Codex File Inspector 0.2.8

This release improves everyday reading and search reliability, especially bounded reads, contextual pagination, error recovery, and short-lived ripgrep processes.

Improvements

• More efficient bounded reads. Large requested ranges avoid repeatedly rebuilding oversized results, reducing temporary allocations. Cancellation is checked during output budgeting and before a successful read is returned.
• Correct CRLF searches. Regex line anchors such as ^MATCH$ work with Windows CRLF text. Matching lines, context, counts, pagination, and subsequent read_file locations remain consistent for LF and CRLF files.
• Reliable pagination. Grep trims context against the matches retained on the current page and preserves the same file order across whole and paginated results. A minimum result that cannot fit the output budget fails explicitly instead of returning an empty success page with a continuation that cannot advance.
• Better failure classification. Already-observed traversal failures remain partial results even when a search stops early for pagination. Error-like words in filenames no longer turn valid searches into query errors. Invalid glob diagnostics identify the input item when it can be determined reliably.
• Clearer invalid-input handling. Validation now covers undefined enum values, null required glob arrays, device-namespace separator variants, and command lines that grow too large after Windows argument escaping. Invalid Windows names are classified as path errors rather than transient I/O failures.
• Useful recovery guidance. Unsupported lookaround and backreference queries explain the existing Rust regex limitation. Certain failed starts with a long working directory now include recovery guidance and discourage repeating the same request unchanged, while retaining the native error code and requested path. These messages do not add PCRE2 or automatically rewrite searches.
• More robust process lifecycle handling. Short searches that finish before Windows Job assignment can have their real output processed under defined completion checks. Their actual exit codes remain significant even when buffered output is paginated. Running or unconfirmed unassigned processes and other Job failures still fail explicitly; failed-start cleanup attempts every release step without masking the original error.
• Startup and instruction improvements. Unrelated project appsettings files no longer affect Server startup. Setup documentation offers two alternative consumer policies: MCP-first with practical Shell fallback, or autonomous selection between MCP and Shell. The user chooses and applies one; the Server does not edit user instructions.

The four tool APIs—read_file, grep, glob, and list_directory—and their structured result schemas are unchanged. Inspected files remain read-only. Optional failure diagnostics remain disabled by default, and successful calls are not logged.

Known limitations

• Very long search roots can still exceed Windows process-start limits. Recovery guidance is a heuristic, not automatic long-path support. Searching from a shorter ancestor requires checking relative globs and ignore/hidden behavior; changing the root does not automatically preserve filtering semantics.
• With ignore rules enabled, parent .gitignore, .ignore, or .rgignore rules outside the nearest repository can still influence searches. The ignore-source boundary has not been changed in this release.
• Standalone CR line endings, including mixed files containing standalone CR, do not yet have fully consistent line positions and counts between grep and read_file. The CRLF improvement does not resolve this separate limitation.
• Job assignment failures for running processes or other unsupported conditions remain explicit errors; completed-process handling is not a general bypass.

Distribution

The Windows x64 build is framework-dependent and requires the .NET 10 x64 runtime. It includes ripgrep 15.2.0. Keep the complete extracted directory together, including assemblies, runtime assets, bundled ripgrep, and license files. Binaries are unsigned.

• CodexFileInspector-0.2.8-win-x64.zip
• CodexFileInspector-0.2.8-source.zip
• SHA256SUMS for the release archives

See the included README for installation, upgrade instructions, and the two consumer-policy templates.

Validation

The exported public source built with zero warnings or errors and passed 388/388 Release tests. Both the published executable and the executable extracted into a separate directory passed 19/19 STDIO tests, with no skipped tests.

All 155 archive entries—101 source and 54 binary entries—were rehashed against the packaging inputs. Extracted files matched their verification records. The runtime-license inventory for 32 packages, all 11 relative documentation links, and release-hygiene checks passed.

The core Server DLL is identical to the 0.2.8 build previously checked through 24 live Codex Host calls with expected outcomes, including intentional error cases. Those Host checks were not repeated as part of this archive-verification run.

Automated tests cover cancellation, but cancellation during a running live Codex Host call was not exercised. These are local verification results, not broad cross-machine certification or an independent security audit.