Expand Up @@ -16,10 +16,6 @@ root = "tests"
#
# That qualifier is load-bearing. Bun resolves bunfig.toml from the CURRENT WORKING
# DIRECTORY, so `cd /tmp && bun test <repo>/tests/x.test.ts` loads no preload at all: nothing
# is sandboxed, and every `getConfigDir()` in that process returns the developer's real
# `~/.opencodex`. On 2026-09-15 such a run deleted one, and in #7407 another wrote auth.json.
# The write guard now arms itself from the test-file entry in that case (src/lib/test-home-guard.ts),
# so such writes throw instead of landing, but a test that needs a config directory still pins its
# own OPENCODEX_HOME instead of trusting this line, and `tests/ci-workflows/test-home-guard.test.ts`
# enforces that for the destructive case.
# is sandboxed. The guard still arms from the test-file entry (src/lib/test-home-guard.ts)
# and refuses writes to the real home. Tests that write config must pin an isolated OPENCODEX_HOME.
preload = ["./tests/preload.ts"]
2 changes: 1 addition & 1 deletion desktop/src-tauri/Cargo.lock

Some generated files are not rendered by default. Learn more about how customized files appear on GitHub.

2 changes: 1 addition & 1 deletion desktop/src-tauri/Cargo.toml
Original file line number Diff line number Diff line change
@@ -1,6 +1,6 @@
[package]
name = "opencodex-desktop"
version = "2.100.0-vadymhimself.2"
version = "2.100.0"
description = "OpenCodex desktop shell"
authors = ["OpenCodex contributors"]
license = "MIT"
Expand Down
2 changes: 2 additions & 0 deletions desktop/src-tauri/src/provider_icons.rs
Original file line number Diff line number Diff line change
Expand Up @@ -44,9 +44,11 @@ const ALIASES: &[(&str, &str)] = &[
("fireworks", "fireworks-color.svg"),
("github", "github-copilot-color.svg"),
("github-copilot", "copilot-color.svg"),
("m365-copilot", "copilot-color.svg"),
("gitlab-duo", "gitlab-duo-color.svg"),
("google", "gemini-color.svg"),
("google-antigravity", "antigravity-color.svg"),
("antigravity-acp", "antigravity-color.svg"),
("google-vertex", "gemini-color.svg"),
("groq", "groq-color.svg"),
("huggingface", "huggingface-color.svg"),
Expand Down
2 changes: 1 addition & 1 deletion desktop/src-tauri/tauri.conf.json
Original file line number Diff line number Diff line change
@@ -1,7 +1,7 @@
{
"$schema": "https://schema.tauri.app/config/2",
"productName": "OpenCodex",
"version": "2.100.0-vadymhimself.2",
"version": "2.100.0",
"identifier": "com.opencodex.desktop",
"build": {
"frontendDist": "../ui",
Expand Down
Original file line number Diff line number Diff line change
@@ -0,0 +1,159 @@
# Terminal-repair PR: bounded credential-masking comparison

This is the initial record at `dad1813b3`. The subsequent production fix and
new base/head measurements are in [the profiled follow-up](020_profiled_redaction_fix.md).

## Question and result

PR #8774 reuses the incremental SSE buffer in Responses terminal repair. Its
`test changed` job failed the existing large-terminal credential-excerpt test.
This record answers the maintainer's request to compare the original PR head
with its base without skipping assertions, increasing deadlines, or retrying CI
until it passes.

The focused comparison reproduces the same 1,500 ms failure on **both** commits.
It does not establish a regression caused by terminal-repair buffering, and it
does not establish the exact scheduling cause of the Linux CI failure. All
observed target-test failures were at the elapsed-time assertion, after the
credential-content assertion. The failing CI remains a hold.

No production code, test assertions, test deadlines, test selection rules, or
workflow configuration were changed for this follow-up.

## Revisions and original CI evidence

- Base: `1d10758fa6ced80cdd780cb000bc5c4b83f84703`.
- Original PR head: `ada6a63b052da9b95f71f23e738ddaee08c78303`.
- CI: run `43721018165`, job `133247879390` (`test changed`), Ubuntu hosted
runner, Bun `1.4.4` (`38afee6c8`), four test workers.
- CI result: 23,214 passed, 75 skipped, one failed. The existing
`tests/responses/sse-untyped-credential-mask.test.ts:552` assertion observed
`1555.7457600000198 ms`, requiring `<1500 ms`. Its content assertion passed.
- This follow-up did not dispatch or rerun hosted CI.

## Comparison protocol

The comparison ran on one native Windows machine with eight logical processors,
using Bun `1.4.4` and the same installed dependencies. The base used a detached
worktree with junctions to the head checkout's dependency directories; nothing
was installed or upgraded for the comparison. Each invocation used a fresh
process and the repository's ordinary test wrapper and home isolation. No other
agent test or build was run concurrently with these measurements.

The sample count and order were fixed before measurement: three pairs per
scenario, in the order **base, head, head, base, base, head**. No sample was
discarded and no failed invocation was retried to obtain green output.

Two focused scenarios used the same command on both revisions:

1. The complete, unchanged credential-masking test file, with four workers
configured; one selected file means there is no other test-file worker load.
2. The same file plus three fixed companion files, with four workers:
`responses-terminal-repair.test.ts`, `sse-payload-rewrite.test.ts`, and
`sse-failed-tail.test.ts`, all under `tests/responses/`. The head includes its
four new terminal-repair cases; this intentionally compares the actual PR
workload, rather than removing its new tests.

An identical diagnostic preload on both revisions captured the two real
`performance.now()` values used by the target assertion. It returned the
original clock values and restored its spy after each case. Stack attribution
was checked before the first timestamp, using the unchanged line 548 call site;
only that case emitted a measurement, so parallel companion output could not
be mistaken for its timer. No fake clock or altered expectation was used.

The preload was:

```ts
import { beforeEach, afterEach, spyOn } from "bun:test";
let reads: number[] = [];
let target = false;
let clock: ReturnType<typeof spyOn> | undefined;
beforeEach(() => {
reads = [];
target = false;
const now = performance.now.bind(performance);
clock = spyOn(performance, "now").mockImplementation(() => {
if (reads.length === 0) target = new Error().stack?.includes("sse-untyped-credential-mask.test.ts:548:") ?? false;
const value = now();
reads.push(value);
return value;
});
});
afterEach(() => {
clock?.mockRestore();
if (target && reads.length === 2) console.error(`PR8774_ELAPSED_MS=${reads[1]! - reads[0]!}`);
});
```

Save it outside test discovery and pass its absolute path as `PROBE` in each
checkout. The commands were equivalent to:

```sh
bun run test -- --parallel=4 --preload "$PROBE" \
tests/responses/sse-untyped-credential-mask.test.ts
bun run test -- --parallel=4 --preload "$PROBE" \
tests/responses/sse-untyped-credential-mask.test.ts \
tests/responses/responses-terminal-repair.test.ts \
tests/responses/sse-payload-rewrite.test.ts \
tests/responses/sse-failed-tail.test.ts
```

## Every measured sample

Values below are the target assertion's elapsed milliseconds, in per-revision
execution order, rounded only for display. Every invocation exited 1; every
target sample exceeded the unchanged 1,500 ms deadline.

| Scenario | Base samples (ms) | PR samples (ms) | Base median | PR median |
| --- | --- | --- | ---: | ---: |
| Single file | 3851.76, 2714.90, 2859.51 | 3173.92, 2656.73, 2833.17 | 2859.51 | 2833.17 |
| Four files | 3052.95, 3447.86, 2847.55 | 3417.38, 3320.02, 2955.70 | 3052.95 | 3320.02 |

The single-file PR median is 0.9% lower; the four-file PR median is 8.7% higher.
The sample ranges overlap in each scenario. Three samples per revision on
Windows are not enough to declare statistical equivalence, diagnose Linux
runner contention, or dismiss the original failure as a flake.

The single-file runs all reported 95 passed / 3 failed. Four-file base runs
reported 405/3, 405/3, and 406/2; head runs reported 410/2, 409/3, and 409/3.
Other observed failures were the existing elapsed-time assertions for repeated
or self-overlapping credentials and for registry lookup. No content assertion
failed. The four new terminal-repair regression cases passed in every head
bundle.

## Control without measurement instrumentation

After the fixed comparison, one original single-file invocation per revision
omitted the diagnostic preload, with all other command arguments unchanged.
This was an instrumentation control, not a retry policy.

| Revision | Original assertion elapsed | Bun case duration | Full-file result |
| --- | ---: | ---: | --- |
| Base | 2875.48 ms | 2968.08 ms | 95 passed / 3 failed, exit 1 |
| PR | 2621.31 ms | 2710.52 ms | 96 passed / 2 failed, exit 1 |

Both controls failed the original target's elapsed-time assertion; its failure
output supplied the original clock measurement directly. Bun case durations
also include fixture/assertion overhead. The baseline failure persists without
the diagnostic spy.

## Scope and disposition

Between these commits the failing test and its synchronous masking path are
unchanged: `src/server/responses/terminal-error-redaction.ts`,
`src/lib/outbound-credential-registry.ts`, `src/lib/redact.ts`,
`src/server/sse-payload-rewrite.ts`, `src/server/bare-upstream-error.ts`, and
`src/lib/errors.ts` have no diff. The timed callback calls the terminal-error
rewrite; it does not call `relayResponsesSseWithTerminalRepair`. The original PR
changes that relay, its tests, and the byte-accounting documentation.

This supports reporting baseline timing sensitivity rather than claiming a
confirmed terminal-repair regression. It does not justify changing credential
handling or weakening its content or bounded-work contracts. This follow-up
therefore adds the comparison record only. The existing required CI failure
and maintainer hold are retained; passing local runtime/structure checks from
the original PR are not presented as a replacement for that failed CI.

Raw logs, the detached worktree, and machine-specific results remain in ignored
scratch space; no credentials, user paths, profiler data, or CI log dumps are
committed.
Loading