Repository navigation
Conversation
Codex Review SummaryThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 47371bbf67
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".
| raise InitError("foundation shown packet exceeds generation byte cap") | ||
| start, end = item.get("start", 1), item["end"] | ||
| try: | ||
| raw = stage.observer.file_text(stage.record.pin, item["path"]) |
There was a problem hiding this comment.
In an unlimited foundation run with multiple requested facets, this rereads every offered file for each judge call because _judge_evidence calls foundation_evidence per section, and coordinator/native validation calls it repeatedly again. PinnedObserver.file_text starts separate git ls-tree and git cat-file processes, so a maximal 66-file packet with six accepted facets causes thousands of Git subprocesses per job; large feature catalogs can therefore become prohibitively slow or exhaust process resources. Validate the immutable packet once per job/replay context and reuse the resulting evidence for its facet reviews and receipt checks.
Useful? React with 👍 / 👎.
Unlimited foundation extraction offered adjacent source chunks and full bounded files/docs, but citation checks rejected ranges spanning chunk seams and the independent reviewer saw only selected excerpts. Review now receives the same pinned, numbered source/doc packet as extraction, so existing support in callers or configuration remains visible.
The new unlimited prompt version 4 keeps source at 100,000 bytes and docs at 30,000 bytes, validates every shown line and selected anchor, and asks for at most two concise claims without unsupported absence assertions. Native replay binds the exact generator system/input and full reviewer packet. Historical version 3 receipts retain their original system, selected-evidence protocol and raw interval identity; finite-mode prompts remain unchanged.
Validation: All 145 combined context, foundation, native approval, checkpoint/publication, knowledge coverage and subscription cases passed on exact commit 47371bb. Independent source/native review found no actionable issues and passed 129 focused cases. Tests include adjacent/overlap versus real gaps, late and noncited lines, source/doc caps, pin/text/header/read failures, genuine version 3 replay, version 4 native context binding, unchanged rejection, and interrupted/prepared replay. Knowledge and adapter tree hashes are unchanged from the base.