Skip to content

Commit 7bdd1b2

Browse files
authored
Merge pull request #4617 from lidge-jun/codex/260914-preview-2550
[WRONG BRANCH] chore(release): promote 2.55.0-preview.20260914 to preview
2 parents 4d633cb + 704857f commit 7bdd1b2

293 files changed

Lines changed: 17559 additions & 927 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.

devlog/_plan/260913_contributor_carry_train/000_plan.md renamed to devlog/_fin/260913_contributor_carry_train/000_plan.md

File renamed without changes.

devlog/_plan/260913_contributor_carry_train/001_candidate_inventory.md renamed to devlog/_fin/260913_contributor_carry_train/001_candidate_inventory.md

File renamed without changes.

devlog/_plan/260913_contributor_carry_train/010_wave1.md renamed to devlog/_fin/260913_contributor_carry_train/010_wave1.md

File renamed without changes.

devlog/_plan/260913_contributor_carry_train/011_wave1_outcome.md renamed to devlog/_fin/260913_contributor_carry_train/011_wave1_outcome.md

File renamed without changes.

devlog/_plan/260913_contributor_carry_train/020_wave1_merge.md renamed to devlog/_fin/260913_contributor_carry_train/020_wave1_merge.md

File renamed without changes.

devlog/_plan/260913_contributor_carry_train/030_wave2.md renamed to devlog/_fin/260913_contributor_carry_train/030_wave2.md

File renamed without changes.

devlog/_plan/260913_contributor_carry_train/040_wave2_merge_regression.md renamed to devlog/_fin/260913_contributor_carry_train/040_wave2_merge_regression.md

File renamed without changes.

devlog/_plan/260913_contributor_carry_train/050_disposition.md renamed to devlog/_fin/260913_contributor_carry_train/050_disposition.md

File renamed without changes.

devlog/_plan/260913_contributor_carry_train/060_outcome.md renamed to devlog/_fin/260913_contributor_carry_train/060_outcome.md

Lines changed: 37 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -1,7 +1,8 @@
11
# Outcome — the contributor carry train
22

33
Twelve lanes were dispatched to land the open contributor work scored 60 or
4-
higher. Eleven landed, one needed nothing, and one is recorded separately below.
4+
higher. Eleven landed and one needed nothing, because the work it was sent to carry was
5+
already on dev.
56

67
## What landed
78

@@ -18,6 +19,7 @@ higher. Eleven landed, one needed nothing, and one is recorded separately below.
1819
| 2 | I3 | #4500 | 94063d0798 | #4467 jaychou0642-create, issue #3775 |
1920
| 2 | I4 | #4498 | 8e6c99608c | issue #4454 |
2021
| 2 | H | — | — | nothing to carry; #3663 was already on dev |
22+
| 2 | I5 | #4515 | cb2e15ba6f | issue #4429, scoped slice |
2123

2224
Every merge used the same gate: a Cross-platform CI run concluded success on the
2325
exact tip head SHA, the merge commit was verified with
@@ -89,6 +91,40 @@ Both directions are the same defect in the plan, not in the lanes: a snapshot of
8991
the open queue is stale the moment it is taken, and only the lane touching the
9092
code can tell.
9193

94+
## The lane that did not close its issue
95+
96+
Lane I5 is the one outcome in this train that is deliberately partial, and it is
97+
the better result.
98+
99+
Issue #4429 reads as a missing executor: a key-auth gateway echoes hosted web_search as
100+
a client function_call, and webSearchBridge had an executor only for ollama.
101+
Arming the other five backends is what #4515 landed. It does not fix the reported
102+
failure, and the lane said so rather than closing the issue.
103+
104+
The reporter's own probe is why. It ends with two pending client calls, exec and
105+
web_search, which the bridge refuses with web_search_bridge_mixed_tools. The
106+
boundary is the mixed-tool leg, not the missing backend, so a continuation has to
107+
preserve the client's exec call and call_id and their ordering without executing
108+
it proxy-side and without losing hosted-search items the relay already completed.
109+
The issue stays open for that, with the scoping recorded on it.
110+
111+
The DeepSeek case in the same thread stays separate on purpose: it emits no
112+
function calls at all, only assistant text, and making matching prose executable
113+
would turn model output into tool execution.
114+
115+
A scoped slice with an accurate description is what the packet asked for, and
116+
refusing to close the issue is the part that makes it honest.
117+
118+
## Final integration evidence
119+
120+
Cross-platform CI run [34760250023](https://github.com/lidge-jun/opencodex/actions/runs/34760250023)
121+
completed successfully on `cb2e15ba6f8ac17af0620d6ff04fcfa7d88e3dcd`, the
122+
merge commit for the last implementation PR, #4515. This verifies the integrated
123+
batch; conditional jobs remain skips rather than claimed passes.
124+
125+
Issue #4429 remains partially unresolved: the non-Ollama executor slice landed, while
126+
mixed-tool continuation remains open. Issue #4519 separately tracks the Ollama endpoint destination-policy gap; it is not the mixed-tool continuation tracker.
127+
92128
## Honest limits of the proof
93129

94130
Non-tip pull requests merged without their own `ci` check, under the recorded
Lines changed: 64 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,64 @@
1+
# 260914 — Round 2/3 delivery roadmap
2+
3+
## Where this starts
4+
5+
The 2026-09-14 triage delivery loop closed with #4519, #4534, #4535, #4536,
6+
#4539, #4555, #4562 and #4563 on dev. What it did not do is reduce the open
7+
surface: 63 pull requests and 63 issues are still open, and a large share of the
8+
pull requests are contributor work that is already green and simply never got
9+
landed. This unit plans two more delivery rounds against that backlog.
10+
11+
## Objective
12+
13+
Rebuild the lane roadmap, run two merge rounds, and finish with a closure sweep
14+
that closes at least 10 issues and at least 10 pull requests with written
15+
reasons. Every close names either a merge commit or the change that supersedes
16+
it.
17+
18+
## Delivery constraints carried from the session
19+
20+
- Lanes are **worktree Codex threads**, one per branch. Subagents inside a lane
21+
are that lane's workers; they never own a branch of their own.
22+
- Lane orchestrators split between `anthropic/claude-opus-5` and `kimi/k3[1m]`.
23+
- Inside a lane, implementation is delegated to `devin/swe-2` and
24+
`xai/grok-4.6` subagents at roughly a 2:3 ratio.
25+
- Each lane runs `cxc-loop` itself: its own goal, its own PABCD cycle.
26+
- **No local suite.** `bun run test`, `bun run typecheck` and `bun install` are
27+
not run locally. Proof is hosted CI at the exact final head, nothing else.
28+
- Pushes use `--no-verify`.
29+
- Merges into dev are admin squash merges, taken on maintainer judgment under
30+
the MAINTAINERS.md single-maintainer dev integration policy, with exact-head
31+
CI recorded.
32+
- Landing another author's work carries a `Co-authored-by` trailer in a branch
33+
commit so it survives the squash.
34+
35+
## Round shape
36+
37+
Two merge rounds, then one closure sweep.
38+
39+
| Round | Work | Merge target |
40+
|---|---|---|
41+
| 1 (wp2) | 5 lane threads on issues that have no pull request, plus review-and-land of the already-green contributor queue | dev |
42+
| 2 (wp3) | 4 lane threads on the second issue cluster, plus the second contributor batch | dev |
43+
| Sweep (wp4) | Close resolved issues with merge references, close superseded and abandoned pull requests with reasons | — |
44+
45+
## Acceptance criteria
46+
47+
1. This roadmap unit exists with per-lane issue assignment, dispatch models and
48+
acceptance criteria.
49+
2. Round 1 lanes were dispatched as separate worktree threads and their pull
50+
requests reached green hosted CI at their exact final head.
51+
3. Round 1 pull requests were admin squash-merged into dev with recorded merge
52+
commits.
53+
4. Round 2 lanes were dispatched and squash-merged into dev the same way.
54+
5. At least 10 issues are closed with a merge-commit reference or a written
55+
supersession reason.
56+
6. At least 10 pull requests are closed — merged, or closed with a written reason
57+
naming the superseding change.
58+
59+
## What would make this fail
60+
61+
The backlog is contributor work, so the failure mode is not "no code lands", it
62+
is "code lands that nobody reviewed". A green check on a three-day-old head is
63+
not evidence about the head being merged. Every merge in this unit re-reads the
64+
diff at the head it is about to squash, and re-reads CI at that same SHA.

0 commit comments

Comments
 (0)