Skip to content

Commit 190a457

Browse files
committed
merge dev into preview for the 2.29.0 preview sync
2 parents d2c700c + 7881319 commit 190a457

279 files changed

Lines changed: 16821 additions & 889 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.
51.6 KB
Loading

‎README.md‎

Lines changed: 9 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -129,8 +129,16 @@ see the [installation docs](https://opencodex.me/getting-started/installation/).
129129

130130
- **Use any LLM with Codex, Claude Code, Claude Desktop, and Grok Build** — 40+ providers out of
131131
the box, each keeping its own native UI.
132-
- **Pool ChatGPT accounts safely** — thread affinity, quota-aware auto-switching, cooldown and
132+
- **Pool ChatGPT accounts** — thread affinity, quota-aware auto-switching, cooldown and
133133
fail-closed auth handling.
134+
135+
> **Provider-policy note:** Account pooling is for routing and operational resilience only; it does
136+
> not guarantee protection from provider rate limits, enforcement, suspension, or other account
137+
> actions. OpenCodex does not endorse using additional accounts to circumvent provider limits or
138+
> sharing account credentials between people. You are responsible for complying with each
139+
> provider's current terms. See the
140+
> [Codex Auth account-pool guidance](https://opencodex.me/guides/web-dashboard/#codex-auth-and-account-pools)
141+
> and [OpenAI's current Terms of Use](https://openai.com/policies/terms-of-use/).
134142
- **Combos** — one virtual model id with failover or weighted round-robin across providers. See
135143
the [combo guide](https://opencodex.me/guides/combos/).
136144
- **Sub-agents on any model** — feature routed models in Codex's sub-agent picker, with v1/v2

‎devlog/_plan/260814_bug_resolution_campaign/030_wave3_cursor.md‎

Lines changed: 15 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -48,6 +48,21 @@ Cursor tool/continuation/edit 경로의 correctness fix를
4848
- teardown 문제 (정상 완료를 aborted/expectedClose:false로 기록)는
4949
별도 작은 PR로 먼저 고친다
5050

51+
2026-08-18 로컬 조사/prototype 메모 (fix/cursor-checkpoint-continuation, 아직 upstream PR 아님):
52+
53+
- 병목의 1차 원인은 JSON 포맷 자체가 아니라, 매 턴 rootPromptMessages/conversationTurns로
54+
과거 대화를 다시 만드는 full replay semantics다.
55+
- ConversationStateStructure checkpoint를 다음 conversationState로 재사용하면 no-tool
56+
follow-up에서 로컬 rootBytes가 history와 같이 커지지 않는다. grok-4.6 live 3턴에서
57+
2·3턴이 continuationMode=checkpoint였고 ALPHA-7을 기억했다.
58+
- 공식 cursor-agent 같은 계정 대조: 1턴 cacheReadTokens 0 / input 18937, 같은 세션 2턴
59+
cacheReadTokens 18816 / 새 input 331 / 답 ALPHA-7. OpenCodex Cursor wire는 usedTokens만
60+
주므로 이쪽 usage로 cache hit를 주장하면 안 된다.
61+
- tool-result는 마지막 정상 완료 턴 checkpoint + suffix replay가 live에서 동작했다.
62+
client-tool suspend 턴 자체는 온전한 checkpoint가 없어 commit하지 않는다.
63+
- 아직 미해결: 큰 context / 429 / kimi-k3 premature completion 재현, stateful live MCP
64+
bridge, 정상 완료 teardown을 aborted로 분류하는 별건.
65+
5166
### Step 5: #1623 분할 (behavior fix 안정화 후)
5267

5368
1. refactor/adapter-registry-authority
Lines changed: 119 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,119 @@
1+
# 090 — Merge log: the bug-PR backlog landing on dev
2+
3+
Unit: 260820_bug_pr_backlog_consolidation
4+
Work-phases: wp21-wp25.
5+
dev before: `31ee7a683`. dev after: `a584890f8`.
6+
7+
## The override that was not needed
8+
9+
All 19 PRs sat at CHANGES_REQUESTED from @Ingwannu with green CI, and the plan was to merge
10+
with admin authority. Three independent read-only lanes read every blocking review first, and
11+
the answer was the same in all three: **the objection was factually current, not stale.**
12+
13+
The recurring complaint was "this head is N commits behind dev". Measured, it was true
14+
everywhere — 16 to 25 commits, and `dev` had itself advanced to `31ee7a683` while the reviews
15+
were being written. A green check on a stale head validates an integration state that no longer
16+
exists, which is a real merge-readiness defect rather than a formality to wave through.
17+
18+
So the resolution was to rebase all 19 branches onto current `dev`, not to override. Admin
19+
authority can bypass a gate; it cannot make an untested integration state tested.
20+
21+
Two reviews named genuine code defects, and both were fixed rather than overridden:
22+
23+
- **#2166** — `addRequestLog` is exported and bypassed the sanitizer: the reviewer reproduced a
24+
raw 111-character value in the `/api/logs` ring against a sanitized 37-character value on
25+
disk. Fixed at the shared ingress so both surfaces read from one normalized entry, with the
26+
direct-ingress regression the reviewer asked for.
27+
- **#2162** — a prompt mutation writing into the outbound user turn needed its content-shape
28+
boundaries pinned. Added: an already-framed turn stays single, image-only content keeps its
29+
image block behind the preamble, assistant-only block content keeps its tail before the
30+
synthesized `(continue)` turn.
31+
32+
One more objection dissolved on rebase: #2148 was carrying eight `devlog/_plan/` planning files
33+
inherited from its branch point. The rebase removed them.
34+
35+
## What landed
36+
37+
19 PRs, merged bottom-up. Every merge commit verified present in `git log origin/dev`.
38+
39+
| PR | dev merge commit | Absorbed from |
40+
|---|---|---|
41+
| #2134 | `930840ca4` | maintainer fix |
42+
| #2160 | `114e9e543` | #2067 @waw4303 |
43+
| #2162 | `087c3c368` | #2082 @yzxcj797 |
44+
| #2164 | `31750b094` | #2027 @yzxcj797 |
45+
| #2165 | `41689b374` | #2155 @waw4303 |
46+
| #2166 | `5fbe65570` | #2163 @Ingwannu |
47+
| #2137 | `be12328bc` | issue #2132 |
48+
| #2146 | `aa07bc308` | #2101 @Ingwannu |
49+
| #2138 | `81492fd10` | #2102 @lilinxiong |
50+
| #2140 | `3ad9c7bf4` | #2100 + #2077 @ntdatt812 |
51+
| #2141 | `1cc35c560` | #2056 @Ingwannu |
52+
| #2142 | `52a463dd6` | #2131 @bet4it |
53+
| #2144 | `8c8a66816` | #2105 @lilinxiong |
54+
| #2145 | `83d5ffa3c` | #2040 @Ingwannu |
55+
| #2147 | `7fc50846b` | #2104 @olddonkey |
56+
| #2148 | `86ed9ed83` | #2109 + #2110 @drakonkat |
57+
| #2149 | `17e8e916b` | #2053 @Ingwannu |
58+
| #2150 | `9a7801547` | #2127 @agentHits |
59+
| #2151 | `a584890f8` | #2075 @olddonkey |
60+
61+
Verification that the rebased content is what actually landed, split by what is still
62+
re-runnable:
63+
64+
- **Re-runnable today:** every merge SHA in the table above satisfies
65+
`git merge-base --is-ancestor <sha> origin/dev` (verified 2026-08-21, 19/19 true) — the
66+
recorded merge commits are exactly the commits on `dev`, so the landed content is the
67+
table's content by construction.
68+
- **Historical assertion, no longer re-runnable:** six of the merges recorded a pre-rebase
69+
branch SHA in their description. At merge time each rebased branch was compared with
70+
`git diff --name-only origin/dev <branch> -- <that PR's own src/ and tests/ paths>` and
71+
returned 0 differing files. The source branches were deleted in the wp0 cleanup
72+
(260820 unit, executed record), so those comparisons cannot be reproduced from this log;
73+
they stand as recorded assertions, not evidence, and the re-runnable ancestor check above
74+
is the durable audit trail.
75+
76+
## Security surfaces, named rather than merged silently
77+
78+
`MAINTAINERS.md` reserves auth, credential handling, OAuth, workflows, and release automation
79+
for explicit human review. Seven of the merged PRs touch that surface, and the user's merge
80+
authorization is the human decision of record for each:
81+
82+
- **#2137 / #2146** — bearer admission and stored-credential substitution; entitlement-gated
83+
model discovery sends the selected account's access token.
84+
- **#2144** — decides whether the shell hook exposing `ANTHROPIC_AUTH_TOKEN` is installed.
85+
- **#2145** — lowers third-party `function_call` output and restores it as a client-executed
86+
private `tool_search_call`.
87+
- **#2147** — OAuth 401 refresh and access-token replay.
88+
- **#2148** — operator-selected destinations for Anthropic/Antigravity OAuth bearers. The
89+
transport gate still rejects public cleartext HTTP; only explicitly opted-in local/private
90+
relays may use it.
91+
- **#2149** — OAuth credential commit ownership under the store lock.
92+
93+
## Issue closure
94+
95+
Every issue named by a merged PR is closed: #2133, #2132, #2092, #2047, #1950, #2097, #1886,
96+
#2125, #2074, #1924, plus the superseded contributor PRs. Verified by `gh api` state, not by
97+
assuming GitHub auto-closed them — these PRs targeted `dev`, not `main`, so auto-close does not
98+
fire.
99+
100+
## Release readiness — not a release
101+
102+
At `a584890f8`:
103+
104+
- `bun run test` — 13716 pass / 10 skip / 0 fail across 866 files.
105+
- `bun x tsc --noEmit` — exit 0.
106+
- `bun run privacy:scan` — passed.
107+
- `package.json` version line: **2.27.0** (unchanged by this campaign).
108+
109+
Changed surfaces: provider registry and transport headers, quota dispatch, routing capability
110+
resolution, the Anthropic/OpenAI-chat/Google adapters, Responses core and compact, OAuth store
111+
and credential commit, request logging and usage persistence, subagent roster management.
112+
113+
Not done, and deliberately: no `scripts/release.ts`, no npm publish, no tag, no change to
114+
`main` or `preview`. Release execution needs its own authorization.
115+
116+
## Still open
117+
118+
**#2054** (@keepitmello) stays open by explicit instruction, carrying the wire-probe request.
119+
**#2167** (@ntdatt812) arrived after this campaign and is untriaged.
Lines changed: 105 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,105 @@
1+
# 110 — Three open bug issues: #2152, #2157, #2156
2+
3+
Unit: 260820_bug_pr_backlog_consolidation
4+
Verification host: `ssh lidge:~/ci-wp3/opencodex`.
5+
6+
Three issues, three different shapes of answer. Two shipped fixes; one is honestly blocked.
7+
8+
## #2152 — Windows CI, three groups (PR #2178)
9+
10+
Three read-only lanes read the three groups. The shape PR #2178 already had was right, and two
11+
of the issue's own premises turned out to be wrong — the fix follows the evidence:
12+
13+
- **Group 1** is not "the case budget is too small". `A-reduced` failed at 79,978 ms against a
14+
150 s ceiling, so the outer budget was never the constraint. The real abort came from
15+
`Fixture.request`'s unscaled 10 s `AbortSignal` firing from inside. And `E` does not start
16+
`ocx` at all — it starts two lock helpers, and its holder released after a fixed 3 s wait
17+
that a Windows contender's spawn can outlast.
18+
- **Group 2** is not "an unprivileged Windows user cannot create symlinks". The runner can, so
19+
`canSymlink` was true and the cases ran — then failed on Unix mode semantics that a Windows
20+
directory cannot satisfy. Production already returns `windows_skip` for exactly that reason,
21+
which is what makes the platform guard a correct skip rather than a masked failure.
22+
- **Group 3** needed the crash retry the macOS leg already had.
23+
24+
### The defect I found in the fix
25+
26+
Group 3's retry grepped for `panic(thread`. This repository already worked that out and wrote
27+
it down: `devlog/_fin/260731_pr_issue_triage_round/050_windows_ci_flake_rca.md:172` says not to
28+
key on it, because Bun emits **both** `panic(thread 2852)` and `panic(main thread)` for the
29+
same failure, and names `Internal assertion failure` as the stable fingerprint. Probed:
30+
31+
```text
32+
MATCH panic(thread 3960): Internal assertion failure
33+
MISS panic(main thread): Internal assertion failure
34+
MISS panic(main thread): PANIC: reached unreachable code
35+
```
36+
37+
The retry would have failed the shard on roughly half the crashes it exists to absorb, while
38+
looking correct.
39+
40+
Three copies of that signature list exist — macOS inline, the new Windows inline, and
41+
`is_bun_runtime_crash` — and the workflow comment already said to keep them in sync with
42+
nothing enforcing it. They had drifted. All three now match, and the contract test pins **the
43+
sync itself**, not the literal text, so the same drift cannot recur. RED-proven: restoring the
44+
bad signature fails it with `windows:Internal assertion failure:false`.
45+
46+
`hasShellCommandHead` was added because the existing exact-whole-line matcher rejected the
47+
`| tee` the retry requires — that is why the Windows step assertion went red — while still
48+
rejecting an echoed or commented-out copy.
49+
50+
**What no local run can prove:** whether 45 s suffices under real Windows contention, the actual
51+
skip result on the runner, and `PIPESTATUS` under Git Bash. Those need a `workflow_dispatch`,
52+
and a useful proof run must exercise the crash path — a green Windows run shows the suite runs,
53+
not that the retry fires.
54+
55+
## #2157 — shadow-helper observability (PR #2179)
56+
57+
The dashboard half of the attribution field #2166 landed. Delivered.
58+
59+
## #2156 — muse-spark truncation: BLOCKED, deliberately
60+
61+
PR #2180 was opened claiming to fix this. An adversarial review found the attribution wrong, and
62+
the source agrees:
63+
64+
- A stall abort emits `response.incomplete` / `upstream_stall_timeout` (`bridge.ts:1371-1396`),
65+
after which the bridge has cancelled upstream, closed, and explicitly discards any late
66+
adapter event (`bridge.ts:837-845`).
67+
- The reporter's error is emitted only after `reader.read()` returns EOF with tool calls still
68+
pending (`openai-chat.ts:1819-1827`), surfacing as `response.failed`.
69+
70+
Different path, different client frame. The heartbeat **cannot** produce the reported error.
71+
72+
What the heartbeat does fix is real and worth landing alone: tool-call deltas are buffered, the
73+
bridge arms its watchdog on adapter activity rather than socket activity, so a large argument
74+
payload was indistinguishable from a hung upstream. The `#2156` references were reworded to
75+
"found while investigating", and the closing keyword removed.
76+
77+
**Second finding, partially addressed.** Making the adapter emit one heartbeat per delta
78+
exposed retention in `guardTerminalEventStream`'s `seen`, which feeds both the continuation
79+
analysis and the rebuilt request. Two event classes must be distinguished: #2180 stopped
80+
retaining `heartbeat` events, but the terminal guard still retains every OTHER nonterminal
81+
event (tool-call deltas included), so a large argument payload can still grow `seen` without
82+
bound wherever `terminalContinuationGuard` is on. The empty-completion guard passes heartbeats
83+
through unretained; matching it for the remaining nonterminal classes is follow-up work, not
84+
something this record's PR landed.
85+
86+
**Why blocked rather than fixed.** The adapter is reporting truthfully: that stream really did
87+
end. What cannot be determined from here is why it ended for ocx and not for Pi direct.
88+
`hadUsage: false` is suggestive — ocx does send `stream_options.include_usage` — but a provider
89+
may simply ignore that flag, so it is not decisive either way.
90+
91+
Asked the reporter for the one thing that settles it: redacted raw SSE captures from a
92+
Pi-direct success and an ocx failure for an equivalent request, through socket close, and
93+
whether either carried `finish_reason`, `[DONE]`, or a usage-only final chunk. If ocx's
94+
upstream closes without a terminal frame while Pi's does not, the difference is in what we send
95+
or how we read it and it is ours. If both close identically and Pi is merely more tolerant, the
96+
right answer is the buffered-mode fallback the reporter suggested — and that choice should rest
97+
on their capture, not on a guess.
98+
99+
## Verification
100+
101+
At the branch tips, on `ssh lidge`:
102+
103+
- #2178: `bun test` 13719 pass / 15 skip / 0 fail; `tests/ci-workflows.test.ts` 132 pass / 0 fail.
104+
- #2180: `bun test` 13722 pass / 15 skip / 0 fail; focused trio 112 pass / 0 fail.
105+
- `bun x tsc --noEmit` exit 0 and `bun run privacy:scan` passed on both.

0 commit comments

Comments
 (0)