## What
Record a model-visible `<turn_aborted>` marker in history when a turn is
interrupted, and treat it as a session prefix.
## Why
When a turn is interrupted, Codex emits `TurnAborted` but previously did
not persist anything model-visible in the conversation history. On the
next user turn, the model can’t tell the previous work was aborted and
may resume/repeat earlier actions (including duplicated side effects
like re-opening PRs).
Fixes: https://github.com/openai/codex/issues/9042
## How
On `TurnAbortReason::Interrupted`, append a hidden user message
containing a `<turn_aborted>…</turn_aborted>` marker and flush.
Treat `<turn_aborted>` like `<environment_context>` for session-prefix
filtering.
Add a regression test to ensure follow-up turns don’t repeat side
effects from an aborted turn.
## Testing
`just fmt`
`just fix -p codex-core`
`cargo test -p codex-core -- --test-threads=1`
`cargo test --all-features -- --test-threads=1`
---------
Co-authored-by: Skylar Graika <sgraika127@gmail.com>
Co-authored-by: jif-oai <jif@openai.com>
Co-authored-by: Eric Traut <etraut@openai.com>
This PR fixes a small issue with chained (layered) config.toml file
merging. The old logic didn't properly handle profiles.
In particular, if a lower-layer config overrides a profile defined in a
higher-layer config, the override did not take effect. This prevents
users from having project-specific profile overrides and contradicts the
(soon-to-be) documented behavior of config merging.
The change adds a unit test for this case. It also exposes a function
from the config crate that is needed by the app server code paths to
implement support for layered configs.
## Summary
We have a variety of things we refer to as instructions in the code
base: our current canonical terms are:
- base instructions (raw string)
- developer instructions (has a type in protocol)
- user instructions
We also have `instructions` floating around in various places. We should
standardize on the above, and start using types to prevent them from
ending up in the wrong place. There will be additional PRs, but I'm
going to keep these small so we can easily follow them!
## Testing
- [x] Tests pass, this is purely a file move
Add support for returning threads by either `created_at` OR `updated_at`
descending. Previously core always returned threads ordered by
`created_at`.
This PR:
- updates core to be able to list threads by `updated_at` OR
`created_at` descending based on what the caller wants
- also update `thread/list` in app-server to expose this (default to
`created_at` if not specified)
All existing codepaths (app-server, TUI) still default to `created_at`,
so no behavior change is expected with this PR.
**Implementation**
To sort by `updated_at` is a bit nontrivial (whereas `created_at` is
easy due to the way we structure the folders and filenames on disk,
which are all based on `created_at`).
The most naive way to do this without introducing a cache file or sqlite
DB (which we have to implement/maintain) is to scan files in reverse
`created_at` order on disk, and look at the file's mtime (last modified
timestamp according to the filesystem) until we reach `MAX_SCAN_FILES`
(currently set to 10,000). Then, we can return the most recent N
threads.
Based on some quick and dirty benchmarking on my machine with ~1000
rollout files, calling `thread/list` with limit 50, the `updated_at`
path is slower as expected due to all the I/O:
- updated-at: average 103.10 ms
- created-at: average 41.10 ms
Those absolute numbers aren't a big deal IMO, but we can certainly
optimize this in a followup if needed by introducing more state stored
on disk.
**Caveat**
There's also a limitation in that any files older than `MAX_SCAN_FILES`
will be excluded, which means if a user continues a REALLY old thread,
it's possible to not be included. In practice that should not be too big
of an issue.
If a user makes...
- 1000 rollouts/day → threads older than 10 days won't show up
- 100 rollouts/day → ~100 days
If this becomes a problem for some reason, even more motivation to
implement an updated_at cache.
moving `web_search` rollout serverside, so need a way to explicitly
disable search + signal eligibility from the client.
- Add `x‑oai‑web‑search‑eligible` header that signifies whether the
request can have web search.
- Only attach the `web_search` tool when the resolved `WebSearchMode` is
`Live` or `Cached`.
When an invalid config.toml key or value is detected, the CLI currently
just quits. This leaves the VSCE in a dead state.
This PR changes the behavior to not quit and bubble up the config error
to users to make it actionable. It also surfaces errors related to
"rules" parsing.
This allows us to surface these errors to users in the VSCE, like this:
<img width="342" height="129" alt="Screenshot 2026-01-13 at 4 29 22 PM"
src="https://github.com/user-attachments/assets/a79ffbe7-7604-400c-a304-c5165b6eebc4"
/>
<img width="346" height="244" alt="Screenshot 2026-01-13 at 4 45 06 PM"
src="https://github.com/user-attachments/assets/de874f7c-16a2-4a95-8c6d-15f10482e67b"
/>
This is an alternate PR to solving the same problem as
<https://github.com/openai/codex/pull/8227>.
In this PR, when Ollama is used via `--oss` (or via `model_provider =
"ollama"`), we default it to use the Responses format. At runtime, we do
an Ollama version check, and if the version is older than when Responses
support was added to Ollama, we print out a warning.
Because there's no way of configuring the wire api for a built-in
provider, we temporarily add a new `oss_provider`/`model_provider`
called `"ollama-chat"` that will force the chat format.
Once the `"chat"` format is fully removed (see
<https://github.com/openai/codex/discussions/7782>), `ollama-chat` can
be removed as well
---------
Co-authored-by: Eric Traut <etraut@openai.com>
Co-authored-by: Michael Bolin <mbolin@openai.com>
Elevated Sandbox NUX:
* prompt for elevated sandbox setup when agent mode is selected (via
/approvals or at startup)
* prompt for degraded sandbox if elevated setup is declined or fails
* introduce /elevate-sandbox command to upgrade from degraded
experience.
Added an agent control plane that lets sessions spawn or message other
conversations via `AgentControl`.
`AgentBus` (core/src/agent/bus.rs) keeps track of the last known status
of a conversation.
ConversationManager now holds shared state behind an Arc so AgentControl
keeps only a weak back-reference, the goal is just to avoid explicit
cycle reference.
Follow-ups:
* Build a small tool in the TUI to be able to see every agent and send
manual message to each of them
* Handle approval requests in this TUI
* Add tools to spawn/communicate between agents (see related design)
* Define agent types
# External (non-OpenAI) Pull Request Requirements
Before opening this Pull Request, please read the dedicated
"Contributing" markdown file or your PR may be closed:
https://github.com/openai/codex/blob/main/docs/contributing.md
If your PR conforms to our contribution guidelines, replace this text
with a detailed and high quality description of your changes.
Include a link to a bug report or enhancement request.
Historically, `accept_elicitation_for_prompt_rule()` was flaky because
we were using a notification to update the sandbox followed by a `shell`
tool request that we expected to be subject to the new sandbox config,
but because [rmcp](https://crates.io/crates/rmcp) MCP servers delegate
each incoming message to a new Tokio task, messages are not guaranteed
to be processed in order, so sometimes the `shell` tool call would run
before the notification was processed.
Prior to this PR, we relied on a generous `sleep()` between the
notification and the request to reduce the change of the test flaking
out.
This PR implements a proper fix, which is to use a _request_ instead of
a notification for the sandbox update so that we can wait for the
response to the sandbox request before sending the request to the
`shell` tool call. Previously, `rmcp` did not support custom requests,
but I fixed that in
https://github.com/modelcontextprotocol/rust-sdk/pull/590, which made it
into the `0.12.0` release (see #8288).
This PR updates `shell-tool-mcp` to expect
`"codex/sandbox-state/update"` as a _request_ instead of a notification
and sends the appropriate ack. Note this behavior is tied to our custom
`codex/sandbox-state` capability, which Codex honors as an MCP client,
which is why `core/src/mcp_connection_manager.rs` had to be updated as
part of this PR, as well.
This PR also updates the docs at `shell-tool-mcp/README.md`.
This is a step towards removing the need to know `model` when
constructing config. We firstly don't need to know `model_info` and just
respect if the user has already set it. Next step, we don't need to know
`model` unless the user explicitly set it in `config.toml`
- Inline response recording during streaming: `run_turn` now records
items as they arrive instead of building a `ProcessedResponseItem` list
and post‑processing via `process_items`.
- Simplify turn handling: `handle_output_item_done` returns the
follow‑up signal + optional tool future; `needs_follow_up` is set only
there, and in‑flight tool futures are drained once at the end (errors
logged, no extra state writes).
- Flattened stream loop: removed `process_items` indirection and the
extra output queue
- - Tests: relaxed `tool_parallelism::tool_results_grouped` to allow any
completion order while still requiring matching call/output IDs.
This PR moves `ModelsFamily` to `openai_models`. It also propagates
`ModelsManager` to session services and use it to drive model family. We
also make `derive_default_model_family` private because it's a step
towards what we want: one place that gives model configuration.
This is a second step at having one source of truth for models
information and config: `ModelsManager`.
Next steps would be to remove `ModelsFamily` from config. That's massive
because it's being used in 41 occasions mostly pre launching `codex`.
Also, we need to make `find_family_for_model` private. It's also big
because it's being used in 21 occasions ~ all tests.
- Introduce `openai_models` in `/core`
- Move `PRESETS` under it
- Move `ModelPreset`, `ModelUpgrade`, `ReasoningEffortPreset`,
`ReasoningEffortPreset`, and `ReasoningEffortPreset` to `protocol`
- Introduce `Op::ListModels` and `EventMsg::AvailableModels`
Next steps:
- migrate `app-server` and `tui` to use the introduced Operation
This change prototypes support for Skills with the CLI. This is an
**experimental** feature for internal testing.
---------
Co-authored-by: Gav Verma <gverma@openai.com>
This introduces a new feature to Codex when it operates as an MCP
_client_ where if an MCP _server_ replies that it has an entry named
`"codex/sandbox-state"` in its _server capabilities_, then Codex will
send it an MCP notification with the following structure:
```json
{
"method": "codex/sandbox-state/update",
"params": {
"sandboxPolicy": {
"type": "workspace-write",
"network-access": false,
"exclude-tmpdir-env-var": false
"exclude-slash-tmp": false
},
"codexLinuxSandboxExe": null,
"sandboxCwd": "/Users/mbolin/code/codex2"
}
}
```
or with whatever values are appropriate for the initial `sandboxPolicy`.
**NOTE:** Codex _should_ continue to send the MCP server notifications
of the same format if these things change over the lifetime of the
thread, but that isn't wired up yet.
The result is that `shell-tool-mcp` can consume these values so that
when it calls `codex_core::exec::process_exec_tool_call()` in
`codex-rs/exec-server/src/posix/escalate_server.rs`, it is now sure to
call it with the correct values (whereas previously we relied on
hardcoded values).
While I would argue this is a supported use case within the MCP
protocol, the `rmcp` crate that we are using today does not support
custom notifications. As such, I had to patch it and I submitted it for
review, so hopefully it will be accepted in some form:
https://github.com/modelcontextprotocol/rust-sdk/pull/556
To test out this change from end-to-end:
- I ran `cargo build` in `~/code/codex2/codex-rs/exec-server`
- I built the fork of Bash in `~/code/bash/bash`
- I added the following to my `~/.codex/config.toml`:
```toml
# Use with `codex --disable shell_tool`.
[mcp_servers.execshell]
args = ["--bash", "/Users/mbolin/code/bash/bash"]
command = "/Users/mbolin/code/codex2/codex-rs/target/debug/codex-exec-mcp-server"
```
- From `~/code/codex2/codex-rs`, I ran `just codex --disable shell_tool`
- When the TUI started up, I verified that the sandbox mode is
`workspace-write`
- I ran `/mcp` to verify that the shell tool from the MCP is there:
<img width="1387" height="1400" alt="image"
src="https://github.com/user-attachments/assets/1a8addcc-5005-4e16-b59f-95cfd06fd4ab"
/>
- Then I asked it:
> what is the output of `gh issue list`
because this should be auto-approved with our existing dummy policy:
af63e6eccc/codex-rs/exec-server/src/posix.rs (L157-L164)
And it worked:
<img width="1387" height="1400" alt="image"
src="https://github.com/user-attachments/assets/7568d2f7-80da-4d68-86d0-c265a6f5e6c1"
/>
## 🐛 Problem
Users running commands with non-ASCII characters (like Russian text
"пример") in Windows/WSL environments experience garbled text in
VSCode's shell preview window, with Unicode replacement characters (�)
appearing instead of the actual text.
**Issue**: https://github.com/openai/codex/issues/6178
## 🔧 Root Cause
The issue was in `StreamOutput<Vec<u8>>::from_utf8_lossy()` method in
`codex-rs/core/src/exec.rs`, which used `String::from_utf8_lossy()` to
convert shell output bytes to strings. This function immediately
replaces any invalid UTF-8 byte sequences with replacement characters,
without attempting to decode using other common encodings.
In Windows/WSL environments, shell output often uses encodings like:
- Windows-1252 (common Windows encoding)
- Latin-1/ISO-8859-1 (extended ASCII)
## 🛠️ Solution
Replaced the simple `String::from_utf8_lossy()` call with intelligent
encoding detection via a new `bytes_to_string_smart()` function that
tries multiple encoding strategies:
1. **UTF-8** (fast path for valid UTF-8)
2. **Windows-1252** (handles Windows-specific characters in 0x80-0x9F
range)
3. **Latin-1** (fallback for extended ASCII)
4. **Lossy UTF-8** (final fallback, same as before)
## 📁 Changes
### New Files
- `codex-rs/core/src/text_encoding.rs` - Smart encoding detection module
- `codex-rs/core/tests/suite/text_encoding_fix.rs` - Integration tests
### Modified Files
- `codex-rs/core/src/lib.rs` - Added text_encoding module
- `codex-rs/core/src/exec.rs` - Updated StreamOutput::from_utf8_lossy()
- `codex-rs/core/tests/suite/mod.rs` - Registered new test module
## ✅ Testing
- **5 unit tests** covering UTF-8, Windows-1252, Latin-1, and fallback
scenarios
- **2 integration tests** simulating the exact Issue #6178 scenario
- **Demonstrates improvement** over the previous
`String::from_utf8_lossy()` approach
All tests pass:
```bash
cargo test -p codex-core text_encoding
cargo test -p codex-core test_shell_output_encoding_issue_6178
```
## 🎯 Impact
- ✅ **Eliminates garbled text** in VSCode shell preview for non-ASCII
content
- ✅ **Supports Windows/WSL environments** with proper encoding detection
- ✅ **Zero performance impact** for UTF-8 text (fast path)
- ✅ **Backward compatible** - UTF-8 content works exactly as before
- ✅ **Handles edge cases** with robust fallback mechanism
## 🧪 Test Scenarios
The fix has been tested with:
- Russian text ("пример")
- Windows-1252 quotation marks (""test")
- Latin-1 accented characters ("café")
- Mixed encoding content
- Invalid byte sequences (graceful fallback)
## 📋 Checklist
- [X] Addresses the reported issue
- [X] Includes comprehensive tests
- [X] Maintains backward compatibility
- [X] Follows project coding conventions
- [X] No breaking changes
---------
Co-authored-by: Josh McKinney <joshka@openai.com>
This PR threads execpolicy2 into codex-core.
activated via feature flag: exec_policy (on by default)
reads and parses all .codexpolicy files in `codex_home/codex`
refactored tool runtime API to integrate execpolicy logic
---------
Co-authored-by: Michael Bolin <mbolin@openai.com>
## Summary
Enables shell_command for windows users, and starts adding some basic
command parsing here, to at least remove powershell prefixes. We'll
follow this up with command parsing but I wanted to land this change
separately with some basic UX.
**NOTE**: This implementation parses bash and powershell on both
platforms. In theory this is possible, since you can use git bash on
windows or powershell on linux. In practice, this may not be worth the
complexity of supporting, so I don't feel strongly about the current
approach vs. platform-specific branching.
## Testing
- [x] Added a bunch of tests
- [x] Ran on both windows and os x
## Overview
Adds LM Studio OSS support. Closes#1883
### Changes
This PR enhances the behavior of `--oss` flag to support LM Studio as a
provider. Additionally, it introduces a new flag`--local-provider` which
can take in `lmstudio` or `ollama` as values if the user wants to
explicitly choose which one to use.
If no provider is specified `codex --oss` will auto-select the provider
based on whichever is running.
#### Additional enhancements
The default can be set using `oss-provider` in config like:
```
oss_provider = "lmstudio"
```
For non-interactive users, they will need to either provide the provider
as an arg or have it in their `config.toml`
### Notes
For best performance, [set the default context
length](https://lmstudio.ai/docs/app/advanced/per-model) for gpt-oss to
the maximum your machine can support
---------
Co-authored-by: Matt Clayton <matt@lmstudio.ai>
Co-authored-by: Eric Traut <etraut@openai.com>
Implements:
```
thread/list
thread/start
thread/resume
thread/archive
```
along with their integration tests. These are relatively light wrappers
around the existing core logic, and changes to core logic are minimal.
However, an improvement made for developer ergonomics:
- `thread/start` and `thread/resume` automatically attaches a
conversation listener internally, so clients don't have to make a
separate `AddConversationListener` call like they do today.
For consistency, also updated `model/list` and `feedback/upload` (naming
conventions, list API params).
This is just a refactor of `conversation_history` file by breaking it up
into multiple smaller ones with helper. This refactor will help us move
more functionality related to context management here. in a clean way.
In this PR, I am exploring migrating task kind to an invocation of
Codex. The main reason would be getting rid off multiple
`ConversationHistory` state and streamlining our context/history
management.
This approach depends on opening a channel between the sub-codex and
codex. This channel is responsible for forwarding `interactive`
(`approvals`) and `non-interactive` events. The `task` is responsible
for handling those events.
This opens the door for implementing `codex as a tool`, replacing
`compact` and `review`, and potentially subagents.
One consideration is this code is very similar to `app-server` specially
in the approval part. If in the future we wanted an interactive
`sub-codex` we should consider using `codex-mcp`
This adds an RPC to the app server to the the `ConversationSummary` via
a rollout path. Now that the VS Code extension supports showing the
Codex UI in an editor panel where the URI of the panel maps to the
rollout file, we need to be able to get the `ConversationSummary` from
the rollout file directly.
Currently we collect all all turn items in a vector, then we add it to
the history on success. This result in losing those items on errors
including aborting `ctrl+c`.
This PR:
- Adds the ability for the tool call to handle cancellation
- bubble the turn items up to where we are recording this info
Admittedly, this logic is an ad-hoc logic that doesn't handle a lot of
error edge cases. The right thing to do is recording to the history on
the spot as `items`/`tool calls output` come. However, this isn't
possible because of having different `task_kind` that has different
`conversation_histories`. The `try_run_turn` has no idea what thread are
we using. We cannot also pass an `arc` to the `conversation_histories`
because it's a private element of `state`.
That's said, `abort` is the most common case and we should cover it
until we remove `task kind`
1. Adds AgentMessage, Reasoning, WebSearch items.
2. Switches the ResponseItem parsing to use new items and then also emit
3. Removes user-item kind and filters out "special" (environment) user
items when returning to clients.
Add proper feature flag instead of having custom flags for everything.
This is just for experimental/wip part of the code
It can be used through CLI:
```bash
codex --enable unified_exec --disable view_image_tool
```
Or in the `config.toml`
```toml
# Global toggles applied to every profile unless overridden.
[features]
apply_patch_freeform = true
view_image_tool = false
```
Follow-up:
In a following PR, the goal is to have a default have `bundles` of
features that we can associate to a model
This adds a queryable auth status for MCP servers which is useful:
1. To determine whether a streamable HTTP server supports auth or not
based on whether or not it supports RFC 8414-3.2
2. Allow us to build a better user experience on top of MCP status
## Summary
- Factor `load_config_as_toml` into `core::config_loader` so config
loading is reusable across callers.
- Layer `~/.codex/config.toml`, optional `~/.codex/managed_config.toml`,
and macOS managed preferences (base64) with recursive table merging and
scoped threads per source.
## Config Flow
```
Managed prefs (macOS profile: com.openai.codex/config_toml_base64)
▲
│
~/.codex/managed_config.toml │ (optional file-based override)
▲
│
~/.codex/config.toml (user-defined settings)
```
- The loader searches under the resolved `CODEX_HOME` directory
(defaults to `~/.codex`).
- Managed configs let administrators ship fleet-wide overrides via
device profiles which is useful for enforcing certain settings like
sandbox or approval defaults.
- For nested hash tables: overlays merge recursively. Child tables are
merged key-by-key, while scalar or array values replace the prior layer
entirely. This lets admins add or tweak individual fields without
clobbering unrelated user settings.
# Tool System Refactor
- Centralizes tool definitions and execution in `core/src/tools/*`:
specs (`spec.rs`), handlers (`handlers/*`), router (`router.rs`),
registry/dispatch (`registry.rs`), and shared context (`context.rs`).
One registry now builds the model-visible tool list and binds handlers.
- Router converts model responses to tool calls; Registry dispatches
with consistent telemetry via `codex-rs/otel` and unified error
handling. Function, Local Shell, MCP, and experimental `unified_exec`
all flow through this path; legacy shell aliases still work.
- Rationale: reduce per‑tool boilerplate, keep spec/handler in sync, and
make adding tools predictable and testable.
Example: `read_file`
- Spec: `core/src/tools/spec.rs` (see `create_read_file_tool`,
registered by `build_specs`).
- Handler: `core/src/tools/handlers/read_file.rs` (absolute `file_path`,
1‑indexed `offset`, `limit`, `L#: ` prefixes, safe truncation).
- E2E test: `core/tests/suite/read_file.rs` validates the tool returns
the requested lines.
## Next steps:
- Decompose `handle_container_exec_with_params`
- Add parallel tool calls
