codex — Engineering Performance
33 engineers all time · Apr 2025 – Nov 2026 · built 2026-11-22 · GitHub
Performance snapshot
Today's rolling 90-day reading for codex, compared with the start of the series. Pick a window to move that comparison point.
Avg. perf / dev / mo
+432.9%
1.73 → 9.21 ETV
Active engineers
+242.9%
7.0 → 24.0
Features
−27.0pp
60.1% → 33.1%
vs. OpenAI
0.95x
1.6x → 0.95x · −5% below
codex vs. OpenAI
Per-engineer ETV for codex against OpenAI as a whole. Both lines are 90-day rolling averages scaled to a 30-day month, so they share one axis and can be read against each other at any point. Pick a window to zoom the chart to it.
Performance Composition
Each month's output split by type of work: Features (new value), Maintenance (sustaining systems), Tests, Docs, and Fixes (rework). The yellow line is output per engineer, so when it rises each engineer is delivering more, whatever the team size did. Unit: Engineering Throughput Value (ETV).
Engineering capacity
Effective engineers behind codex, against its pre-AI baseline. Each subject has its own: codex's is 1.73 ETV / dev / mo, its first reading in Q2 2025. Per-engineer ETV divided by that gives a capacity multiple, and that multiple applied to the engineers active in the trailing 90 days turns it into engineer-equivalents. The line is the real headcount, so the gap between line and area is what the leverage is worth. Because each baseline is its own, every subject opens at 1.0x on its first day: multiples measure improvement and are not comparable between subjects.
Knowledge concentration
How dependent is this repo on a small number of engineers? Higher top-1 share = higher key-person risk.
jif owns 23.8 % of commits.
Behind the numbers
Written summary of the work completed each month.
In September 2026, development activity surged across the repository with 1,617 total commits, delivering major enhancements in asynchronous classification for Guardian v2, exec-server streaming, and thread persistence. Total output increased across every category, with Features up 63% and Tests rising 135% compared to the 5-month average. The period represented a genuine step up in volume and delivery, exceeding both the 5-month average and the preceding two months.
Highlights
- Delivered conversation mode for Guardian v2 asynchronous classification via 726f1492 (felixxia-oai) and retained conversation sampling support via 5f300d3f (felixxia-oai).
- Added capability-gated writable file streams and chunked write protocol support to exec-server in e7798c99 (Adam), 875bf920 (Adam), and c51f5bfb (Adam).
- Promoted API-key model discovery to stable and enabled it by default in 7d4c7a07 (andrewgu-oai), while decoupling cyber access programs in f151a0f5 (Brian).
- Enhanced thread-store durability with authoritative replay history restoration in d7b0d4aa (felixxia-oai), snapshot reuse on resume in 92bc601a (felixxia-oai), and explicit goal persistence in de020167 (felixxia-oai).
- Introduced automated SQLite corruption detection and recovery mechanisms in 3620b2ca (Dmitriy) alongside typed error classification in 596ae94f (Dmitriy).
Observations
- Total commits reached 1,617 (a 64% increase over the 986-commit 5-month average), yielding a total output score of 772 versus the 394 5-month average (+96%).
- Features score rose to 254 compared to the 156 5-month average (+63%), driven by major feature additions in Guardian classification, exec-server protocols, and TUI capabilities.
- Fixes score reached 126 vs the 52 5-month average (+143%), involving targeted fixes for session lifecycle interruptions in ec0cfa5d (Benjamin), TUI paste handling in cca44069 (Benjamin), and thread naming persistence in 1f77c0cf (Raunak).
- Significant performance refactoring moved blocking I/O operations off async runtimes, including session index I/O in 4f699cd6 (Charlie), thread history projection in ecc205ed (Charlie), and rollout directory scanning in 5aa92804 (Charlie).
- Tests score surged to 292 (+135% above the 124 5-month average), reflecting comprehensive snapshot, regression, and unit test suites across TUI, exec-server, and state recovery modules.
Based on 1,617 commits463.2 ETVUpdated Nov 22, 2026, 5:01 PM
Top engineers
Most impactful commits
Top 10 by ETV in the all-time window.
- 8.0ETVAdd selection and copying to the transcript viewer (#46732) ## What changed - Add mouse selection, word and line selection, drag autoscrolling, and keyboard selection with `Ctrl+Space`. Support copying selected text and opening links with a modified click. - Anchor reading positions and selections to source text so resizing, streaming completion, and history pagination preserve the displayed content. Preserve hard line breaks, indentation, tabs, and graphemes when copying, while excluding display gutters and soft wraps. - Rejoin activity groups split across history pages while retaining the reader's displayed revision. - Distinguish confirmed clipboard writes from unconfirmed terminal requests, and retain native clipboard ownership after the transcript overlay closes. - Manage mouse capture and keyboard modes across overlay transitions, editor handoffs, and suspend/resume. ## Testing Add regression tests for selection during live updates and pagination, source-text copying across resize, clipboard confirmation and lifetime, and terminal mode restoration. GitOrigin-RevId: 6b2438f576567cbba755b2405915a25acf009b8dEric Traut · d7d9f2b9 · 2026-09-19
- 7.7ETVSimplify app-server integration test setup (#34786) ## What changed - Add typed helpers for initialized app-server instances, JSON-RPC requests, responses, notifications, and thread startup. - Add a composable `MockResponsesConfig` builder while preserving the existing mock config writers. - Migrate the app-server integration suite to the shared helpers, removing repeated request ID management, deserialization, initialization, and config assembly. ## Testing - Add unit coverage for composing mock provider, feature, and extra config, and for compatibility of the existing mock config writer. GitOrigin-RevId: 62da177c1d6e8299400de64bd6766567bf1ba3d8pakrym-oai · 10cc57c9 · 2026-07-22
- 7.3ETVchore: migrate from Config::load_from_base_config_with_overrides to ConfigBuilder (#8276) https://github.com/openai/codex/pull/8235 introduced `ConfigBuilder` and this PR updates all call non-test call sites to use it instead of `Config::load_from_base_config_with_overrides()`. This is important because `load_from_base_config_with_overrides()` uses an empty `ConfigRequirements`, which is a reasonable default for testing so the tests are not influenced by the settings on the host. This method is now guarded by `#[cfg(test)]` so it cannot be used by business logic. Because `ConfigBuilder::build()` is `async`, many of the test methods had to be migrated to be `async`, as well. On the bright side, this made it possible to eliminate a bunch of `block_on_future()` stuff.Michael Bolin · 3d4ced3f · 2025-12-19
- 5.3ETVKeep same-named MCP servers separate across environments (#60905) ## Why MCP servers discovered on different executors can share a name. Resolving them by name alone exposes only one server and prevents tools and resources from being addressed independently. ## What changed - Add the disabled-by-default `multi_environment_mcp` feature to expose environment-scoped servers, including multiple installations of the same plugin. - Carry exact `McpServerKey` identities through catalogs, connections, tool calls, resources, and approval tracking. Give environment-scoped tools stable namespaces derived from their environment. - Include `environment_id` in scoped resource listings and accept it in resource requests. Require it for environment-owned servers when multiple environments are selected. - Preserve server ownership through hooks and use the owning environment's permissions for scoped elicitation review. ## Testing Add coverage for same-named servers across environments, independent calls and approvals, resource selection, plugin discovery, and rejection of hook changes to the tool environment. GitOrigin-RevId: 5593372b06a15c16aa3220faa3b19f5d3b20e67esayan-oai · 561f0f5c · 2026-11-11
- 5.0ETVHarden unsandboxed patch filesystem access (#39659) ## Why An `apply_patch` path can be replaced with a symlink after verification, allowing an unsandboxed patch operation to reach a different file than the one that was approved. ## What changed - Add `follow_symlinks` options to executor filesystem reads, writes, metadata lookups, directory creation, and removal, including the corresponding `followSymlinks` protocol fields. - Implement no-follow filesystem operations on Unix and Windows that reject links in any path component and restrict file access to regular files. - Run `apply_patch` with symlink traversal disabled when an otherwise-required sandbox is bypassed, while retaining the existing follow-symlink default for standalone callers. ## Testing - Cover leaf and ancestor symlinks across patch add, update, delete, and move operations, including a path swap after verification. - Exercise local and remote no-follow filesystem behavior, concurrent directory creation, special-file rejection, and Windows reparse points. GitOrigin-RevId: 43fd479084891493ce13564fbd894b98f329c6ddpakrym-oai · e3e5ad28 · 2026-08-20
- 4.8ETVAdd transcript search and per-activity detail controls (#46734) ## What changed - Add incremental, case-insensitive literal search across transcript details, loading older history as needed. Use `F3` to start Find, or `/` in the transcript pager, and navigate matches in either direction. Closing Find restores the previous presentation and reading position. - Add clickable activity detail controls and `F4` keyboard focus for inspecting individual activity groups in the owned transcript. Preserve expansion state as live activity becomes history. - Expose configurable `global.find_transcript`, `global.focus_activity`, and `pager.find` bindings under `tui.keymap`. Let new defaults yield to existing custom bindings and show available completions for pending chords. ## Testing Add regression coverage for Unicode matching, bounded scanning, paginated and changing history, selection during search, composer input isolation, activity controls, and custom shortcut conflicts. GitOrigin-RevId: 5bcbcdfbca7c5c296c635ec9b079ccd7a908e336Eric Traut · df7f717c · 2026-09-19
- 4.6ETVfeat: spreadsheet artifact (#13345)jif-oai · 8c5e50ef · 2026-03-03
- 4.5ETVImprove TUI contrast, keyboard hints, and picker layouts (#46680) ## Why Theme colors and dimmed text can lose contrast against terminal and shaded backgrounds. Pickers also need to keep controls and the selected result visible when space is limited. ## What changed - Add shared blue accents and contrast-aware foregrounds that account for painted backgrounds and terminal color depth, with terminal-default fallbacks for limited palettes. - Apply readable colors to footer text, status lines, prompt attachment labels, and diff content. Remove dimming from syntax-highlighted diff text so it does not undo contrast correction. - Use compact shortcut labels such as `ctrl+c`, emphasize complete key combinations, and separate composer footer actions with ` · `. - Add an opt-in picker presentation with full-width selection, filled tabs, separate overflow indicators, and navigation based on visible rows. Enable it for `/keymap`, reserving result space to keep controls stable across filtering and tab changes. ## Testing Add regression tests and snapshots for contrast after palette conversion, shaded diff backgrounds, shortcut typography, picker resizing and pagination, Unicode tab truncation, and stable `/keymap` geometry. GitOrigin-RevId: f3c71879d3bcc4e3b29d70e63cf377dea5914916Eric Traut · 132c2be2 · 2026-09-19
- 4.2ETVPersist text elements through TUI input and history (#9393) Continuation of breaking up this PR https://github.com/openai/codex/pull/9116 ## Summary - Thread user text element ranges through TUI/TUI2 input, submission, queueing, and history so placeholders survive resume/edit flows. - Preserve local image attachments alongside text elements and rehydrate placeholders when restoring drafts. - Keep model-facing content shapes clean by attaching UI metadata only to user input/events (no API content changes). ## Key Changes - TUI/TUI2 composer now captures text element ranges, trims them with text edits, and restores them when submission is suppressed. - User history cells render styled spans for text elements and keep local image paths for future rehydration. - Initial chat widget bootstraps accept empty `initial_text_elements` to keep initialization uniform. - Protocol/core helpers updated to tolerate the new InputText field shape without changing payloads sent to the API.charley-oai · eb90e20c · 2026-01-20
- 4.2ETVStabilize transcript and composer interactions in the TUI (#46749) ## Why Composer suggestions and transient hints should preserve the draft and transcript context. Submitting a local command while reading older history should reveal its output. ## What changed - Separate persistent status from transient footer hints in the owned transcript, sharing layout calculations for rendering, measurement, and cursor placement. - Overlay command and mention suggestions on history without moving the composer; hide retained suggestions while transcript interactions own input. - Show responsive, grouped shortcut help above the composer and let `Esc` dismiss it before interrupting a task or navigating history. - Resume transcript following for submissions, slash commands, and last-response copy results while preserving the reading position during background updates. - Reset transcript interaction state after prompt reverts and refresh retained usage information without rewriting terminal history. ## Testing Add regression tests and snapshots for popup placement, clipped layouts, footer focus, shortcut dismissal, local-command scrolling, prompt reverts, and pagination while browsing prompts. GitOrigin-RevId: 890147313d53f050503165a57610c381ffb93ef5Eric Traut · de6ae71a · 2026-09-20