You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Commit 60ac8ef
Browse filesBrowse the repository at this point in the historyBrowse files
Copy file name to clipboardExpand all lines: structure/runtime.md
+1-1Lines changed: 1 addition & 1 deletion
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -1,6 +1,6 @@
1
1
# Runtime
2
2
3
-
The minute sweep checks persisted activation deadlines locally; only missing deadlines trigger metadata discovery. See the [quota activation contract](providers/openai-tiers.md#public-provider-contract). Codex control-plane ids are defined in the dependency-free `src/codex/control-plane-models.ts`; `src/router.ts` routes these exact ids unchanged to canonical OpenAI, retaining the existing error when no OpenAI provider is enabled. They do not widen the public native roster. Catalog requests and scheduled source refreshes follow the [nonblocking catalog observation contract](catalog.md#shared-catalog). Async runtime persistence awaits the fresh [selection admission](codex-cli-update.md#worker-realm-and-the-selection-commit-window), rechecks observed selection/epoch and caller-currentness changes after that await, and returns detached frozen results; signal cancellation makes the persistence wrapper return null without writing, while raw async resolver rejections and exceptions without signal cancellation retain their rejection behavior; bounded exec rejects promptly and releases output readers on deadline or abort, independently awaits actual child exit, and escalates SIGTERM to SIGKILL after the shared 2-second grace (Windows termination is forceful). Known Codex.app and ChatGPT.app commands at any location or alias follow the [execution trust contract](clients/chatgpt-desktop.md#codex-app-execution-trust); automatic discovery also requires matching bundle identity; the existing ChatGPT bundle trust policy (safe ownership, permissions, non-symlink paths, and OpenAI signatures for both the bundle and the exact executable) is enforced at every execution boundary in src/codex/exec-boundary.ts before the resolved canonical target is spawned, not once at selection. Version probes retain that execution's bundle classification for selection, so a bare PATH alias of an app cannot supersede a discovered persisted runtime; selection does not resolve PATH again after the probe. Stored and displayed command spelling remains unchanged, and this observation grants no trust to a later execution. Claude overflow normalization examines at most 64 Ki characters, extracts stated context/request counts in separate forward scans, retains the bounded rewritten prefix, and leaves other error classifications intact; the managed native Messages SSE route follows the same contract (`tests/claude-integration/claude-prompt-too-long.test.ts`). Optional native desktop observation follows the [Codex Desktop startup gate and awaited teardown](clients/codex-desktop.md#proxy-startup-preference). Claude Code tier defaults are projected by `src/claude/model-default-settings.ts` into user settings only after explicit preview/apply, independently of launcher environment injection; see the [management contract](gui-and-management-api.md). The explicit projection pairs selected aliases with readable labels from `src/claude/context-windows.ts`, preserving custom labels and clearing only a label matching its current alias when an explicitly selected tier becomes a non-alias; unset tiers retain both settings model IDs and labels. No startup or periodic sync writes these defaults.
3
+
The minute sweep checks persisted activation deadlines locally; only missing deadlines trigger metadata discovery. See the [quota activation contract](providers/openai-tiers.md#public-provider-contract). Codex control-plane ids are defined in the dependency-free `src/codex/control-plane-models.ts`; `src/router.ts` routes these exact ids unchanged to canonical OpenAI, retaining the existing error when no OpenAI provider is enabled. They do not widen the public native roster. Catalog requests and scheduled source refreshes follow the [nonblocking catalog observation contract](catalog.md#shared-catalog). Async runtime persistence awaits the fresh [selection admission](codex-cli-update.md#worker-realm-and-the-selection-commit-window), rechecks observed selection/epoch and caller-currentness changes after that await, and returns detached frozen results; signal cancellation makes the persistence wrapper return null without writing, while raw async resolver rejections and exceptions without signal cancellation retain their rejection behavior; bounded exec rejects promptly and releases output readers on deadline or abort, independently awaits actual child exit, and escalates SIGTERM to SIGKILL after the shared 2-second grace (Windows termination is forceful). Known Codex.app and ChatGPT.app commands at any location or alias follow the [execution trust contract](clients/chatgpt-desktop.md#codex-app-execution-trust); automatic discovery also requires matching bundle identity; the existing ChatGPT bundle trust policy (safe ownership, permissions, non-symlink paths, and OpenAI signatures for both the bundle and the exact executable) is enforced at every execution boundary in src/codex/exec-boundary.ts before the resolved canonical target is spawned, not once at selection. Version probes retain that execution's bundle classification for selection, so a bare PATH alias of an app cannot supersede a discovered persisted runtime; selection does not resolve PATH again after the probe. Stored and displayed command spelling remains unchanged, and this observation grants no trust to a later execution. Claude overflow normalization examines at most 64 Ki characters, extracts stated context/request counts in separate forward scans, retains the bounded rewritten prefix, and leaves other error classifications intact; the managed native Messages SSE route follows the same contract (`tests/claude-integration/claude-prompt-too-long.test.ts`). On the translated Messages-to-Responses lane, `isContextOverflowError` in `src/claude/outbound.ts` decides overflow for the streamed `response.failed` handler, the non-stream error branch and the failed-JSON branch: a coded `context_length_exceeded` always counts; an uncoded or generic-coded (`server_error`, `upstream_error`, `api_error`, `internal_error`) refusal counts only on wording that asserts the request exceeds a size limit, never with a 401/403/429 status, a rate-limit or credential type, throughput wording or another specific code, and then answers 400 `prompt is too long` so Claude Code compacts (#8127, `tests/claude-integration/claude-uncoded-context-overflow.test.ts`). Optional native desktop observation follows the [Codex Desktop startup gate and awaited teardown](clients/codex-desktop.md#proxy-startup-preference). Claude Code tier defaults are projected by `src/claude/model-default-settings.ts` into user settings only after explicit preview/apply, independently of launcher environment injection; see the [management contract](gui-and-management-api.md). The explicit projection pairs selected aliases with readable labels from `src/claude/context-windows.ts`, preserving custom labels and clearing only a label matching its current alias when an explicitly selected tier becomes a non-alias; unset tiers retain both settings model IDs and labels. No startup or periodic sync writes these defaults.
0 commit comments