Repository navigation
fix: preserve CLI evidence through interruption and output reuse - #19
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: 9e8904bd3d
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".
|
@codex review |
|
@codex review |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: dada7c5f28
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".
|
Exact final head The two valid initial hosted Codex findings have concrete fix/regression replies and resolved threads: checkpoint writer/recovery bounds (c97bc86) and nonregular blocking input (8a514e8). A fresh supplemental hosted Codex review was requested on the exact final head and remains distinct from Copilot. Required GitHub Copilot review remains blocked: documented This is stacked on PR18 at |
|
@codex review |
|
Codex Review: Didn't find any major issues. Can't wait for the next one! Reviewed commit: ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
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". |
|
Final head The supplemental hosted Codex review completed on this exact head with no new findings: #19 (comment) . All three earlier valid Codex findings have fixes, meaningful regressions, concrete replies and verified resolved discussions (writer/recovery bounds, nonregular input blocking, and fetcher/recovery HTTP status domain). This is not a GitHub Copilot review. GitHub Copilot remains a blocker: requests returned success but no reviewer registered, no review_requested event exists, and no completed Copilot review covers this head. Parent PR18 at |
Fixes #15
The CLI previously opened report paths with truncation before checking targets, and Ctrl-C could leave invalid JSON or completed HTML rows only in memory. File runs now refuse reused paths, atomically checkpoint every completed row with fsync before progress, and atomically publish a fully rendered final artifact.
--forceexplicitly permits final replacement; checkpoints remain no-clobber.--recoverrenders saved results offline with explicit incomplete/completed/pending counts and never reruns targets.CLI JSON now wraps unchanged result dictionaries under
resultsplus run metadata; CSV appends run fields and empty summaries, and HTML labels run completeness. README, reference and installed skill document the format migration, conservative retained checkpoints and 64 MiB recovery bound. Stdout cannot recover from forced termination; in-progress uncommitted results and hardware storage failure remain outside durability guarantees.Validation: six initial regression methods fail the unchanged parent; additional checkpoint-bound/nonregular-file regressions pass; 182 local tests pass (25 Windows platform skips); all eleven focused durability tests pass under Linux via WSL. Coverage includes abrupt child-process exit followed by offline HTML reconstruction, all-format partial recovery, existing-output preservation, failed explicit replacement, publication races, malformed recovery and CSV formula protection. Actual CLI help, skill validator and full committed whitespace checks pass. No external requests, scans, credentials or production changes.
Dependency: stacked on PR #18 at immutable
0915b00da362b9c08a5ba7f52771c53c969f48ff; this PR targetsfix/audit-14-origin-scopefor a focused diff. Parent must actually merge before this PR, followed by coordinator base reconciliation and fresh affected checks/review. Hosted Copilot review of the exact final head and required CI are still mandatory; no merge authorized by this PR body.The final recovery regression also covers a real loopback HTTP999 response: recovery accepts the fetcher's exact-integer100..999/null domain while still rejecting unsafe types and exploitability attestation.