🤖 feat: AI-generated sidebar status via small-model loop#3238
Conversation
|
Preview deployment for your docs. Learn more about Mintlify Previews.
💡 Tip: Enable Workflows to automatically generate PRs for you. |
|
@codex review Please review this PR. Key areas:
|
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: f1c982f6c7
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
|
@codex review Addressed the partial.json feedback: |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: b52db3d1d4
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
|
@codex review Addressed the fairness feedback: |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 265f907bf5
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
|
@codex review Fixed the persist-vs-hash ordering. |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 3d2ac73d1c
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
|
@codex review Made |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 3b118b76b2
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
|
@codex review Two follow-ups:
|
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 2264113ab8
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
|
@codex review Addressed the lifecycle hazard: |
|
Codex Review: Didn't find any major issues. Breezy! ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
If Codex has suggestions, it will comment; otherwise it will react with 👍. Codex can also answer questions or update the PR. Try commenting "@codex address that feedback". |
|
@codex review Refactor pass for simplicity / minimal LoC / maintenance burden. Net delta: -438 lines.
All previously-resolved review concerns (round-robin fairness, persist-then-hash, partial-stream inclusion, lifecycle stop guard, transient status precedence) are preserved and still tested. |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 5a67d08ddf
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
|
@codex review Addressed P1: |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 5325acf7a2
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
|
@codex review Addressed P2: added a |
|
Codex Review: Didn't find any major issues. More of your lovely PRs please. ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
If Codex has suggestions, it will comment; otherwise it will react with 👍. Codex can also answer questions or update the PR. Try commenting "@codex address that feedback". |
|
@codex review Addressed three review threads:
Net diff: +83 / -163. All previously-resolved Codex concerns (recency preservation, pre-generator stop guard, partial-stream inclusion, round-robin fairness, persist-then-hash, lifecycle stop guard) are preserved. |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 9b9aafe40b
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
|
@codex review Addressed P2: inverted the precedence in the active-workspace branch so Added a regression test that gives the workspace BOTH live aggregator todos AND a persisted activity todoStatus, asserting the persisted one wins. |
|
Codex Review: Didn't find any major issues. You're on a roll. ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
If Codex has suggestions, it will comment; otherwise it will react with 👍. Codex can also answer questions or update the PR. Try commenting "@codex address that feedback". |
|
@codex review Bug fix prompted by deployed behavior: most workspaces were showing the literal phrase "Awaiting next task" in the sidebar. Root cause was the small-model prompt itself, which both:
Reframed around "most recent activity", dropped the special idle branch, and explicitly forbid generic placeholders so the model names the concrete activity from the transcript instead. |
|
Codex Review: Didn't find any major issues. Already looking forward to the next diff. ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
If Codex has suggestions, it will comment; otherwise it will react with 👍. Codex can also answer questions or update the PR. Try commenting "@codex address that feedback". |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: fc33fa5cbe
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
|
@codex review Fixed the empty-transcript starvation path: non-racy empty transcripts now consume observed recency so they do not keep sorting ahead of populated workspaces under max-concurrency=1, while the recent first-recency guard still protects the sendMessage recency/history race. |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 40d38dd213
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
|
@codex review Fixed the pre-provider failure starvation path: config/auth failures now consume the observed recency priority while leaving |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: bb21f74090
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
|
@codex review Addressed the pre-provider/history race by tracking the last transcript hash the scheduler examined separately from the settled dedup hash. Recent recency bumps now wait if history still shows the same transcript, including pre-provider retry states, while older/config-failure recency is still consumed to avoid starvation. |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 37a9dc9a72
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
|
@codex review Fixed stale dedup recency priority: once the recent history-catchup guard has expired, unchanged transcript dedup skips now consume observed recency so the workspace stops bypassing cadence and cannot starve other workspaces. |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: be1bf725a0
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
|
@codex review Fixed startup behavior too: AgentStatusService.start() now kicks off one immediate runTick in addition to scheduling future interval ticks, matching the no-startup-delay comment while preserving max-concurrency=1. |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 07e2ef0c88
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
|
@codex review Fixed in-flight stale writes: after generation returns, AgentStatusService now reloads the workspace snapshot and drops the result if recency advanced beyond the value observed at dispatch, so a slow provider response cannot persist/emit a pre-pivot status after a newer user turn. |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: c8f8d35d80
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
|
@codex review Fixed the atomicity issue: ExtensionMetadataService.setSidebarStatus now accepts a serialized-mutation guard ( |
|
Codex Review: Didn't find any major issues. Swish! ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
If Codex has suggestions, it will comment; otherwise it will react with 👍. Codex can also answer questions or update the PR. Try commenting "@codex address that feedback". |
The post-provider-failure / placeholder-rejection / successful-persist branches in runForWorkspace each ran the same two-line pair: markRecencyObserved() to consume any observed recency bump, then state.lastInputHash = inputHash to advance the dedup hash so the next tick won't regenerate against the same transcript. Hoist that pair into a settleOnTranscript() closure (next to the existing markRecencyObserved closure) so the three settlement branches read as a single named intent. The pre-provider-failure and empty/dedup-hit branches keep using bare markRecencyObserved() because they should still retry against the same transcript when conditions change — captured in a comment on the new helper so a future reader doesn't reintroduce the duplicated pair. Pure refactor — emitted JS, all six callsite branches, and the existing 25-test agentStatusService suite (including the post-provider/placeholder/success/pre-provider/pre-provider-retry/dedup regression cases added in #3238) are all unchanged. 🤖 Generated with mux Co-Authored-By: Mux <noreply@coder.com>
The post-provider-failure / placeholder-rejection / successful-persist branches in runForWorkspace each ran the same two-line pair: markRecencyObserved() to consume any observed recency bump, then state.lastInputHash = inputHash to advance the dedup hash so the next tick won't regenerate against the same transcript. Hoist that pair into a settleOnTranscript() closure (next to the existing markRecencyObserved closure) so the three settlement branches read as a single named intent. The pre-provider-failure and empty/dedup-hit branches keep using bare markRecencyObserved() because they should still retry against the same transcript when conditions change — captured in a comment on the new helper so a future reader doesn't reintroduce the duplicated pair. Pure refactor — emitted JS, all six callsite branches, and the existing 25-test agentStatusService suite (including the post-provider/placeholder/success/pre-provider/pre-provider-retry/dedup regression cases added in #3238) are all unchanged. 🤖 Generated with mux Co-Authored-By: Mux <noreply@coder.com>
The post-provider-failure / placeholder-rejection / successful-persist branches in runForWorkspace each ran the same two-line pair: markRecencyObserved() to consume any observed recency bump, then state.lastInputHash = inputHash to advance the dedup hash so the next tick won't regenerate against the same transcript. Hoist that pair into a settleOnTranscript() closure (next to the existing markRecencyObserved closure) so the three settlement branches read as a single named intent. The pre-provider-failure and empty/dedup-hit branches keep using bare markRecencyObserved() because they should still retry against the same transcript when conditions change — captured in a comment on the new helper so a future reader doesn't reintroduce the duplicated pair. Pure refactor — emitted JS, all six callsite branches, and the existing 25-test agentStatusService suite (including the post-provider/placeholder/success/pre-provider/pre-provider-retry/dedup regression cases added in #3238) are all unchanged. 🤖 Generated with mux Co-Authored-By: Mux <noreply@coder.com>
The post-provider-failure / placeholder-rejection / successful-persist branches in runForWorkspace each ran the same two-line pair: markRecencyObserved() to consume any observed recency bump, then state.lastInputHash = inputHash to advance the dedup hash so the next tick won't regenerate against the same transcript. Hoist that pair into a settleOnTranscript() closure (next to the existing markRecencyObserved closure) so the three settlement branches read as a single named intent. The pre-provider-failure and empty/dedup-hit branches keep using bare markRecencyObserved() because they should still retry against the same transcript when conditions change — captured in a comment on the new helper so a future reader doesn't reintroduce the duplicated pair. Pure refactor — emitted JS, all six callsite branches, and the existing 25-test agentStatusService suite (including the post-provider/placeholder/success/pre-provider/pre-provider-retry/dedup regression cases added in #3238) are all unchanged. 🤖 Generated with mux Co-Authored-By: Mux <noreply@coder.com>
The post-provider-failure / placeholder-rejection / successful-persist branches in runForWorkspace each ran the same two-line pair: markRecencyObserved() to consume any observed recency bump, then state.lastInputHash = inputHash to advance the dedup hash so the next tick won't regenerate against the same transcript. Hoist that pair into a settleOnTranscript() closure (next to the existing markRecencyObserved closure) so the three settlement branches read as a single named intent. The pre-provider-failure and empty/dedup-hit branches keep using bare markRecencyObserved() because they should still retry against the same transcript when conditions change — captured in a comment on the new helper so a future reader doesn't reintroduce the duplicated pair. Pure refactor — emitted JS, all six callsite branches, and the existing 25-test agentStatusService suite (including the post-provider/placeholder/success/pre-provider/pre-provider-retry/dedup regression cases added in #3238) are all unchanged. 🤖 Generated with mux Co-Authored-By: Mux <noreply@coder.com>
The post-provider-failure / placeholder-rejection / successful-persist branches in runForWorkspace each ran the same two-line pair: markRecencyObserved() to consume any observed recency bump, then state.lastInputHash = inputHash to advance the dedup hash so the next tick won't regenerate against the same transcript. Hoist that pair into a settleOnTranscript() closure (next to the existing markRecencyObserved closure) so the three settlement branches read as a single named intent. The pre-provider-failure and empty/dedup-hit branches keep using bare markRecencyObserved() because they should still retry against the same transcript when conditions change — captured in a comment on the new helper so a future reader doesn't reintroduce the duplicated pair. Pure refactor — emitted JS, all six callsite branches, and the existing 25-test agentStatusService suite (including the post-provider/placeholder/success/pre-provider/pre-provider-retry/dedup regression cases added in #3238) are all unchanged. 🤖 Generated with mux Co-Authored-By: Mux <noreply@coder.com>
## Summary Long-lived auto-cleanup PR that accumulates low-risk, behavior-preserving refactors picked from recent `main` commits. ## Changes ### Collapse `getCommandGhostHint` experiment gates into a table The slash-command goal feature added in coder#3235 introduced a third experiment-gating block to `getCommandGhostHint` (`src/browser/utils/slashCommands/registry.ts`), structurally identical to the existing `heartbeat` block: each matched a specific `commandKey`, wrapped `isExperimentEnabled(<flag>)` in a `try/catch`, and returned `null` on either disabled-experiment or thrown-check. The two blocks already drifted — the goal branch's `catch` had no comment, the heartbeat branch's had one explaining that hiding on a thrown check is the safer default — and a fourth gated command would have copy-pasted yet another nine-line `if (commandKey === "...") { try { ... } catch { ... } }` block. Hoist the (commandKey → ExperimentId) mapping into a `COMMAND_GHOST_HINT_EXPERIMENT_GATES` table and collapse both `if`-blocks into a single lookup + `try/catch`. Non-gated commands (`/compact`, `/model`, etc.) still skip the gate check identically because the lookup returns `undefined`, so the surrounding `null != undefined` guard short-circuits exactly the way the absent `if (commandKey === ...)` branches did. Adding a fourth gated command becomes a one-line table entry, with the safer "hide on thrown check" comment captured once on the shared `catch` instead of one branch's `catch` (and the other's accidentally missing it). Pure refactor — emitted JS, the gate semantics for `/goal` and `/heartbeat` (disabled experiment hides the ghost-hint; thrown experiment check also hides it, per the heartbeat branch's pre-existing comment), the fall-through `SLASH_COMMAND_DEFINITION_MAP.get(commandKey)?.inputHint ?? null` lookup for non-gated commands, and the existing 10-test `ghostHint` suite (plus the 136-test full `slashCommands` suite) are all unchanged. <details> <summary>Previous cleanups</summary> ### Drop dead `?? []` fallback for `sanitizeSchema.tagNames` The JSX-tag preservation plugin added in coder#3256 built an `ALLOWED_RAW_HTML_TAG_NAMES` Set by lowercasing `sanitizeSchema.tagNames`, guarded with a `(sanitizeSchema.tagNames ?? []).map(...)` fallback. `sanitizeSchema` is constructed locally in the same module via `{ ...defaultSchema, tagNames: [...] }` with a concrete array literal — after the spread/override, `tagNames` is unconditionally a non-empty `string[]`, so the outer `?? []` fallback can never fire and only serves to obscure that invariant from future readers. Drop the outer fallback and add a comment capturing the rationale (the inner `...(defaultSchema.tagNames ?? [])` spread still guards the library type, which is genuinely optional). A future refactor that touches the schema construction now has to spell out any change that could legitimately make `tagNames` undefined, instead of inheriting the dead guard by copy-paste. Pure refactor — emitted JS, the `rehypePreserveUnknownRawHtml` plugin's behavior (parsed via the same lowercased Set), the unknown-tag-as-text preservation semantics added in coder#3256, and the existing `MarkdownRenderer.raw-html` test suite are all unchanged. ### Extract `readInlineHeightPx` helper in `useAutoResizeTextarea` The chat input auto-resize stabilization in coder#3263 added a second `Number.parseFloat(el.style.height)` + `Number.isFinite(...)` pair to verify that the cached pixel height still reflects the textarea's inline style, sitting a handful of lines from the existing pair in the `canOnlyGrow` first-render fallback. Both spell out the same intent — read the textarea's inline `height` style as a finite px number, or treat it as missing — but as two independent expressions they were one stray edit away from drifting on what counts as a usable inline height (e.g. how to handle an empty string, a non-px unit, or the `NaN` that `parseFloat` returns for either). Extract a private `readInlineHeightPx(el): number | null` helper that captures the parseFloat + isFinite pair in one place, with the rationale captured on the helper itself so a future caller (e.g. another resize-coordination path that also needs to read the inline height) doesn't reintroduce the duplicated check. The two call sites collapse to a single named call, and the surviving branches replace `Number.isFinite(appliedHeight)` / `Number.isFinite(currentInlineHeight)` with `!== null` checks (functionally identical because the helper already filters non-finite values, plus TypeScript narrows `number | null` cleanly without a separate type guard). Pure refactor — emitted writes, the `canOnlyGrow` grow-only semantics, the post-resize "restore the px height after `auto` writes or external clears" verification added in coder#3263, and the existing 5-test `useAutoResizeTextarea` suite (including the regression cases for capped large-text deletion and external inline-height clears added in coder#3263) are all unchanged. ### Extract `settleOnTranscript` helper in `agentStatusService` The new `AgentStatusService.runForWorkspace` (added by the AI-generated sidebar status loop in coder#3238) had the same two-line pair — `markRecencyObserved()` to consume any observed recency bump, then `state.lastInputHash = inputHash` to advance the dedup hash — repeated across the three branches that produce a definitive outcome for the current transcript: post-provider failure (model refused tool / rate limit / persistent provider error), placeholder rejection (defense-in-depth filter for generic "Awaiting next task" output), and successful persist. The other three branches (empty workspace, idle/frozen dedup hit, pre-provider auth/config failure) intentionally call only `markRecencyObserved()` because they should still retry against the same transcript when conditions change. Hoist the duplicated pair into a `settleOnTranscript()` closure declared next to the existing `markRecencyObserved` closure inside `runForWorkspace`, with a comment capturing why the three settlement branches advance the dedup hash and the three retry branches don't, so a future reader doesn't reintroduce the duplicated pair (or accidentally swap one for the other). The three settlement call sites collapse to a single named call, and the three retry call sites keep their bare `markRecencyObserved()` shape. Pure refactor — emitted JS, the six branch behaviors (settle vs retry), the dedup-hash semantics documented on the `lastInputHash` field, and the existing 25-test `agentStatusService` suite (including the post-provider / placeholder / success / pre-provider / pre-provider-retry / dedup regression cases from coder#3238) are all unchanged. ### Collapse `ReviewPanel` selection-validity branches `ReviewPanel.tsx`'s post-filter selection-validity effect (touched by the immersive sidebar fix in coder#3249) split into two structurally identical arms: an `isImmersive` branch that validated `selectedHunkId` against `hunks` (the unfiltered diff) and a non-immersive branch that validated against `filteredHunks`. Both arms then ran the same `setSelectedHunkId(filteredHunks[0].id)` reset on miss, with the immersive arm needing its own early `return` to avoid double-running the reset. The duplication invited drift between the two modes — e.g. a future hide-read-style filter that adjusts the immersive reset would have to keep both arms in lockstep by hand. Pick the validity list up front (`const validityList = isImmersive ? hunks : filteredHunks;`) and run a single `selectedHunkId && validityList.some(...)` check, with a short comment capturing the rationale so a future reader doesn't reintroduce the split. The early-return on `filteredHunks.length === 0`, the dependency list (`[filteredHunks, hunks, isImmersive, selectedHunkId, setSelectedHunkId]`), and the immersive-aware "only reset when the hunk has truly disappeared from the diff" semantics added in coder#3249 are all preserved. Pure refactor — emitted JS, the selection-reset trigger conditions, and the existing 5-test `ImmersiveReviewView` suite (including the two immersive regression tests added in coder#3249) are all unchanged. ### Extract `renameAliasField` helper for bash tool preprocess The bash tool's `z.preprocess` shim (in `src/common/utils/tools/toolDefinitions.ts`) normalizes quirky model emissions to canonical field names. After the DeepSeek v4 fix in coder#3247 added a second `description` → `display_name` rename block (mirroring the existing `command` → `script` rename), the two blocks were structurally identical: skip when canonical is already a string, drop the alias via destructure, re-spread with the canonical name. Each alias still required its own `as Record<string, unknown> & { <alias>: string }` cast plus the same three-line destructure/spread. Extract a private `renameAliasField(obj, alias, canonical)` helper that captures the rename pattern in one place, with the rationale for why aliases exist (and why they stay undocumented) noted on the helper. The call site collapses to two named lines that read as the intent (`rename command to script`, `rename description to display_name`) instead of fifteen lines of mostly-identical destructure/spread. A future alias (e.g. another quirky-model field) becomes a one-line addition. Pure refactor — emitted JS, the canonical-field-wins precedence, the "no-op when neither alias nor canonical is a string" branches, and the existing 36-test `toolDefinitions` suite (including the four `command`/`description` alias precedence cases added in coder#3247) are all unchanged. ### Derive `TokenRecord` from `BrowserBridgeTokenPayload` `BrowserBridgeTokenManager` (in `src/node/services/browser/BrowserBridgeTokenManager.ts`, touched by the new other-workspace session picker in coder#3243) declared two structurally identical interfaces side by side: a private `TokenRecord` (the stored token state) and an exported `BrowserBridgeTokenPayload` (the `validate()` return shape), with `TokenRecord` differing only by the extra `expiresAtMs: number` deadline. `validate()` then rebuilt the payload by listing each of the four shared fields a third time. The coder#3243 commit made this drift surface especially visible — adding `allowOtherWorkspaceSession` required parallel edits in `TokenRecord`, `BrowserBridgeTokenPayload`, the `mint()` insert, and the `validate()` rebuild. Collapse `TokenRecord` to `extends BrowserBridgeTokenPayload` (with the rationale captured in a short comment so a future reader doesn't reintroduce the duplicated field list) and rewrite the `validate()` rebuild as a rest-spread destructure (`const { expiresAtMs, ...payload } = record; return payload;`). The eslint config already enables `ignoreRestSiblings: true`, so the unused `expiresAtMs` binding doesn't trigger `no-unused-vars`. A future payload field (e.g. another scoping flag) now lives in exactly one place — the exported `BrowserBridgeTokenPayload` interface — and automatically flows through both the stored record shape and the validate-time rebuild. Pure refactor — emitted JS, the TTL semantics, the `null`-on-expired path, and the existing 5-test `BrowserBridgeTokenManager` suite (including the new "preserves explicit other-workspace session scope" case) are all unchanged. ### Extract `detachLanguageModelCleanup` helper in `languageModelCleanup` `moveLanguageModelCleanup` and `runLanguageModelCleanup` (in the new `src/node/services/languageModelCleanup.ts` from the OpenAI WebSocket transport opt-in PR coder#3241) both implemented the same "look up the attached cleanup, delete the symbol slot, return it" sequence inline. The duplication is structurally identical — same `LanguageModelWithCleanup` cast, same symbol read, same `delete` — so both call sites were one stray edit away from drifting on whether the slot gets cleared before or after the cleanup runs (which matters because `attachLanguageModelCleanup` asserts the slot is empty). Extract a private `detachLanguageModelCleanup` helper that does the single-shot pop in one place and returns the cleanup (or `undefined`). Both surviving public functions reduce to their intent: `move` re-attaches the popped cleanup to the target, and `run` invokes it inside the existing `try/catch` that swallows + logs failures. A short comment captures the rationale on the helper itself so the next caller (e.g. a future "cancel without invoking" path) can't accidentally leak a slot. Pure refactor — emitted JS, the symbol's single-shot semantics, and the existing 6-test `languageModelCleanup` suite are all unchanged. ### Extract `isAgentTaskActiveStatus` predicate in `task_await` Three call sites in `src/node/services/tools/task_await.ts` gated on the active agent-task subset (`"queued" | "running" | "awaiting_report"`) by inlining the three-arm equality check. The new task_await elapsed-timing commit (coder#3234) extended both the `timeoutMs === 0` branch and the `timed out` catch-branch symmetrically with `...getAgentTaskElapsedField(taskId)`, making the two structurally identical `{ status, taskId, ...elapsed }` returns sit a handful of lines from a third copy in the `ForegroundWaitBackgroundedError` branch that picks between the same three values to coalesce against a `"running"` fallback. Add a local `AgentTaskActiveStatus` subset alias plus an `isAgentTaskActiveStatus` type predicate near the existing `coerceTimeoutMs` / `parseTimestampMs` helpers, and collapse all three inline checks to a single call. The predicate narrows the nullable `AgentTaskStatus | null` return of `getAgentTaskStatus` to the active subset, so the surviving `{ status, taskId, ... }` returns keep their narrowed `status` field with no `as const` gymnastics. The new comment captures the rationale so a future field added to one of the active-status branches won't silently drift away from the others. Pure refactor — emitted return shapes (including `elapsed_ms`), narrowed `status` literals, and the existing 17-test `task_await` suite are all unchanged. ### Drop dead length guard in `parseBedrockModelName` `secondPart` The early `dotParts.length < 2` return at the top of `parseBedrockModelName` (in `src/common/utils/ai/modelDisplay.ts`) already guarantees `dotParts.length >= 2` by the time `secondPart` is computed, which makes the `dotParts.length > 1 ? dotParts[1].toLowerCase() : ""` ternary's empty-string branch unreachable. The DeepSeek V4 commit (coder#3237) extended the surrounding formatter without touching this site, but the new DeepSeek branch made the asymmetry more obvious — the line directly above (`firstPart`) already accesses `dotParts[0].toLowerCase()` without a guard, so the ternary on `secondPart` was the odd one out. Inline to a direct `dotParts[1].toLowerCase()` access (matching `firstPart`'s shape) and capture the rationale in a one-line comment so a future reader doesn't reintroduce the guard. Pure dead-code cleanup — emitted JS, runtime behavior, and the existing 18-test `modelDisplay` suite (including the new DeepSeek + existing Bedrock cases) are all unchanged. ### Drop unused `workspaceName` param from `parseRuntimeString` The second argument to `parseRuntimeString` (in `src/browser/utils/chatCommands.ts`) was named `_workspaceName` (underscore-prefixed = unused) when it was introduced in the original SSH runtime PR, and has never been referenced by the function body — error messages are all generic and don't include any workspace-specific context. The `/new`-mirrors-`/fork` simplification (coder#3230) made the noise especially visible: the new `createNewWorkspace` call site had to pass `options.workspaceName ?? "(auto-generated)"` purely to satisfy the signature, and added a comment claiming `parseRuntimeString only uses the name for error reporting context` — but the function never reads it. Mobile already had the cleaner shape (`parseRuntimeStringForMobile(runtime?: string)`). Drop the parameter from the signature, drop the placeholder + misleading comment at the only non-test caller (`createNewWorkspace`), and drop the `workspaceName = "test-workspace"` constant + 23 second-arguments in `chatCommands.test.ts`. Pure dead-parameter cleanup — emitted JS, error messages, and runtime behavior are all identical, and the desktop signature now matches mobile's. ### Drop redundant `isPlanHandoffAgent` boolean in `streamContextBuilder` The `isPlanHandoffAgent` boolean in `buildPlanInstructions` was extracted when the gate was `effectiveAgentId === "exec" || effectiveAgentId === "orchestrator"`. After coder#3224 ripped out the Orchestrator agent, the boolean collapsed to a single equality check (`effectiveAgentId === "exec"`), and the trailing `else if (isPlanHandoffAgent && chatHasStartHerePlanSummary)` redundantly re-evaluated the same gate just to log a debug line. Replace the flat `if/else if` with a nested `if (effectiveAgentId === "exec") { … }` that tests the Start Here summary inside, removing the duplicate gate re-check and the now-meaningless boolean. A short comment captures the rationale so a future reader doesn't reintroduce the alias. Pure refactor — emitted control flow, the debug log, and `planContentForTransition` assignments are identical. ### Extract `seedScrollDirectionBaseline` helper in `useAutoScroll` `useAutoScroll` (touched by coder#3226) seeds `lastScrollTopRef` from two call sites — `jumpToBottom` and `disableAutoScroll` — to keep `handleScroll`'s released-branch direction check (`currentScrollTop > previousScrollTop`) honest after a programmatic ownership transfer that may not produce a follow-up scroll event. Both sites repeated the same write (`lastScrollTopRef.current = contentRef.current?.scrollTop ?? 0`) and a multi-line block comment re-explaining the same shared rationale. Extract a `seedScrollDirectionBaseline` `useCallback` with the shared rationale captured once on the helper itself. Each call site reduces to a single call plus a one-line comment naming the site-specific reason it doesn't get a free scroll-event refresh (`stickToBottom` skips the write at max; `disableAutoScroll` never fires a scroll event itself). The dependency arrays for `jumpToBottom` and `disableAutoScroll` add the new helper, which has empty deps (`useCallback(..., [])`) so its identity is stable across renders — no extra re-creations of the surrounding callbacks. This shrinks the surface area for future drift: a third path that needs to flip `autoScrollRef`/`programmaticDisableRef` without a guaranteed scroll-event tail (e.g. the deferred workspace-switch hydration race called out in coder#3226's "Pains") can call the helper instead of duplicating the rationale a third time. Pure refactor — emitted writes, write order, and ref values are identical, and the existing 25-test `useAutoScroll` suite (including the 6 new regression tests added in coder#3226) passes unchanged. ### Extract `pushStreamErrorRow` helper in `StreamingMessageAggregator` `buildDisplayedMessagesForMessage` now has two branches that synthesize a `stream-error` `DisplayedMessage`: the existing `message.metadata?.error` path and the `finishReason === "length"` path added in coder#3223. Both pushes were structurally identical, differing only in `id` suffix, `error` string, and `errorType` — the parent-message-derived fields (`historyId`, `historySequence`, `model`, `routedThroughGateway`, `timestamp`) were duplicated across both call sites. Extract a local `pushStreamErrorRow` closure that captures the shared fields once. Each branch reduces to a single call passing the three differing values. The `model` access switches from `message.metadata.model` (which relied on the outer `if (message.metadata?.error)` narrowing) to `message.metadata?.model`, which is functionally identical when `metadata` is defined and falls through to `undefined` otherwise — same emitted value either way. This shrinks the surface area for future drift: the next branch added (e.g. a different finish-reason synthesis) can't accidentally pass a stale `model` or forget `routedThroughGateway`. Pure refactor — emitted `DisplayedMessage` objects are identical, and the existing 77-test SMA suite (including the 6 new max-tokens tests) passes unchanged. ### Drop redundant `GuardAnchors` type alias in `file_edit_insert` In `src/node/services/tools/file_edit_insert.ts`, `GuardAnchors` was defined as `Pick<InsertContentOptions, "insert_before" | "insert_after">`, but `InsertContentOptions` itself is already `Pick<FileEditInsertToolArgs, "insert_before" | "insert_after">` after the `.nullish()` strict-mode refactor in coder#2250 stripped the `InsertContentOptions` interface down to those same two fields. The two aliases became structurally identical, so `GuardAnchors` is dead. Drop the alias and use `InsertContentOptions` for the two callers (`insertWithGuards`, `resolveGuardAnchor`). Both names were file-local; no exports change. The function names already convey "guard" context, so the parameter type doesn't need to repeat it. This noise was especially visible to the new guardless-empty-file path added in coder#3220 since it sits next to `insertContent` which uses `InsertContentOptions`. </details> ## Validation - `bun test src/browser/utils/slashCommands/ghostHint.test.ts` — 10/10 pass. - `bun test src/browser/utils/slashCommands/` — 136/136 pass (full slashCommands suite, covers `parser`, `suggestions`, `goal`, `heartbeat`, `idle`, `new`, `fork`, `compact`, `parser_multiline`, and `ghostHint`). - `make static-check` — eslint, tsgo (×2), prettier, shfmt, ruff, code-to-docs link check all pass. ## Risks Minimal — purely a local table-extraction inside `getCommandGhostHint`. The table-driven lookup short-circuits exactly the way the two if-blocks did: for non-gated commands `COMMAND_GHOST_HINT_EXPERIMENT_GATES[commandKey]` returns `undefined`, the `gateExperimentId != null` guard fails, and execution falls through to the existing `SLASH_COMMAND_DEFINITION_MAP.get(...)` lookup. For `/goal` and `/heartbeat` the only behaviour change is that the goal branch now inherits the heartbeat branch's "hide on thrown experiment check" comment (the heartbeat branch's `catch` already had the comment; the goal branch's `catch` did not). No imports, types, or experiment IDs are introduced beyond the existing `ExperimentId` type-only import from `@/common/constants/experiments`. --- Auto-cleanup checkpoint: 23440dd --- _Generated with `mux` • Model: `anthropic:claude-opus-4-7` • Thinking: `xhigh`_ <!-- mux-attribution: model=anthropic:claude-opus-4-7 thinking=xhigh --> --------- Co-authored-by: mux-bot[bot] <264182336+mux-bot[bot]@users.noreply.github.com> Co-authored-by: mux-bot <mux-bot@coder.com> Co-authored-by: mux <bot@mux.coder.com> Co-authored-by: ammar-agent <ammar+ai@ammar.io> Co-authored-by: mux-auto-cleanup <noreply@anthropic.com> Co-authored-by: Mux <noreply@coder.com>
2 participants
Summary
Adds an
AgentStatusServicethat produces the sidebar agent-status text using the same "small model" path as workspace title generation (NAME_GEN_PREFERRED_MODELS— Claude Haiku → GPT-mini), instead of inferring it from the workspace's TODO list. The status is regenerated on a focus-aware cadence, skips when the trailing transcript hash is unchanged, and is fed at most ~8k trailing transcript tokens.Background
Today the sidebar's per-workspace status string is derived from
deriveTodoStatus, which picks the first in-progress / pending / last-completed TODO. That's a noisy proxy for "what is the agent doing right now":todo_write, so a long bash run or a multi-step assistant turn shows a stale TODO.The user asked for the status to come from a small LLM call instead, with sensible cost controls (token cap, cadence, dedup, focus gating).
Implementation
Constants (
src/constants/agentStatus.ts)30s, unfocused cadence120s, scheduler tick10s, startup delay30s.80messages, ≤8000tokens, with a per-message char cap of4000.1global generation in flight.Small-model path (
src/node/services/workspaceStatusGenerator.ts)workspaceTitleGenerator.ts.streamText+ a newpropose_statustool (added totoolDefinitions.ts); reusesmapNameGenerationError/mapModelCreationErrorso all the existing 401/402/403/429/5xx classifications apply.<transcript>markers and instructs the model to focus on current activity rather than overall task scope.propose_statusis excluded from defaulttools.tsregistration with the same comment aspropose_name.Scheduler (
src/node/services/agentStatusService.ts)setIntervalticking every 10s. Per-workspacelastRanAt+ a current focused/unfocused interval gate the actual generation.getLastMessages(80)+readPartial→ format → trim oldest entries until under the token budget (usingTokenizerServicewith a fallback model constant), then SHA-256 the transcript.lastRanAtascending soMAX_CONCURRENT=1doesn't starve workspaces deeper in the list (round-robin without an explicit queue).runTickawaits all dispatched workspaces (via trackedinFlightPromises) instead of fire-and-forget, so production schedulers and tests can observe completion deterministically.if (this.stopped) returnguards (post-generation, post-persist) so astop()mid-flight never leaks writes past the service's lifecycle.Focus tracking (
src/node/services/windowService.ts)WindowService.isFocused()delegates toBrowserWindow.isFocused()directly. Falls back totruein non-desktop contexts (CLI server, tests) so backend services don't throttle themselves to "unfocused" cadence when there is no window.Schema + persistence
aiStatusfield onExtensionMetadata+WorkspaceActivitySnapshotSchema.extensionMetadata.setAiStatusdirectly (bypassing the historicalemitWorkspaceActivityUpdatewrapper that would have swallowed disk errors) and emits the activity snapshot viaworkspaceService.emitWorkspaceActivityonly after the disk write resolves successfully.ExtensionMetadataService.save()re-throws on failure to support this;initialize()keeps its startup-safety semantics by wrapping its mutation calls in a logged try/catch per AGENTS.md.Frontend (
src/browser/stores/WorkspaceStore.ts)transientStatus ?? aiStatus ?? todoStatus.transientStatuscollapsesdisplayStatus(inactive workspaces) andaggregator.getAgentStatus()(active workspaces) so explicit system-set messages always beat aiStatus regardless of which branch produces them.applyWorkspaceActivitySnapshotdiffsaiStatusso a fresh small-model status triggers re-render.Validation
workspaceStatusGenerator.test.ts: prompt structure (transcript delimiters, empty-transcript sentinel) + empty-candidates short-circuit.agentStatusService.test.ts: generates on first tick, dedups when transcript is unchanged, regenerates after history changes, includes the partial assistant message mid-stream, skips empty workspaces, focused-vs-unfocused cadence boundaries, round-robin fairness across multiple workspaces, archived workspaces are skipped, stops cleanly mid-generation, persist failures don't advance the dedup hash.WorkspaceStore.test.ts: aiStatus beats todoStatus; displayStatus still beats aiStatus.make static-checkclean (eslint, typecheck, prettier, doc-link check).make testshows only pre-existing flaky-timing failures present onmain(CLI auth + GitStatusStore subset). My new files pass cleanly.Risks
mapNameGenerationErrorretry budget capped at 3 candidates.Codex review history (resolved)
chat.jsonl. Fixed by appendingreadPartialso the hash refreshes mid-stream.MAX_CONCURRENT=1and fixed iteration order, deep workspaces would starve. Fixed withlastRanAtascending sort.updateAiStatuswent through a wrapper that silently swallowed disk errors. Fixed by callingsetAiStatusdirectly and updating the hash only on success.save()swallowing errors:ExtensionMetadataService.save()itself swallowed errors. Fixed by re-throwing;initialize()wraps its calls in try/catch to preserve startup-safety.aggregator.getAgentStatus()was being buried under aiStatus for active workspaces. Fixed with thetransientStatustier.runForWorkspacecould persist/emit afterstop(). Fixed with twothis.stoppedchecks at await boundaries.Simplification refactor
After the original implementation landed and Codex approved, the diff was simplified for minimal LoC / maintenance burden (8 files, +232 / -670 lines):
aiStatusInputHashround-trip turned out to add ~50 LoC of complexity (hydration sweep, getter method, conditional-spread coercion) for ~one extra LLM call per workspace per app launch. Removed the field from the type, the coercion, the setter signature, andgetAiStatusInputHashentirely; in-session dedup still works.WindowService: droppedEventEmitterinheritance,setFocused,focusedfield, focus/blur listeners, and the test-stub defensive check.isFocused()delegates toBrowserWindow.isFocused()directly.AgentStatusService: dropped__test__exports,inFlightCount(useinFlightPromises.size),markInFlighthelper,drainInFlighthelper,hashesHydrated+hydratePersistedHashes,resolveTokenizerModel. Mergedtick()+runTick(). UsedFALLBACK_TOKENIZER_MODELconstant inbuildTrailingTranscriptso we don't double-callgetWorkspaceTitleModelCandidatesper workspace.Generated with
mux• Model:anthropic:claude-opus-4-7• Thinking:max• Cost:$58.06