The provider now lists gpt-6.2-sol at the top of every account's /models
catalog. When that request fails (it has 5 s, DNS included), Xcoder falls
back to its bundled table, which only knew GPT-6.1-Sol: sessions without a
configured model started on 6.1, and 6.2 ran on fallback metadata
("Model metadata for `gpt-6.2-sol` not found").
Add the provider's own gpt-6.2-sol entry to the bundled table with
priority 0, so it is the default with or without the live catalog. A test
checks that a failed catalog fetch defaults to the newest bundled Sol.
Windows support is removed: the Windows sandbox crates, PowerShell and
cmd shell support, Windows-only code paths, config keys, app-server
requests, TUI prompts, the standalone Windows updater and the Windows CI
and installer scripts.
The TUI header, /status, the exec banner, analytics and turn metadata
show the tracked Codex version (at least 0.169.0) instead of 0.0.0.
Builds:
- Ziran build scripts no longer watch the missing ziran.local.toml;
Cargo reran them on every build, recompiling most of the workspace.
- The build wrapper no longer caps builds at 2 GiB RAM in its own scope;
memory comes from the surrounding build queue.
- make builds a `local` profile (no LTO, 16 codegen units, incremental);
published releases keep `release` (XCODER_CARGO_PROFILE=release in CI).
- codex-protocol depends on a small codex-network-policy crate instead
of the network proxy, the slowest crate in the build.
- Unused dependencies are removed.
A rebuild after a change in codex-config now takes about 90 s, and the
full local suite builds and passes in under 3 minutes.
Diffs come from the Ziran diff package (codex-utils-text-diff) instead
of similar and diffy, and the git baseline repository and remote URL
sanitizing use the Ziran git package instead of gix and gix-url. All
four remain only as test oracles: similar and diffy on generated edits
and damaged patches, gix-url on 200,000 generated remotes, and the git
CLI checks written objects with fsck --strict.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
extract_apply_patch_from_bash now runs xcoder_apply_patch_heredoc in
native/apply_patch.zi instead of a tree-sitter query. It accepts the
same two whole-script forms (apply_patch/applypatch <<DELIM ... DELIM,
optionally after `cd <one bare word, "string" or 'string'> &&`) and
returns the same body and cd path. Heredoc bodies end only at a line
that is exactly the delimiter (leading tabs stripped for <<-), as in
Bash: tree-sitter's scanner also ended them at lines that merely start
with the delimiter (EOFX), cutting such patches short. An unterminated
heredoc does not match, as before.
The query stays as the test oracle: on 20,000 generated realistic
scripts Ziran returns the same trimmed body and cd path wherever
tree-sitter's own tree has no errors; new unit tests pin the Bash rules
for look-alike, indented and unterminated delimiters. tree-sitter and
tree-sitter-bash are now dev-dependencies of codex-apply-patch.
Local tests: apply-patch (75 unit, 28 integration), core apply_patch
(29).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
parse_shell_lc_literal_commands, which dangerous-command detection uses
to look at every command in a script, now runs xcoder_literal_commands
in native/shell.zi instead of walking a tree-sitter-bash tree. It
descends into lists, pipelines, subshells, { } groups, if/while/until/
for/select/case and function bodies, $(...), backticks, <(...) and
>(...), double quotes and unquoted heredoc bodies; it drops redirection
targets and leading assignments and keeps only literal words, as the
tree walk did. Words are collected on a stack so commands nested in an
argument never interleave with it, and commands come out in post-order,
which reversed is the tree walk's order. Scripts whose compound commands
or operators do not fit together fail, as Bash refuses to run them.
The tree walk stays as the test oracle. The invariant is about safety:
whenever tree-sitter's commands make a script dangerous, Ziran's must
too, unless Bash itself rejects the script or reads that text as a
heredoc body (checked with bash -n). It holds on 40,000 generated
scripts (32,943 with identical command lists) and on 1,005 real agent
commands, where Ziran additionally flags two scripts whose rm commands
tree-sitter failed to parse.
Local tests: shell-command (195 + corpus), core danger (16) and
safety (15) tests.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
parse_shell_script_into_commands, which exec policy, command
canonicalization, plugin attribution and parse_command use to prove a
script is only plain commands joined by &&, ||, ;, | and newlines, now
runs native/shell.zi instead of walking a tree-sitter-bash parse.
The recognizer is deliberately conservative and is checked against the
tree-sitter walk, kept as the test oracle: it must never accept a script
tree-sitter rejects or return different words. The parity tests found
and pinned tree-sitter's rules for command names (a single bare word,
not a number, not dashes alone, no `=`, not a reserved word even when
followed by punctuation) and for word parts starting with `=`. On
50,000 generated scripts there are no parity problems (3,630 accepted
by both; 1,327 adversarial ones rejected only by Ziran); on 1,005 real
agent commands (local corpus, not committed) the two agree exactly.
tree-sitter-bash still serves the literal-command extraction for
dangerous-command detection, shell snapshots and apply_patch heredocs.
Local tests: shell-command (193 + corpus run), core exec_policy (104)
and command_canonicalization (4).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Core turned the streaming patch parser's full snapshot into a progress
event for every delta the model sent: every hunk was cloned and every
update diff reformatted, though most deltas are a few characters and
change nothing. Streaming a long patch was quadratic. The parser now
has feed(), which reports whether a delta completed a line, and
hunks(), which borrows the current snapshot; core skips deltas that
complete no line, so the events it emits are unchanged. push_delta
keeps its API, and the parser splits deltas on newlines instead of
pushing one character at a time.
Local tests: apply-patch (73 unit, 28 integration, including a new
test that feed matches push_delta character by character) and core's
29 apply_patch tests.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Exec-policy .rules files were evaluated by a full Starlark interpreter,
but they only ever use a declarative subset. native/execpolicy.zi now
parses that subset into a node tree: top-level calls and NAME = value
assignments over strings (quoted, raw and triple-quoted, with every
Starlark escape), lists, tuples, integers, True/False/None, names and
`+`, with comments, continuations and semicolons. A small Rust
evaluator resolves names and `+` and binds arguments to prefix_rule,
network_rule and host_executable as Starlark did (positional or
keyword, None as omitted, errors for unknown, missing, duplicate or
mistyped parameters); the existing validation is unchanged.
Functions, loops, conditions, load and f-strings are rejected with
their line. Errors keep the `path:line:column: starlark error:` shape
core parses for the location. The starlark crates and about 60
dependencies they pulled in are removed, along with two security
advisory exceptions only they needed.
Local tests: all execpolicy tests plus six new rules-language tests,
and core's 104 exec_policy tests (run with a 16 MiB test stack; one
agent-spawn test overflows the default stack in this debug build).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
derive_new_contents_from_chunks now hands the file and its chunks to
xcoder_apply_chunks in native/apply_patch.zi, which splits the file
into line offsets (keeping LF, CRLF and CR endings, the first one
preferred), finds every chunk with the Ziran matcher, collects and
stably sorts the replacements, and writes the new contents once into a
buffer sized up front. Lines are no longer copied into strings several
times. Error messages are formatted in Rust exactly as before.
The Rust pipeline (SourceFile + compute_replacements) remains as the
fallback for text above 2 GiB and as the oracle: results and error
messages match on 3,000 generated patches with mixed endings, context
lines, end-of-file and failing hunks. A 20-hunk update of a 50,000-line
file takes 3.2 ms instead of 7.7 ms (make benchmark-apply-patch).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
seek_sequence now runs in native/apply_patch.zi behind the same Rust
signature. It borrows the file's and the pattern's lines through
pointer and length tables, compares bytes with memcmp, trims Unicode
White_Space from both ends without rescanning lines, and decodes only
non-ASCII scalars. The four passes (exact, trim_end, trim, trim with
typographic punctuation normalised) and their order are unchanged.
The Rust matcher stays as the fallback for lines above 2 GiB and as
the oracle for parity tests over 4,000 generated cases, edge cases and
a 20,000-line file. On a 50,000-line file (make benchmark-apply-patch):
punctuation-tolerant and failing searches are 4-5x faster (3.9 ms vs
20.5 ms, 3.1 ms vs 16.1 ms); exact and trimmed matches cost 0.4-0.7 ms
more, mostly building the line tables.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Parallel cells could fail with "Text file busy" when another thread
forked while a freshly written cell executable still had a writer
descriptor; spawning now retries ETXTBSY briefly. The upstream Codex
version check runs on a detached thread with its own runtime, so a
stalled network cannot hold up process exit, and the offline CLI tests
and memory benchmark disable it.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Report the newest upstream Codex release to the service, separately
from Xcoder's own version. The reported version is the larger of the
built-in 0.169.0 and the latest openai/codex release, which the CLI
checks on GitHub in the background at most every 20 hours and caches in
CODEX_HOME/xcoder-upstream-version.json (XCODER_UPSTREAM_VERSION_CHECK=0
turns the check off). `xcoder --version` now shows Xcoder's own number,
0.1.0.
Fix two regressions found by the local suite:
- Shell commands that ran before the session's shell snapshot existed
used the login profile's PATH, so a profile's apply_patch could win
over Xcoder's. The Codex-owned PATH entries are now replayed after
login startup without a snapshot too.
- Tool cells failed with "unexpected end of file" when they exited,
because the callback channels a cell never used closed. A clean close
between requests now retires that channel.
The installer no longer keeps copies of the binaries it replaces. The
validation fixture covers every native test, and the dead --oss
plumbing is gone.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Idle sessions give their memory back: after XCODER_IDLE_RELEASE_SECONDS
(default 120, 0 disables) a watcher written in Ziran (native/idle.zi)
trims the allocator, pages private memory out to swap and unmaps clean
program pages. Measured with 600 tool calls: 415 MB resident drops to
23 MB, and the next keypress takes 79 ms. The TUI's replay copy of a
finished command keeps only the first and last parts of output above
16 KiB. make benchmark-memory and the scripts/memory allocation tracer
measure this.
Remove what Xcoder never uses under Harmony: the cloud, tcp-tunnel,
responses-api-proxy, exec-server and debug app-server subcommands with
their crates, the Ollama and LM Studio providers with the TUI's
local-provider picker, and the AWS SDK behind Bedrock sign-in. --oss
now stops at startup with an explanation.
Model requests identify as the tracked upstream release (0.169.0) in
their User-Agent and version header, as catalog requests already did.
A fresh ChatGPT-account session refused gpt-6.1-sol while requests
still said 0.0.0. The rule lives in codex_protocol::client_version.
Also carried over from the shared working tree: native Ziran fuzzy
matching and Mermaid adapters, Linux packaging and its workflow, and
Landlock/seccomp cell sandboxing in place of bwrap.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Xcoder keeps upstream's 0.0.0 development workspace version, and the
Codex models endpoint hides newer models such as gpt-6.1-sol from 0.0.0
clients. Catalog requests and cache entries now report the tracked
upstream release, 0.169.0, so the full model list is returned and stale
0.0.0 cache entries are refreshed.
## Why
MCP request metadata could forward a client-supplied `product_sku` to unrelated servers and connectors.
## What changed
Remove `product_sku` from turn metadata before forwarding it. For Node REPL-backed servers and the browser connector, populate it from the turn's `apps_mcp_product_sku` configuration, mapping unrecognized values to `other` and omitting empty or missing values.
## Testing
Add regression coverage for configured REPL and browser attribution, unrelated servers and connectors, unrecognized or absent configuration, and spoofed client metadata.
GitOrigin-RevId: bc24062526048ea1c1621099e609053cf167602f
## Why
Narrow composer footers can clip hints or crowd out mode indicators and actions.
## What changed
Use a shared layout for navigation, actions, mode indicators, context, and warnings. Shorten or drop whole labels in priority order instead of clipping them. Add a second row when input modes and queue or active side-conversation actions fit separately but not together, keeping navigation and modes above actions.
## Testing
Add coverage for fullscreen and scrollback layouts, narrow Plan mode hints, active side-conversation actions, oversized goal indicators, and warning hit areas.
GitOrigin-RevId: 849a53d45ef1b36b40c52282e329b48ec7d7d12e
## Why
Long built-in status lines were truncated to a single row even when the footer had room for more content.
## What changed
- Wrap status text to the available width and row budget, preserving styles and hyperlink targets across rows.
- Show an unlinked ellipsis when content exceeds the available space.
- Refresh wrapped content when the status text, agent label, or width changes.
- Keep reasoning-effort transitions on one row and restore wrapping after the final animation frame.
## Testing
Add coverage for wrapping and hyperlink boundaries across footer layouts, content and agent-label updates, resizing, and wrapping restoration after an effort transition. Update transcript and startup snapshots for multiline status text.
GitOrigin-RevId: 30e0795c473efd62fc4d0dca4deb6029b625a9e8
## Why
Status content and footer hints need to fit within the composer's height budget while preserving room for editing and active search controls.
## What changed
- Share row allocation between composer measurement and rendering. Cap status and hints at two rows in inline scrollback, and scale the fullscreen cap from two to four rows with the available budget.
- Reserve space for active footer controls, give suggestions and image previews priority over additional status rows, and avoid reserving an empty hints row when status is enabled.
- Keep textarea row counts and scroll offsets as `usize`, saturating only the terminal height so very tall drafts remain scrollable with a visible caret.
## Testing
Add layout and snapshot coverage for constrained height budgets, status and hints, image previews, slash suggestions, and transcript search cursors. Add a regression test for rendering a draft beyond the `u16` row limit.
GitOrigin-RevId: 20fb0384ffc976dfd12e648004763feff0597dda
## Why
Composer hints could replace configured status content, and mode indicators and warnings reduced the width available to status extensions.
## What changed
- Give enabled status content a dedicated area below the composer in both inline and fullscreen layouts, preserving it during transient hint states.
- Keep shortcuts, queue hints, mode indicators, and warnings in the hint row.
- Measure extension status content at the full available width, allowing up to four rows while yielding space to the composer when height is limited.
- Suppress composer hints behind centered dialogs.
## Testing
Update layout assertions and snapshots for status persistence, extension resizing and clipping, warning placement, and reasoning-effort transitions.
GitOrigin-RevId: f11c2e4fb53c35b69c2fbdbe80cef31bb28faca4
## Why
A live parent thread continues receiving lifecycle notifications while its side chat has focus, so returning to it does not need another liveness read.
## What changed
- Skip the parent liveness read when returning from a side chat while the parent is still attached and online. Keep the read when reattachment is needed.
- Stop scheduling a deferred screen clear when handling a transcript reset that preserves the screen.
## Testing
Add regression coverage for parent `thread/read` requests with and without reattachment, and assert that side-chat navigation leaves no deferred screen clear.
GitOrigin-RevId: a07f161ec86b0ae51250c5a0156402d328c42bdf
## Why
An accepted agent message may never appear in a later inference request or compaction snapshot. Rollout traces need to retain the delivery and its interaction edge even when the receiving agent makes no further request.
## What changed
Capture agent message payloads in `HistoryItemAccepted` events and create conversation items when those receipts are replayed. Reuse each item's ID when it later appears in a request or checkpoint, updating its body to the model-visible representation. Repeated receipts preserve that representation without duplicating the item.
## Testing
Extend the spawn-edge regression test to cover delivery before any child inference, stable item IDs across a later request, and preservation of trimmed content after a repeated receipt.
GitOrigin-RevId: c193d691e8240645800ce004c11f21e3f1c82915
## What changed
Expose history initialization and the effective history mode to thread-ready contributors. Add `EvidenceCursor::at_thread_ready` with paginated-history validation:
- Start new or cleared threads at position zero without requiring an existing rollout.
- Start resumed, forked, or supplied-history threads at the durable tail to avoid backfilling existing evidence.
- Skip the initial `SessionMeta` header when capturing evidence from position zero.
## Testing
Extend evidence tests to cover initialization before a rollout exists, capture of the first original record, rejection of resume without a rollout, and exclusion of existing records on resume.
GitOrigin-RevId: 32aa8e2afb71eb4e882b5f1fb107ac8616b33c8e
## Why
Submodule worktree scans can execute clean filters selected by task-writable Git configuration.
## What changed
Pass `--ignore-submodules=dirty` to `git status` in `get_has_changes_in_repo` to skip uncommitted submodule contents while retaining gitlink changes.
## Testing
Add a regression test that configures a submodule clean filter and verifies the repository change check does not execute it. The test also confirms that `git status` without the option executes the filter.
GitOrigin-RevId: 5330402f0354c3bf9613358575f8bedc9f312886
## Why
Code Mode needs a way to forward mixed MCP tool results while preserving their content types and order.
## What changed
Add `content(block)` and `content(iterable)` so callers can forward tool output with `content(result.content)`. Text, image, and audio blocks become their corresponding output items; resource links and embedded resources are serialized as text. Invalid blocks and encrypted text are rejected, and iterators are closed when a block is rejected.
Share MCP block conversion and image detail types with the protocol layer, while preserving existing `image()` and `audio()` input handling.
## Testing
Add coverage for mixed content ordering, iterator behavior and cleanup, invalid blocks, image metadata, short audio clips, and output suppression after exit. Add an integration scenario verifying that MCP text and image blocks reach the model in order.
GitOrigin-RevId: 83832ec6c530dd35ea1e3373680627997c6e13c1
## Why
Remote commands can use workspace paths from a different operating system. Preparing them on the client should preserve those paths and command arguments for the execution host.
## What changed
- Build exec-server requests directly, carrying the native command and sandbox context without applying the client's sandbox transform.
- Resolve workspace permissions and runtime read access in `SandboxManager` on the execution host, including helper binaries and validated managed-network CA bundles.
- Add explicit backend selection and required platform sandbox selection, rejecting launches when the required sandbox is unavailable.
## Testing
Extend request tests to cover foreign workspace paths, Unicode arguments, and empty arguments. Extend the remote process test on Unix to verify that restricted reads allow the sandbox helper and validated CA bundle while denying an untrusted certificate file.
GitOrigin-RevId: 5fc02ac4208912ecefb80a212742389afc14ac05
## Why
Gateway sign-in needs a way to complete authentication over SSH or when no local browser is available.
## What changed
- Prefer device authentication for gateways that support it in SSH sessions, headless Linux environments, or when no browser handler is available.
- Check the selected session's provider and complete required gateway sign-in before requesting models, including when resuming a thread.
- Display the verification URL, one-time code, and cancellation instructions. Handle sign-in results, cancellation, and connection failures while the prompt is open.
- Skip device login for older servers that do not support it while preserving errors from current servers.
## Testing
Add protocol tests for early sign-in completion, connection loss, provider selection, resumed threads, and legacy-server compatibility, plus snapshots of the device prompt and gateway sign-in loading text.
GitOrigin-RevId: 4f3675ab2fc7fccafbad17796c72a7736ebb86b5
## What changed
Add experimental `environments` fields to `turn/settings/update` and `thread/settings/update`. Turn updates apply to later steps of the running turn; thread updates apply to future turns. Omission or `null` preserves the selection, while `[]` clears it. Providing `cwd` in a thread update without an environment list selects the defaults for that directory.
Include the ordered environment selection in `thread/settings/updated`. Wait for Core to accept thread settings updates before returning success, and return rejections as request errors.
## Testing
Add integration coverage for attaching, replacing, preserving, and clearing environments; rejected updates; and the distinction between later steps and future turns.
GitOrigin-RevId: bfebeda57c676be94427aeb0c567c0b94fd9c387
## Why
Cloud conversations previously used the model returned on activation without exposing model discovery or supporting model changes from the TUI.
## What changed
- Fetch the account-scoped Cloud catalog on demand through `/v2/models`, independently of the submission and attachment queue.
- Let users stage a supported model and reasoning effort for the next new turn in the current conversation. Send these settings with the user message and show acceptance after successful admission.
- Show a loading picker when the catalog is empty, preserve compatibility with Cloud servers that lack discovery, and discard stale catalog results and selections when the account changes.
## Testing
Add coverage for catalog request parameters and queue bypass, deferred discovery and compatibility, model and effort serialization, admission outcomes, account invalidation, and opening the picker after discovery.
GitOrigin-RevId: 2cd0e7fe5f760d4da9258eefe9f0dcbcd23f9b07
## Why
When a workspace has an unreadable ancestor, PowerShell can start with `$PWD` unset and its filesystem drive at the root, breaking relative paths.
## What changed
Prefix PowerShell commands in the MXC sandbox with a guarded workaround that restores the location from the process working directory when `$PWD` is unset. Preserve locations already initialized by startup or a profile.
## Testing
Extend the MXC PowerShell integration test to cover an unreadable ancestor, checking that `$PWD` matches the process working directory, relative file access works, and the parent and sibling file remain inaccessible.
GitOrigin-RevId: 4fb067db52a7f3c6e819169016fe6bd104a277a1
## Why
User-configured model catalogs can supply Guardian prompt templates that omit policy slots or add conflicting instructions. Managed rules need a trusted template to reach the reviewer.
## What changed
- Use the bundled Guardian policy template when managed policy or additional rules are present, while preserving explicit host-provided templates.
- Remove `auto_review.extra_policy` and `auto_review.experimental_policy_template` from configuration and the schema. Warn when they are used, directing additional rules to `guardian_extra_policy` in managed `requirements.toml`.
## Testing
Add regression coverage for managed rules with user-supplied catalog templates, explicit host templates, and migration warnings for the removed settings.
GitOrigin-RevId: 7f00bfaa85051272ed70ceb2fcb64ffc297b8755
Pass `can_save: true` when opening the consent dialog and assert that the
popup displays “Include logs and session history?”.
GitOrigin-RevId: d6035a9a1b3e5114a6e92a6fd31547f02c4980fc
## Why
Waiting for host acceptance of a service-tier update in the submission loop can block interrupts while the host receipt is pending.
## What changed
Move the host acceptance wait out of the submission loop while keeping explicit thread settings replies pending until acceptance completes. Report handoff failures to the caller while preserving the locally committed settings, and allow a subsequent explicit request for the same tier to retry a failed handoff.
## Testing
Add a WebSocket regression test covering interrupts with pending host receipts, failure reporting, same-tier retries, and continued thread operation after startup caller cancellation.
GitOrigin-RevId: 1b27d156eba3ebb47e717a987f41f4f595635be8
## Why
Holding the session MCP gate during `tools/list` blocks other runtime access. Releasing it requires preventing stale refresh results from publishing after configuration or authentication changes.
## What changed
- Capture the Apps client under the session gate, then release the gate before fetching tools.
- Validate runtime and authentication state before publishing the shared cache and client catalog; retry a superseded refresh once.
- Wait for in-flight Apps refreshes before capturing model tools, preparing Apps calls, or shutting down the runtime.
## Testing
Add regression coverage for stale refreshes after runtime invalidation or logout, publication validation after call preparation drains, and a model step that waits for refreshed Apps tools while MCP status queries remain available.
GitOrigin-RevId: 42d4be8c4a9218933f8e3d07fc7f32f2da9cbd9d
## Why
MCP tools marked with `readOnlyHint: true` could still require approval in automatic mode when `destructiveHint` was also true.
## What changed
Give `readOnlyHint` precedence over `destructiveHint` and `openWorldHint` when deciding whether annotations require approval. Explicit prompt mode and strict automatic review continue to require their respective approval checks.
## Testing
Update unit tests for conflicting annotations, prompt mode, and strict automatic review. Extend app integration coverage to verify that read-only calls proceed while writes prompt in both automatic and writes approval modes.
GitOrigin-RevId: 11a930b857886bf41d09fd54c3a38607ae3d1b4e
## What changed
- Replace footer chord hints with a floating guide that shows available completions after a 350 ms delay, including custom prefixes.
- Share guide rendering across conversations, transcript overlays, the session picker, startup drafts, hook review, and daemon recovery. Preserve workspace-specific pane labels and filter startup drafts to supported actions.
- Adapt the guide to short inline viewports, retain `Esc` cancellation, and redraw when pending chords clear. Consume clicks on the guide so they do not activate controls beneath it.
## Testing
Extend tests and snapshots for delayed reveal, timeout, custom prefixes, compact and inline layouts, startup-safe completions, and picker redraws after chord completion or cancellation.
GitOrigin-RevId: afa2fe3ae1b5fbea4cd871c17d35206e749c842c
## What changed
- Add cloud archive and unarchive requests and support listing archived threads.
- Detach successfully archived threads and emit archive notifications. Report loaded cloud threads from live attachments rather than cached resume metadata, including after archive failures.
- Scope resume picker archive state and transcript previews by backend and thread so archiving a cloud session preserves a local session with the same ID.
- Limit cloud task archive cleanup to the selected thread, preserving child agents and their active session state, and update the confirmation text accordingly.
## Testing
Add regression coverage for archive acknowledgments, archived-thread discovery and restoration, attachment state after archive failures, same-ID sessions across backends, and preservation of attached child agents.
GitOrigin-RevId: bd155be85ee911375d4d326b18f9b5abe8e82b3d
## Why
CRT argument escaping adds literal backslashes before quotes in `cmd.exe /c` scripts, breaking quoted paths and executable names.
## What changed
Serialize single-script `cmd /c` invocations with `/s /c` and an outer quote pair, preserving the script's quotes. Share this handling across Windows sandbox process creation, ConPTY, and MXC execution requests.
## Testing
Add a regression test for `certutil -hashfile` with a quoted path containing spaces, and update the command-line test for a quoted PowerShell executable.
GitOrigin-RevId: 32b4b7b55f6af4dba206813eac150310f37ae1d4
## Why
Locally configured MCP servers may differ from those reported by the app server. Waiting for unreported servers can leave startup progress busy after the observed servers have finished.
## What changed
- Track startup progress from received server status updates, settling when no observed starts remain pending. Handle standalone outcomes after attachment or lag, deduplicate repeated diagnostics, and reactivate progress on new starts.
- Read effective MCP configuration through `config/read` to prevent task tools from shadowing the `codex_tui` namespace, including when configuration changes after startup.
- Use the same configuration reader to disable MCP servers for temporary structured threads.
- Remove client-side MCP requirement checks for the task-tools server and MCP requirements from debug configuration output, leaving enforcement to the app server.
## Testing
Update regression coverage for unreported local servers, startup rounds after lag, diagnostic handling, namespace conflicts, and server-enforced managed requirements.
GitOrigin-RevId: 370767b90ee55f8b0b1403872ef437d53257586b
## Why
`thread/sample` previously required the source's latest turn to finish before it could be sampled.
## What changed
Add `allowInProgressTurn`, defaulting to `false`, to sample a snapshot of the latest turn while the source continues running. The requested turn must still be the latest, but source history changes during sampling no longer invalidate the result when this option is enabled.
For an in-progress snapshot, include a context note explaining that the source was not stopped and pending tool calls may appear as `aborted` in the copy. Suppress the copy's default interruption message.
## Testing
Add coverage for sampling running turns in legacy and paginated history modes, including pending tool calls and mid-turn compaction. Verify that the source continues independently and that finished-turn samples retain their existing context.
GitOrigin-RevId: 771b979bf677e4ff0686476bfaf21aff92f7e7ac
## What changed
Allow GCP Responses requests to use `incremental_tools_legacy`. When enabled, tool declarations are sent through `additional_tools` inputs instead of the top-level `tools` array.
## Testing
Extend Vertex request coverage to check incremental tool declarations, including namespaces, tool search, and `exec_command`. Add GCP cases to the scenario that verifies tool catalog updates are appended without rewriting history, with incremental tools enabled and disabled.
GitOrigin-RevId: 8f4c02f8d62575e24e695d6fae0472a09625409f
## What changed
- Show a refusal status while preparing alternative suggestions and remove the pending notice above the composer.
- Clarify that choosing a rewrite starts a new chat with earlier context, and update overview suggestions to explicitly say users can ask Codex.
- Add an option to report safety-check concerns through the consent flow when reporting is enabled.
## Testing
Update safety recourse snapshots and extend coverage for reporting consent, returning to the picker, and Daybreak notices when enabled or unavailable for the selected model.
GitOrigin-RevId: 99f565ebec7064ec55d388e73acede2dfe639474
## Why
Opening a saved side chat waits for thread resume and history loading. Keep the chooser usable during that work and switch conversations only if the user still wants to open the chat.
## What changed
- Prepare saved chats in the background and show opening, waiting, and retry states in the sidebar. Allow cancellation and keep only the latest queued destination, with at most two preparations in flight.
- Ignore stale completions after navigation, archiving, disconnects, or backend changes, and release unclaimed subscriptions on their original connection.
- Preserve live events, settings, read receipts, and pending approvals received during loading. Anchor paginated history to the resumed snapshot to avoid overlap with live events.
- Carry held-key suppression across conversation switches.
## Testing
Add regression tests for retry and supersession, cancellation, stale completions, subscription cleanup, live updates during attachment, and held keys, plus snapshots for waiting and retry rows.
GitOrigin-RevId: 2c7d94482078f16895049ce79b242d3eb0d92f40
## What changed
Expose `discover_authorization_metadata` using RMCP and the existing OAuth HTTP adapter with a caller-supplied HTTP client. Return published authorization-server metadata, including discovery through protected-resource metadata, while excluding synthesized legacy endpoints. Return `None` when authorization metadata is unavailable and propagate discovery errors.
## Testing
Add tests for published versus synthesized endpoints, discovery failures and issuer mismatches, and discovery through a separate authorization server without sending credentials.
GitOrigin-RevId: b273067bb3b33d11301ae0e2017b28cd8e25b548
## Why
Title and content searches exclude attached threads, preventing clients from finding their saved conversations alongside ordinary threads.
## What changed
- Add `includeAttachedThreads` to `thread/list` and `thread/search`. For `thread/list`, require a non-empty `searchTerm` and preserve the usual provider and source filters.
- Expose `attachedToThreadId` on thread responses to identify the owning parent independently of fork and subagent lineage.
- Support inclusive searches through SQLite and filesystem fallback, repairing missing database entries with saved thread names during title searches.
## Testing
Add regression coverage for opt-in title and content searches, attached-parent metadata after restart, default search filters, filesystem searches without SQLite, and discovery and repair of unindexed title matches.
GitOrigin-RevId: ac12ec3c730ad7f1722bdf8e21600d17f12017f0
## Why
The existing loopback listener mode does not provide the stdin/stdout transport needed by OpenSSH's `ProxyCommand`.
## What changed
- Relay one SSH session over HTTP/3 CONNECT with `--environment-id ENV --port PORT`, defaulting to the public Portals gateway and port 22. Also accept an explicit `--proxy-url` with `--target ENV:PORT`.
- Require ChatGPT OAuth login, enforce cloud configuration and application network policy, and restrict OAuth credentials to the built-in gateway origin. Leave SSH authentication and host-key verification to OpenSSH.
- Preserve binary stdio, propagate input half-close, exit on gateway EOF, and report rejected CONNECT requests by HTTP status without exposing arbitrary peer error text.
- Require ERS-issued v2 tunnel JWTs for Portals listener mode unless OAuth mode is selected, and reject older ERS tunnel JWTs during credential updates.
- Disable UDP segmentation offload for SSH tunnels on Linux to avoid stalled bulk transfers on affected paths. Document an OpenSSH configuration example.
## Testing
Add coverage for argument validation, gateway routing, credential boundaries, JWT startup and renewal checks, binary relay and half-close behavior, gateway EOF and rejection, and cloud policy enforcement before gateway connection.
GitOrigin-RevId: 0d9afbddef01748efcc1c4b88f7e19923deff7e1
## Why
The Windows `notify` backend wakes every 100 ms even when idle.
## What changed
- Patch `notify` 8.2.0 in Bazel to wait indefinitely for events or control requests.
- Wake the watcher when configuration requests are queued and share ownership of the wakeup semaphore with `Arc<OwnedHandle>`.
## Testing
Add a Windows regression test that verifies file changes are reported after unregistering one directory and registering another on the same watcher.
GitOrigin-RevId: b53b3c8bc514008554fdeffaef3f7bf8305a4803
## Why
Reference-backed forks can lack a local opening user message, causing them to disappear from thread listings or lose their inherited preview when a new turn is written. Compaction can also remove the opening message from the model context used to prepare a fork.
## What changed
- Persist inherited previews and preserve them when recording new turns and merging thread-list metadata.
- Include reference-backed forks with inherited history in thread listings before their first local message.
- Recover opening previews from transcript history within the requested fork boundary when compaction omits them from model context.
- Use the shared `user_message_preview` helper when deriving previews from rollout items.
## Testing
Extend thread-fork integration tests to check previews in listings before and after a new turn, after reopening, and when forking at a boundary after compaction.
GitOrigin-RevId: 5962cf71ea7104f07c9013cc77259b105e4c60d7
## Why
A failed side chat's opening prompt and image attachments could be restored into the parent chat's draft when dismissing the failure.
## What changed
Clear the pending opening message when sidebar chat creation fails, and stop restoring the opening message when side conversation creation returns an error.
## Testing
Update the sidebar chooser test to verify that dismissing a failed creation preserves the parent draft and a subsequent prompt without restoring the failed prompt or its image attachment.
GitOrigin-RevId: 1423786b2209e9441e0bc1504ffcf51e74ba7cde
## What changed
- Fold earlier commentary, reasoning, and tool activity beneath an expandable prompt control in the full-screen transcript, keeping partial and final answers visible.
- Label the control with completion timing when available, or a previous-message count otherwise. Keep tool groups independently expandable within the turn.
- Preserve assistant message phases during live completion and replay, and retain untimed completion markers to identify completed turns.
## Testing
Add transcript coverage for nested expansion, keyboard and mouse controls, narrow viewports, search anchor preservation, detailed mode, and legacy messages without phase or timing metadata.
GitOrigin-RevId: b17d9d7c870990178d76d12a13d463ff9c32b96c
## What changed
- Combine adjacent tool calls into one expandable summary, including calls that move from live output into history. Label successful command groups with summaries such as `Explored files` and `Ran 2 commands`.
- Include web searches in tool groups, hide duplicate provisional search entries, and show a single completed search directly. Preserve its reading position when it completes.
- Refresh summaries as history changes while preserving pinned content during selection and reading.
## Testing
Add regression coverage for web search completion, grouped search expansion, loading older history during selection, and command completion with an expanded summary.
GitOrigin-RevId: 98f4e0c49cbf8014c25874cf47d1c0c92ee6b30d