Repository navigation
Return complete worker answers and verify reads before cleanup - #156
pepechka-kun wants to merge 5 commits into
Conversation
Worker answers could be truncated or lost at cleanup, and callers repeated result reads. Preserve bounded masked answer pages, count completed waits as reads and return owner-scoped closed-child receipts.
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. |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 113adccd61
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".
113adcc to
8fb947b
Compare
Captured replies hid process failures, and staging a draft forced rereading an unchanged reply. Preserve bounded failure details and invalidate read receipts only for submitted tasks.
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 8822bba38b
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".
Closing a parent disposed unread descendants and lost their results. Validate the whole disposal branch before closing any card and retain scoped closed receipts for every removed child.
Only agent descendants expose captured results. Skip result lookup for plain terminal descendants while still validating and disposing the entire branch.
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 653de7f3a6
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".
…encodings A failed worker could exceed the MCP response cap after its page was marked read. Bound the complete co-returned result before recording a read receipt, preserving diagnostics and sequential answer pagination.
Superseded by #158, a smaller bugfix-only PR from current main. Optional features and workflow/security-policy changes from this proposal are not included in the replacement. This branch and discussion remain available for reference.
Worker results could be truncated or cleared before the orchestrator read them, while callers repeated wait/get calls. Return bounded masked answer pages, count a completed wait as a read and require the complete current answer before child cleanup.
Failed exits now retain bounded terminal diagnostics beside captured answers. Staging an unsubmitted draft preserves existing read receipts; submitted tasks and changed answer revisions still invalidate them. Unread or partially read answers remain protected from cleanup, including all descendants removed by a cascading parent cancellation. The entire branch is checked before disposal, with scoped closed receipts for removed descendants; plain terminal descendants do not require agent answers. Answer pages include co-returned exit diagnostics in their budget across both JSON encodings, before any read receipt is recorded.
Validation: Latest follow-up: 42 orchestration/delegation tests passed, 1 native-helper test skipped; earlier cascade checks: 59 passed, 1 skipped; typecheck passed. Coverage includes failed/successful exits, short and paginated answers, unsubmitted drafts, submitted follow-ups and cleanup. No running services or profiles changed.
Combined local regression run with all three companion topics: 2,355 passed, 27 skipped, no failures. Typecheck and source secret audit passed.
Companion topics: #155 and #157; each is based independently on main.