prreviewbuddy 0.71.2 → 0.73.1

package/CHANGELOG.md

@@ -1,5 +1,51 @@
 # Changelog
 
+## 0.73.1
+
+**A review's first update says how long updates take.** While an update runs, the page now
+compares it with this review's earlier updates, and on its first update with other updates in the
+same repository: "Previous update took 51s", "Updates to this review usually take about 1 minute",
+or "Updates in this repository usually take about 1 minute". An update is never compared with a
+fresh review, which does much more work. With no finished update in the repository, nothing is
+said.
+
+**The running time is the footer of the progress page.** Under the steps, the running time counts
+to the second ("Running for 1m 08s"), with how long earlier runs took beneath it and Cancel beside
+them as a proper button. The same footer appears while a review runs and while an update runs.
+
+## 0.73.0
+
+**KISS runs now count as reviews in usage telemetry.** With telemetry on, a KISS run sends the same
+two events a standard review does, `review_started` and `review_finished`, with the same agent,
+outcome and duration. Until now it sent neither, so somebody who only ran KISS looked as if they
+had installed PR Review Buddy and never used it. Both review events now carry one more field,
+`reviewType`, which is one of two words, `standard` or `kiss`: the kind of analysis, and nothing
+about your code, your repository or what the analysis said. It is the only field added. Updating a
+review with new commits still sends nothing. `prreviewbuddy config set telemetry-upload off` and
+`PRB_TELEMETRY=off` stop these exactly as before.
+
+## 0.72.1
+
+**Local usage logs count themes on the first read.** The `workspace_opened` log entry waited for
+the review to be marked Reviewed rather than for the analysis to finish, so it rarely fired before
+a reading session ended and the session summary recorded zero themes. It now waits only for the
+analysis. Nothing about what is sent changes: neither event leaves the machine.
+
+## 0.72.0
+
+**Both explanations stay on the card.** Ask Buddy on an issue or question now offers
+**Explain (technical)** and **Explain (simple)**, and either one keeps its answer under the finding,
+labelled **Explanation · Technical** or **Explanation · Simple**. A card holds one explanation: run
+the other kind and it replaces the one there, while both answers stay in Buddy AI. If a new
+explanation fails or is stopped, the card keeps the one it had. The same two names appear wherever
+Ask Buddy offers them, including on a selection.
+
+**Ask more follows up on the explanation you were reading.** If another tab replaces a card's
+explanation after you press Ask more, your question still goes with the one you were reading.
+
+**Simple explanations kept by 0.71 carry over.** They appear on their cards as
+Explanation · Simple, and Ask more and updates use them as before.
+
 ## 0.71.2
 
 **Buddy AI's opening line reads plainly.** Before your first question, Buddy AI now says "Answers

package/dist/{delete_review-BgIZpcsP.js → delete_review-B0f6rmy_.js}

@@ -3277,12 +3277,44 @@ function describeAuthorship(changeSet, pullRequest) {
 }
 //#endregion
 //#region ../../packages/review-harness/src/domain/ai_review/review_session.ts
+/**
+* The Ask Buddy actions whose answer is kept on the finding's card: `explain-detail` is
+* Explain (technical), `explain` is Explain (simple). The ids are the shared action ids in
+* selection_prompts.ts, which predate the labels.
+*/
+var EXPLANATION_ACTIONS = ["explain-detail", "explain"];
+function isExplanationAction(value) {
+	return EXPLANATION_ACTIONS.includes(value);
+}
+function explanationActionOf(explanation) {
+	return explanation.action ?? "explain";
+}
 /** The explanation of this finding, if the reviewer has asked for one. */
-function simpleExplanationOf(session, kind, id) {
-	return session.simpleExplanations?.find((entry) => entry.kind === kind && entry.id === id);
+function explanationOf(session, kind, id) {
+	return session.findingExplanations?.find((entry) => entry.kind === kind && entry.id === id);
 }
-/** The list with this finding's explanation replaced, never a second one beside it. */
-function withSimpleExplanation(entries, explanation) {
+/**
+* A session as stored by 0.71.0 to 0.71.2, which kept explanations under `simpleExplanations`, read
+* as this version stores it. Those were all Explain (simple), which an entry without an action
+* already means, so they move across unchanged. Where both lists name a finding the newer list's
+* entry wins: a later version wrote it. A session without the old list is returned as it was.
+*/
+function withLegacyExplanationsMoved(session) {
+	const legacy = session.simpleExplanations;
+	if (legacy === void 0) return session;
+	const { simpleExplanations: _moved, ...rest } = session;
+	const newer = session.findingExplanations ?? [];
+	const merged = [...(Array.isArray(legacy) ? legacy : []).filter((entry) => !explanationOf(session, entry.kind, entry.id)), ...newer];
+	return {
+		...rest,
+		...merged.length ? { findingExplanations: merged } : {}
+	};
+}
+/**
+* The list with this finding's explanation replaced, never a second one beside it. One slot per
+* finding whichever action wrote it: the newest explanation is the card's.
+*/
+function withExplanation(entries, explanation) {
 	return [...(entries ?? []).filter((entry) => !(entry.kind === explanation.kind && entry.id === explanation.id)), explanation];
 }
 /**
@@ -4431,12 +4463,12 @@ var WORKSPACES_DIR = join(STORE_ROOT, "workspaces");
 * Named so that a long run that rewrites the session, an update above all, can put these back from
 * the copy on disk when it saves: the reviewer may have answered a question or dismissed a finding
 * in the minutes it took. `handleState` writes the first two and nothing else; `handleAsk` writes
-* `simpleExplanations`, and only when the reviewer asked for one.
+* `findingExplanations`, and only when the reviewer asked for one.
 */
 var REVIEWER_SESSION_FIELDS = [
 	"answeredQuestionIds",
 	"dismissedIssues",
-	"simpleExplanations"
+	"findingExplanations"
 ];
 /**
 * A session a long run is about to write, with the reviewer's own fields taken from disk instead.
@@ -4668,7 +4700,13 @@ function deleteWorkspace(id) {
 }
 function loadWorkspace(id) {
 	try {
-		return JSON.parse(readFileSync(workspacePath(id), "utf8"));
+		const stored = JSON.parse(readFileSync(workspacePath(id), "utf8"));
+		if (!stored?.session || typeof stored.session !== "object") return stored;
+		const session = withLegacyExplanationsMoved(stored.session);
+		return session === stored.session ? stored : {
+			...stored,
+			session
+		};
 	} catch {
 		return null;
 	}
@@ -12831,13 +12869,28 @@ var RUNTIME_WORDS = {
 		last: "Last analysis of this review",
 		recent: "Recent analyses of this review",
 		usual: "Usually takes"
-	},
-	update: {
-		last: "Last update of this review",
-		recent: "Recent updates of this review",
-		usual: "Usually takes"
 	}
 };
+/**
+* How long earlier updates took, said from the most specific history there is: this review's own
+* updates, else the repository's, else nothing. Never both, and never a fresh review's runtime,
+* which is a different operation. The rule above holds at both levels: one run is stated exactly
+* and only several earn "usually". Several are always one figure, the median, never a range.
+*/
+function describeUpdateRuntimes(own, repository) {
+	const [ownRuns, repoRuns] = [own.slice(0, 3), repository.slice(0, 3)];
+	if (ownRuns.length === 1) return `Previous update took ${formatDuration(ownRuns[0])}`;
+	if (ownRuns.length > 1) return `Updates to this review usually take ${aboutMinutes(median(ownRuns))}`;
+	if (repoRuns.length === 1) return `Last update in this repository took ${formatDuration(repoRuns[0])}`;
+	if (repoRuns.length > 1) return `Updates in this repository usually take ${aboutMinutes(median(repoRuns))}`;
+	return null;
+}
+/** "less than a minute", "about 1 minute", "about 4 minutes". */
+function aboutMinutes(ms) {
+	const minutes = toMinutes(ms);
+	if (minutes === 0) return "less than a minute";
+	return `about ${minutes} ${minutes === 1 ? "minute" : "minutes"}`;
+}
 //#endregion
 //#region ../../packages/review-harness/src/runtime/store.ts
 var VERSION = 1;
@@ -12944,6 +12997,18 @@ function updateRuntimes(repoPath, reviewId) {
 		reviewType: "standard"
 	});
 }
+/**
+* Updates of any review in the repository, for a review's first update. No subject for the reason
+* `newReviewRuntimes` has none: a review not yet updated has no updates of its own. Still only
+* updates, because a fresh review of the same change is a different operation and much slower.
+*/
+function repoUpdateRuntimes(repoPath) {
+	return durationsOf({
+		operations: ["update"],
+		repoPath,
+		reviewType: "standard"
+	});
+}
 //#endregion
 //#region ../../packages/review-harness/src/workspace/runtime_history.ts
 /**
@@ -12977,10 +13042,15 @@ function analysisHistory(workspace, job) {
 function waitingHistory(workspace, runs) {
 	return {
 		job: isLiveJob(runs.job) ? analysisHistory(workspace, runs.job) : null,
-		update: isLiveJob(runs.updateJob) ? describeRuntimes(updateRuntimes(workspace.repoPath, workspace.id), RUNTIME_WORDS.update) : null,
+		update: isLiveJob(runs.updateJob) ? updateHistory(workspace) : null,
 		kiss: isLiveJob(runs.kissJob) ? analysisHistory(workspace, runs.kissJob) : null
 	};
 }
+/** This review's own updates first; the repository's only for a review never updated before. */
+function updateHistory(workspace) {
+	const own = updateRuntimes(workspace.repoPath, workspace.id);
+	return describeUpdateRuntimes(own, own.length > 0 ? [] : repoUpdateRuntimes(workspace.repoPath));
+}
 //#endregion
 //#region ../../packages/review-harness/src/domain/feedback/feedback_url.ts
 /** How each client names itself in the prefilled message, which is the half a screenshot carries. */
@@ -13273,6 +13343,7 @@ function project(record) {
 		cliVersion: record.cliVersion,
 		platform: record.platform,
 		agent: event.agent,
+		reviewType: event.reviewType,
 		occurredAt: record.at
 	};
 	return event.event === "review_finished" ? {
@@ -13351,7 +13422,7 @@ var LOG_PATH = join(STORE_ROOT, "events.jsonl");
 * disable is a different act from sending to a server, and the consent that suits one does not
 * carry to the other, so `events.jsonl` keeps recording the full event union either way and the
 * question asked on first run is about the wire. Twenty events are written here; three of them,
-* carrying eight fields between them, are the most that is ever sent.
+* carrying nine fields between them, are the most that is ever sent.
 */
 function record(event) {
 	try {
@@ -13693,10 +13764,11 @@ async function drive(jobId, workspaceId, runners, driver, signal) {
 				contextId: context.contextId
 			};
 			saveJob(job);
-			if (job.reviewType !== "kiss") record({
+			record({
 				event: "review_started",
 				reviewId: workspaceId,
 				agent: agentName(job.env.agentId),
+				reviewType: reviewKindOf(job),
 				changedFiles: totalChangedFiles(context.changeSet.data),
 				linesChanged: context.changeSet.data.files.reduce((total, file) => total + (file.additions ?? 0) + (file.deletions ?? 0), 0),
 				analysedFiles: Math.max(0, context.changeSet.data.files.length - context.inventoryOnlyFiles.length)
@@ -13777,11 +13849,11 @@ async function drive(jobId, workspaceId, runners, driver, signal) {
 * finished it. That is the number worth having: it is what somebody actually waited.
 */
 function finished(job, workspaceId, startedAt, ok, now) {
-	if (job.reviewType === "kiss") return;
 	record({
 		event: "review_finished",
 		reviewId: workspaceId,
 		agent: agentName(job.env.agentId),
+		reviewType: reviewKindOf(job),
 		ok,
 		durationMs: now() - startedAt
 	});
@@ -13831,6 +13903,10 @@ function recordAnalysisRuntime(job, workspaceId, from, durationMs, completedAt,
 * named ran under Claude. Recording those as unknown would invent a fourth agent out of the
 * oldest reviews on disk.
 */
+/** A job with no `reviewType` predates KISS and is a standard review. */
+function reviewKindOf(job) {
+	return job.reviewType === "kiss" ? "kiss" : "standard";
+}
 function agentName(id) {
 	const resolved = id ?? DEFAULT_AGENT_ID;
 	return isAgentId(resolved) ? resolved : "other";
@@ -16708,4 +16784,4 @@ function lineageIds(id) {
 	return [.../* @__PURE__ */ new Set([id, ...lineage.map((review) => review.id)])];
 }
 //#endregion
-export { MAX_WORKSPACE_PORT as $, isClaimed as $n, resolveBase as $r, markReadForReview as $t, relativeTime as A, isWorktreeHeld as An, parseLifecycle as Ar, partFiles as At, readIndexToken as B, reviewedShaOf as Bn, issuesOutstanding as Br, isArchitectureId as Bt, track as C, originRemote as Cn, activityAttributes as Cr, mapRuntimes as Ct, endRun as D, MANAGED_ROOT as Dn, lifecycleItemLabel as Dr, ARCHITECTURE_DIR as Dt, completeRun as E, ReviewBeingDeletedError as En, lifecycleChip as Er, describeRuntimes as Et, runBatch as F, describeSubject as Fn, queueStamp as Fr, trackedFiles as Ft, reviewsUrl as G, loadJob as Gn, simpleExplanationOf as Gr, createArchitectureNotification as Gt, clearServerRecordOwnedBy as H, endedAt as Hn, reviewerDispositions as Hr, announceJob as Ht, activateQueued as I, findExistingReview as In, readQueue as Ir, ownersOf as It, stopChild as J, isLiveJob as Jn, WITHHELD_REASON as Jr, eventName as Jt, serverEntryExists as K, saveJob as Kn, withSimpleExplanation as Kr, createPrState as Kt, compactQueue as L, checkCodeFreshness as Ln, doneVerb as Lr, QUOTE_LIMITS as Lt, extractPullRequestLinks as M, removeWorktree as Mn, queueAdd as Mr, repositoryKey as Mt, parsePullRequestLink as N, requestJobStop as Nn, queueAddMany as Nr, storeArchitecture as Nt, reportRun as O, createWorktree as On, lifecyclePill as Or, checkArchitecture as Ot, unsupportedReason as P, describeMovement as Pn, queueRemove as Pr, storedScopes as Pt, DEFAULT_WORKSPACE_PORT as Q, clearClaim as Qn, isCommitReview as Qr, markRead as Qt, queueForReview as R, checkReview as Rn, isIssueOutstanding as Rr, REPOSITORY_SCOPE as Rt, isDriving as S, remoteHost as Sn, LIFECYCLES as Sr, waitingHistory as St, beginRun as T, modelHelpLines as Tn, isLifecycle as Tr, RUNTIME_WORDS as Tt, ensureServer as U, fail as Un, processDiscussion as Ur, branchOf as Ut, bootstrapUrl as V, allJobs as Vn, questionsOutstanding as Vr, writeJsonAtomically as Vt, readServerRecord as W, isTerminal as Wn, recommendationsOf as Wr, subjectOf as Wt, workspaceUrl as X, progressSteps as Xn, resolveTarget as Xr, listNotifications as Xt, stopServer as Y, phasesFor as Yn, kissOutcome as Yr, headline as Yt, writeServerRecord as Z, purposeOf as Zn, DetachedHeadError as Zr, markAllRead as Zt, liveUpdateFor as _, extractJsonText as _n, reviewedRepositories as _r, telemetryUploadConsent as _t, wasBlocked as a, writeAgentPreference as ai, withUsageRecorded as an, kissDecisionRefusal as ar, failureOf as at, recordKissRun as b, forgeResolver as bn, workspaceRevision as br, PACKAGE_NAME as bt, RefreshUnavailableError as c, AGENT_IDS as ci, authorPreview as cn, lineagePosition as cr, reanalyseReview as ct, mergeUpdateInto as d, detectAgents as di, renderMarkdown as dn, mutateWorkspace as dr, append as dt, git as ei, pruneNotifications as en, decideKissRecommendation as er, MIN_WORKSPACE_PORT as et, updateReview as f, AgentCancelledError as fi, CODE_BLOCK_CLASS as fn, positionInLineage as fr, readEvents as ft, UpdateAlreadyRunningError as g, asString as gn, reviewedCommit as gr, recordTelemetryUploadConsent as gt, askCheckout as h, renderInlineText as hn, recentWorkspaces as hr, deliver as ht, liveJobsFor as i, readAgentPreference as ii, unreadCount as in, indexReviewGroups as ir, writePortPreference as it, batchIdentity as j, readMarker as jn, sectionOf as jr, readArchitecture as jt, stopRun as k, holdWorktree as kn, offersLifecycle as kr, loadArchitecture as kt, cancelUpdate as l, agentById as li, hasAuthorContent as ln, loadWorkspace as lr, runJob as lt, isUnchanged as m, escapeHtml as mn, previousKissResult as mr, stamp as mt, deleteReview as n, resolveModel as ni, removeNotification as nn, groupByLineage as nr, resolveWorkspacePort as nt, discardJob as o, CONFIG_PATH as oi, EXPLAIN_SIMPLY_PROMPT as on, kissDecisionsOf as or, prepareReview as ot, refreshPrContext as p, AgentUnavailableError as pi, CODE_COPY_CLASS as pn, presentedKissRunOf as pr, record as pt, spawnServer as q, PHASES as qn, describeAuthorship as qr, createPrUpdate as qt, lineageIds as r, clearAgentPreference as ri, toneOf as rn, indexKeptIds as rr, statedWorkspacePort as rt, FETCH_TIMEOUT_MS as s, STORE_ROOT as si, UNESTABLISHED_MOTIVATION as sn, latestKissRun as sr, settleReview as st, deleteLineage as t, resolveAgent as ti, readNotification as tn, followedRefName as tr, parseWorkspacePort as tt, cancellableHere as u, agentFor as ui, GITHUB_MARKDOWN as un, matchingWorkspaceIds as ur, startJob as ut, runningJobs as v, objectWithAnyOf as vn, summarise as vr, telemetryUploadEnabled as vt, RUN_KINDS as w, fillFileUrlTemplate as wn, activityLabel as wr, recordRuntimeBestEffort as wt, startKissJob as x, forgeFor as xn, workspaceStamps as xr, feedbackUrl as xt, agentEnvOf as y, claimRequestUrl as yn, touchWorkspace as yr, BUILD_VERSION as yt, settleQueue as z, followedBranch as zn, isQuestionOutstanding as zr, ancestorsOf as zt };
+export { MAX_WORKSPACE_PORT as $, isClaimed as $n, DetachedHeadError as $r, markReadForReview as $t, relativeTime as A, isWorktreeHeld as An, parseLifecycle as Ar, partFiles as At, readIndexToken as B, reviewedShaOf as Bn, issuesOutstanding as Br, isArchitectureId as Bt, track as C, originRemote as Cn, activityAttributes as Cr, mapRuntimes as Ct, endRun as D, MANAGED_ROOT as Dn, lifecycleItemLabel as Dr, ARCHITECTURE_DIR as Dt, completeRun as E, ReviewBeingDeletedError as En, lifecycleChip as Er, describeRuntimes as Et, runBatch as F, describeSubject as Fn, queueStamp as Fr, trackedFiles as Ft, reviewsUrl as G, loadJob as Gn, explanationActionOf as Gr, createArchitectureNotification as Gt, clearServerRecordOwnedBy as H, endedAt as Hn, reviewerDispositions as Hr, announceJob as Ht, activateQueued as I, findExistingReview as In, readQueue as Ir, ownersOf as It, stopChild as J, isLiveJob as Jn, withExplanation as Jr, eventName as Jt, serverEntryExists as K, saveJob as Kn, explanationOf as Kr, createPrState as Kt, compactQueue as L, checkCodeFreshness as Ln, doneVerb as Lr, QUOTE_LIMITS as Lt, extractPullRequestLinks as M, removeWorktree as Mn, queueAdd as Mr, repositoryKey as Mt, parsePullRequestLink as N, requestJobStop as Nn, queueAddMany as Nr, storeArchitecture as Nt, reportRun as O, createWorktree as On, lifecyclePill as Or, checkArchitecture as Ot, unsupportedReason as P, describeMovement as Pn, queueRemove as Pr, storedScopes as Pt, DEFAULT_WORKSPACE_PORT as Q, clearClaim as Qn, resolveTarget as Qr, markRead as Qt, queueForReview as R, checkReview as Rn, isIssueOutstanding as Rr, REPOSITORY_SCOPE as Rt, isDriving as S, remoteHost as Sn, LIFECYCLES as Sr, waitingHistory as St, beginRun as T, modelHelpLines as Tn, isLifecycle as Tr, RUNTIME_WORDS as Tt, ensureServer as U, fail as Un, processDiscussion as Ur, branchOf as Ut, bootstrapUrl as V, allJobs as Vn, questionsOutstanding as Vr, writeJsonAtomically as Vt, readServerRecord as W, isTerminal as Wn, recommendationsOf as Wr, subjectOf as Wt, workspaceUrl as X, progressSteps as Xn, WITHHELD_REASON as Xr, listNotifications as Xt, stopServer as Y, phasesFor as Yn, describeAuthorship as Yr, headline as Yt, writeServerRecord as Z, purposeOf as Zn, kissOutcome as Zr, markAllRead as Zt, liveUpdateFor as _, extractJsonText as _n, reviewedRepositories as _r, telemetryUploadConsent as _t, wasBlocked as a, clearAgentPreference as ai, withUsageRecorded as an, kissDecisionRefusal as ar, failureOf as at, recordKissRun as b, forgeResolver as bn, workspaceRevision as br, PACKAGE_NAME as bt, RefreshUnavailableError as c, CONFIG_PATH as ci, authorPreview as cn, lineagePosition as cr, reanalyseReview as ct, mergeUpdateInto as d, agentById as di, renderMarkdown as dn, mutateWorkspace as dr, append as dt, isCommitReview as ei, pruneNotifications as en, decideKissRecommendation as er, MIN_WORKSPACE_PORT as et, updateReview as f, agentFor as fi, CODE_BLOCK_CLASS as fn, positionInLineage as fr, readEvents as ft, UpdateAlreadyRunningError as g, asString as gn, reviewedCommit as gr, recordTelemetryUploadConsent as gt, askCheckout as h, AgentUnavailableError as hi, renderInlineText as hn, recentWorkspaces as hr, deliver as ht, liveJobsFor as i, resolveModel as ii, unreadCount as in, indexReviewGroups as ir, writePortPreference as it, batchIdentity as j, readMarker as jn, sectionOf as jr, readArchitecture as jt, stopRun as k, holdWorktree as kn, offersLifecycle as kr, loadArchitecture as kt, cancelUpdate as l, STORE_ROOT as li, hasAuthorContent as ln, loadWorkspace as lr, runJob as lt, isUnchanged as m, AgentCancelledError as mi, escapeHtml as mn, previousKissResult as mr, stamp as mt, deleteReview as n, git as ni, removeNotification as nn, groupByLineage as nr, resolveWorkspacePort as nt, discardJob as o, readAgentPreference as oi, EXPLAIN_SIMPLY_PROMPT as on, kissDecisionsOf as or, prepareReview as ot, refreshPrContext as p, detectAgents as pi, CODE_COPY_CLASS as pn, presentedKissRunOf as pr, record as pt, spawnServer as q, PHASES as qn, isExplanationAction as qr, createPrUpdate as qt, lineageIds as r, resolveAgent as ri, toneOf as rn, indexKeptIds as rr, statedWorkspacePort as rt, FETCH_TIMEOUT_MS as s, writeAgentPreference as si, UNESTABLISHED_MOTIVATION as sn, latestKissRun as sr, settleReview as st, deleteLineage as t, resolveBase as ti, readNotification as tn, followedRefName as tr, parseWorkspacePort as tt, cancellableHere as u, AGENT_IDS as ui, GITHUB_MARKDOWN as un, matchingWorkspaceIds as ur, startJob as ut, runningJobs as v, objectWithAnyOf as vn, summarise as vr, telemetryUploadEnabled as vt, RUN_KINDS as w, fillFileUrlTemplate as wn, activityLabel as wr, recordRuntimeBestEffort as wt, startKissJob as x, forgeFor as xn, workspaceStamps as xr, feedbackUrl as xt, agentEnvOf as y, claimRequestUrl as yn, touchWorkspace as yr, BUILD_VERSION as yt, settleQueue as z, followedBranch as zn, isQuestionOutstanding as zr, ancestorsOf as zt };

package/dist/main.js

@@ -1,5 +1,5 @@
 #!/usr/bin/env node
-import { $ as MAX_WORKSPACE_PORT, A as relativeTime, B as readIndexToken, Dn as MANAGED_ROOT, F as runBatch, Fn as describeSubject, G as reviewsUrl, Gn as loadJob, In as findExistingReview, Jr as WITHHELD_REASON, Mn as removeWorktree, Pn as describeMovement, Q as DEFAULT_WORKSPACE_PORT, Qn as clearClaim, R as queueForReview, Tn as modelHelpLines, U as ensureServer, V as bootstrapUrl, Vn as allJobs, X as workspaceUrl, Xn as progressSteps, Xr as resolveTarget, Y as stopServer, Yr as kissOutcome, Zn as purposeOf, _ as liveUpdateFor, _r as reviewedRepositories, _t as telemetryUploadConsent, a as wasBlocked, ai as writeAgentPreference, b as recordKissRun, bn as forgeResolver, bt as PACKAGE_NAME, ci as AGENT_IDS, ct as reanalyseReview, d as mergeUpdateInto, di as detectAgents, dr as mutateWorkspace, dt as append, ei as git, et as MIN_WORKSPACE_PORT, f as updateReview, fr as positionInLineage, g as UpdateAlreadyRunningError, gt as recordTelemetryUploadConsent, hr as recentWorkspaces, ht as deliver, i as liveJobsFor, ii as readAgentPreference, it as writePortPreference, j as batchIdentity, jn as readMarker, li as agentById, lr as loadWorkspace, mt as stamp, n as deleteReview$1, ni as resolveModel, oi as CONFIG_PATH, ot as prepareReview, pt as record$1, qn as PHASES, rt as statedWorkspacePort, si as STORE_ROOT, st as settleReview, ti as resolveAgent, tr as followedRefName, tt as parseWorkspacePort, ur as matchingWorkspaceIds, ut as startJob, v as runningJobs, vr as summarise, vt as telemetryUploadEnabled, yt as BUILD_VERSION } from "./delete_review-BgIZpcsP.js";
+import { $ as MAX_WORKSPACE_PORT, A as relativeTime, B as readIndexToken, Dn as MANAGED_ROOT, F as runBatch, Fn as describeSubject, G as reviewsUrl, Gn as loadJob, In as findExistingReview, Mn as removeWorktree, Pn as describeMovement, Q as DEFAULT_WORKSPACE_PORT, Qn as clearClaim, Qr as resolveTarget, R as queueForReview, Tn as modelHelpLines, U as ensureServer, V as bootstrapUrl, Vn as allJobs, X as workspaceUrl, Xn as progressSteps, Xr as WITHHELD_REASON, Y as stopServer, Zn as purposeOf, Zr as kissOutcome, _ as liveUpdateFor, _r as reviewedRepositories, _t as telemetryUploadConsent, a as wasBlocked, b as recordKissRun, bn as forgeResolver, bt as PACKAGE_NAME, ci as CONFIG_PATH, ct as reanalyseReview, d as mergeUpdateInto, di as agentById, dr as mutateWorkspace, dt as append, et as MIN_WORKSPACE_PORT, f as updateReview, fr as positionInLineage, g as UpdateAlreadyRunningError, gt as recordTelemetryUploadConsent, hr as recentWorkspaces, ht as deliver, i as liveJobsFor, ii as resolveModel, it as writePortPreference, j as batchIdentity, jn as readMarker, li as STORE_ROOT, lr as loadWorkspace, mt as stamp, n as deleteReview$1, ni as git, oi as readAgentPreference, ot as prepareReview, pi as detectAgents, pt as record$1, qn as PHASES, ri as resolveAgent, rt as statedWorkspacePort, si as writeAgentPreference, st as settleReview, tr as followedRefName, tt as parseWorkspacePort, ui as AGENT_IDS, ur as matchingWorkspaceIds, ut as startJob, v as runningJobs, vr as summarise, vt as telemetryUploadEnabled, yt as BUILD_VERSION } from "./delete_review-B0f6rmy_.js";
 import { basename, dirname, join, resolve } from "node:path";
 import { spawn } from "node:child_process";
 import { existsSync, mkdirSync, readdirSync, realpathSync, rmSync, statSync, writeFileSync } from "node:fs";

package/dist/server.js

@@ -1,4 +1,4 @@
-import { $ as MAX_WORKSPACE_PORT, $n as isClaimed, $r as resolveBase, $t as markReadForReview, A as relativeTime$1, An as isWorktreeHeld, Ar as parseLifecycle, At as partFiles, B as readIndexToken, Bn as reviewedShaOf, Br as issuesOutstanding, Bt as isArchitectureId, C as track, Cn as originRemote, Cr as activityAttributes, Ct as mapRuntimes, D as endRun, Dn as MANAGED_ROOT, Dr as lifecycleItemLabel, Dt as ARCHITECTURE_DIR, E as completeRun, En as ReviewBeingDeletedError, Er as lifecycleChip, Et as describeRuntimes, F as runBatch, Fr as queueStamp, Ft as trackedFiles, Gn as loadJob, Gr as simpleExplanationOf, Gt as createArchitectureNotification, H as clearServerRecordOwnedBy, Hn as endedAt, Hr as reviewerDispositions, Ht as announceJob, I as activateQueued, In as findExistingReview, Ir as readQueue, It as ownersOf, J as stopChild, Jn as isLiveJob, Jr as WITHHELD_REASON, Jt as eventName, K as serverEntryExists, Kn as saveJob, Kr as withSimpleExplanation, Kt as createPrState, L as compactQueue, Ln as checkCodeFreshness, Lr as doneVerb, Lt as QUOTE_LIMITS, M as extractPullRequestLinks, Mn as removeWorktree, Mr as queueAdd, Mt as repositoryKey, N as parsePullRequestLink, Nn as requestJobStop, Nr as queueAddMany, Nt as storeArchitecture, O as reportRun, On as createWorktree, Or as lifecyclePill, Ot as checkArchitecture, P as unsupportedReason, Pr as queueRemove, Pt as storedScopes, Q as DEFAULT_WORKSPACE_PORT, Qr as isCommitReview, Qt as markRead, Rn as checkReview, Rr as isIssueOutstanding, Rt as REPOSITORY_SCOPE, S as isDriving, Sn as remoteHost, Sr as LIFECYCLES, St as waitingHistory, T as beginRun, Tr as isLifecycle, Tt as RUNTIME_WORDS, Un as fail, Ur as processDiscussion, Ut as branchOf, Vn as allJobs, Vr as questionsOutstanding, Vt as writeJsonAtomically, W as readServerRecord, Wn as isTerminal, Wr as recommendationsOf, Wt as subjectOf, Xn as progressSteps, Xr as resolveTarget, Xt as listNotifications, Yn as phasesFor, Yr as kissOutcome, Yt as headline, Z as writeServerRecord, Zn as purposeOf, Zr as DetachedHeadError, Zt as markAllRead, _ as liveUpdateFor, _n as extractJsonText, _t as telemetryUploadConsent, a as wasBlocked, ai as writeAgentPreference, an as withUsageRecorded, ar as kissDecisionRefusal, at as failureOf$1, bn as forgeResolver, br as workspaceRevision, c as RefreshUnavailableError, ci as AGENT_IDS, cn as authorPreview, cr as lineagePosition, ct as reanalyseReview, d as mergeUpdateInto, di as detectAgents, dn as renderMarkdown, dr as mutateWorkspace, ei as git, en as pruneNotifications, er as decideKissRecommendation, et as MIN_WORKSPACE_PORT, f as updateReview, fi as AgentCancelledError, fn as CODE_BLOCK_CLASS, ft as readEvents, g as UpdateAlreadyRunningError, gn as asString, gr as reviewedCommit, gt as recordTelemetryUploadConsent, h as askCheckout, hn as renderInlineText, hr as recentWorkspaces, i as liveJobsFor, ii as readAgentPreference, in as unreadCount$1, ir as indexReviewGroups, it as writePortPreference, j as batchIdentity, jn as readMarker, jr as sectionOf, jt as readArchitecture, k as stopRun, kn as holdWorktree, kr as offersLifecycle, kt as loadArchitecture, l as cancelUpdate, li as agentById, ln as hasAuthorContent, lr as loadWorkspace, lt as runJob, m as isUnchanged, mn as escapeHtml$1, mr as previousKissResult, n as deleteReview, ni as resolveModel, nn as removeNotification, nr as groupByLineage, nt as resolveWorkspacePort, o as discardJob, on as EXPLAIN_SIMPLY_PROMPT, or as kissDecisionsOf, ot as prepareReview, p as refreshPrContext, pi as AgentUnavailableError, pn as CODE_COPY_CLASS, pr as presentedKissRunOf, pt as record, q as spawnServer, qr as describeAuthorship, qt as createPrUpdate, r as lineageIds, ri as clearAgentPreference, rn as toneOf, rr as indexKeptIds, rt as statedWorkspacePort, s as FETCH_TIMEOUT_MS, si as STORE_ROOT, sn as UNESTABLISHED_MOTIVATION, sr as latestKissRun, st as settleReview, t as deleteLineage, ti as resolveAgent, tn as readNotification, tr as followedRefName, tt as parseWorkspacePort, u as cancellableHere, ui as agentFor$1, un as GITHUB_MARKDOWN, vn as objectWithAnyOf, w as RUN_KINDS, wn as fillFileUrlTemplate, wr as activityLabel, wt as recordRuntimeBestEffort, x as startKissJob, xn as forgeFor, xr as workspaceStamps, xt as feedbackUrl, y as agentEnvOf, yn as claimRequestUrl, yr as touchWorkspace, yt as BUILD_VERSION, z as settleQueue, zn as followedBranch, zr as isQuestionOutstanding, zt as ancestorsOf } from "./delete_review-BgIZpcsP.js";
+import { $ as MAX_WORKSPACE_PORT, $n as isClaimed, $r as DetachedHeadError, $t as markReadForReview, A as relativeTime$1, An as isWorktreeHeld, Ar as parseLifecycle, At as partFiles, B as readIndexToken, Bn as reviewedShaOf, Br as issuesOutstanding, Bt as isArchitectureId, C as track, Cn as originRemote, Cr as activityAttributes, Ct as mapRuntimes, D as endRun, Dn as MANAGED_ROOT, Dr as lifecycleItemLabel, Dt as ARCHITECTURE_DIR, E as completeRun, En as ReviewBeingDeletedError, Er as lifecycleChip, Et as describeRuntimes, F as runBatch, Fr as queueStamp, Ft as trackedFiles, Gn as loadJob, Gr as explanationActionOf, Gt as createArchitectureNotification, H as clearServerRecordOwnedBy, Hn as endedAt, Hr as reviewerDispositions, Ht as announceJob, I as activateQueued, In as findExistingReview, Ir as readQueue, It as ownersOf, J as stopChild, Jn as isLiveJob, Jr as withExplanation, Jt as eventName, K as serverEntryExists, Kn as saveJob, Kr as explanationOf, Kt as createPrState, L as compactQueue, Ln as checkCodeFreshness, Lr as doneVerb, Lt as QUOTE_LIMITS, M as extractPullRequestLinks, Mn as removeWorktree, Mr as queueAdd, Mt as repositoryKey, N as parsePullRequestLink, Nn as requestJobStop, Nr as queueAddMany, Nt as storeArchitecture, O as reportRun, On as createWorktree, Or as lifecyclePill, Ot as checkArchitecture, P as unsupportedReason, Pr as queueRemove, Pt as storedScopes, Q as DEFAULT_WORKSPACE_PORT, Qr as resolveTarget, Qt as markRead, Rn as checkReview, Rr as isIssueOutstanding, Rt as REPOSITORY_SCOPE, S as isDriving, Sn as remoteHost, Sr as LIFECYCLES, St as waitingHistory, T as beginRun, Tr as isLifecycle, Tt as RUNTIME_WORDS, Un as fail, Ur as processDiscussion, Ut as branchOf, Vn as allJobs, Vr as questionsOutstanding, Vt as writeJsonAtomically, W as readServerRecord, Wn as isTerminal, Wr as recommendationsOf, Wt as subjectOf, Xn as progressSteps, Xr as WITHHELD_REASON, Xt as listNotifications, Yn as phasesFor, Yr as describeAuthorship, Yt as headline, Z as writeServerRecord, Zn as purposeOf, Zr as kissOutcome, Zt as markAllRead, _ as liveUpdateFor, _n as extractJsonText, _t as telemetryUploadConsent, a as wasBlocked, ai as clearAgentPreference, an as withUsageRecorded, ar as kissDecisionRefusal, at as failureOf$1, bn as forgeResolver, br as workspaceRevision, c as RefreshUnavailableError, cn as authorPreview, cr as lineagePosition, ct as reanalyseReview, d as mergeUpdateInto, di as agentById, dn as renderMarkdown, dr as mutateWorkspace, ei as isCommitReview, en as pruneNotifications, er as decideKissRecommendation, et as MIN_WORKSPACE_PORT, f as updateReview, fi as agentFor$1, fn as CODE_BLOCK_CLASS, ft as readEvents, g as UpdateAlreadyRunningError, gn as asString, gr as reviewedCommit, gt as recordTelemetryUploadConsent, h as askCheckout, hi as AgentUnavailableError, hn as renderInlineText, hr as recentWorkspaces, i as liveJobsFor, ii as resolveModel, in as unreadCount$1, ir as indexReviewGroups, it as writePortPreference, j as batchIdentity, jn as readMarker, jr as sectionOf, jt as readArchitecture, k as stopRun, kn as holdWorktree, kr as offersLifecycle, kt as loadArchitecture, l as cancelUpdate, li as STORE_ROOT, ln as hasAuthorContent, lr as loadWorkspace, lt as runJob, m as isUnchanged, mi as AgentCancelledError, mn as escapeHtml$1, mr as previousKissResult, n as deleteReview, ni as git, nn as removeNotification, nr as groupByLineage, nt as resolveWorkspacePort, o as discardJob, oi as readAgentPreference, on as EXPLAIN_SIMPLY_PROMPT, or as kissDecisionsOf, ot as prepareReview, p as refreshPrContext, pi as detectAgents, pn as CODE_COPY_CLASS, pr as presentedKissRunOf, pt as record, q as spawnServer, qr as isExplanationAction, qt as createPrUpdate, r as lineageIds, ri as resolveAgent, rn as toneOf, rr as indexKeptIds, rt as statedWorkspacePort, s as FETCH_TIMEOUT_MS, si as writeAgentPreference, sn as UNESTABLISHED_MOTIVATION, sr as latestKissRun, st as settleReview, t as deleteLineage, ti as resolveBase, tn as readNotification, tr as followedRefName, tt as parseWorkspacePort, u as cancellableHere, ui as AGENT_IDS, un as GITHUB_MARKDOWN, vn as objectWithAnyOf, w as RUN_KINDS, wn as fillFileUrlTemplate, wr as activityLabel, wt as recordRuntimeBestEffort, x as startKissJob, xn as forgeFor, xr as workspaceStamps, xt as feedbackUrl, y as agentEnvOf, yn as claimRequestUrl, yr as touchWorkspace, yt as BUILD_VERSION, z as settleQueue, zn as followedBranch, zr as isQuestionOutstanding, zt as ancestorsOf } from "./delete_review-B0f6rmy_.js";
 import { basename, isAbsolute, join, relative, resolve } from "node:path";
 import { execFile } from "node:child_process";
 import { existsSync, mkdirSync, readFileSync, readdirSync, realpathSync, rmSync, statSync, writeFileSync } from "node:fs";
@@ -1097,25 +1097,22 @@ function renderAskBuddyMenu(kind) {
   </div>`;
 }
 //#endregion
-//#region ../../packages/review-harness/src/domain/ui/components/simple_explanation.ts
-/**
-* A finding's simple explanation: Buddy AI's plain-English account of it, kept on the card because
-* the reviewer asked for one.
-*
-* For reading the finding, not part of it. It carries no copy source, so Copy and Prepare feedback,
-* which take the card's first one, can only ever find the finding's own.
-*
-* A native disclosure, closed when the page loads and open when it has just been written. Whether a
-* reader left it open is not kept: the explanation is the durable part.
-*/
-var SIMPLE_EXPLANATION_LABEL = "Simple explanation";
+//#region ../../packages/review-harness/src/domain/ui/components/finding_explanation.ts
 var ASK_MORE_LABEL = "Ask more →";
-function renderSimpleExplanation(text, options = {}) {
-	const body = renderMarkdown(text);
+var EXPLANATION_LABELS = {
+	"explain-detail": "Explanation · Technical",
+	explain: "Explanation · Simple"
+};
+/** The disclosure's label: which kind of explanation the card holds, so the other is one choice away. */
+function explanationLabel(explanation) {
+	return EXPLANATION_LABELS[explanationActionOf(explanation)];
+}
+function renderFindingExplanation(explanation, options = {}) {
+	const body = renderMarkdown(explanation.text);
 	const askMore = options.askMore ? `<button type="button" class="pr-helper-simple-explanation-more" data-simple-explanation-more>${ASK_MORE_LABEL}</button>` : "";
 	const content = askMore && body.endsWith("</p>") ? `${body.slice(0, -4)} ${askMore}</p>` : body + askMore;
-	return `<details class="pr-helper-simple-explanation" data-simple-explanation${options.open ? " open" : ""}>
-        <summary class="pr-helper-simple-explanation-summary">${SIMPLE_EXPLANATION_LABEL}</summary>
+	return `<details class="pr-helper-simple-explanation" data-simple-explanation${!Number.isFinite(explanation.writtenAt) ? "" : ` data-explanation-written-at="${explanation.writtenAt}"`}${options.open ? " open" : ""}>
+        <summary class="pr-helper-simple-explanation-summary">${explanationLabel(explanation)}</summary>
         <div class="pr-helper-simple-explanation-body pr-helper-markdown">${content}</div>
       </details>`;
 }
@@ -2608,7 +2605,7 @@ function renderIssuesTab(session, investigatingId, options) {
       </div>
       ${renderStaleDismissal(stale)}
       ${renderSeverityRevision(issue.priorSeverity, issue.severity)}
-      <p class="pr-helper-issue-desc">${renderInlineText(issue.description)}</p>${renderCardExplanation(simpleExplanationOf(session, "issue", issue.id)?.text, options)}
+      <p class="pr-helper-issue-desc">${renderInlineText(issue.description)}</p>${renderCardExplanation(explanationOf(session, "issue", issue.id), options)}
       ${renderReconciliationNote(issue.reconciliationNote)}
       ${renderInvestigationNote(issue.investigationNote)}
       ${renderExistingDiscussion(issue.existingDiscussion)}
@@ -2642,12 +2639,12 @@ function closedFindingsSummary(counts) {
 	return `Dealt with (${counts.done})`;
 }
 /**
-* The finding's simple explanation, under what the finding says and above where it stands. Nothing
-* at all when the reviewer has not asked for one, so such a card is the card it always was.
+* The finding's explanation, under what the finding says and above where it stands. Nothing at all
+* when the reviewer has not asked for one, so such a card is the card it always was.
 */
-function renderCardExplanation(text, options) {
-	if (!text?.trim()) return "";
-	return renderSimpleExplanation(text, { askMore: options?.askBuddy });
+function renderCardExplanation(explanation, options) {
+	if (!explanation?.text.trim()) return "";
+	return renderFindingExplanation(explanation, { askMore: options?.askBuddy });
 }
 /** The reviewer's own decision, said on the card so the closed pile is never one undifferentiated heap. */
 function renderDismissedBadge(isDismissed) {
@@ -2787,7 +2784,7 @@ function renderQuestionsTab(session, investigatingId, options) {
 	const progress = renderQuestionsProgress(result.questions, answered);
 	const records = session.investigations;
 	const now = Date.now();
-	const card = (q) => renderQuestionCard(q, sections, answered.has(q.id), records, now, investigatingId, options, simpleExplanationOf(session, "question", q.id)?.text);
+	const card = (q) => renderQuestionCard(q, sections, answered.has(q.id), records, now, investigatingId, options, explanationOf(session, "question", q.id));
 	const open = result.questions.filter((q) => isQuestionOutstanding(q, answered));
 	const settled = result.questions.filter((q) => !isQuestionOutstanding(q, answered));
 	const closed = settled.length === 0 ? "" : renderClosedFindings(`Answered (${settled.length})`, settled.map(card).join(""), "questions", open.length === 0);
@@ -5435,20 +5432,6 @@ function conversationNote(report, actions) {
 * not exist in any form until the agent has finished reasoning, and drawing their outlines would
 * promise a shape nobody has produced.
 */
-/**
-* How long this has been going, to the second.
-*
-* Seconds rather than the strip's rounded minutes, and that is the whole point of a separate
-* helper: "Running for under a minute" does not change for a full minute, so the one number on
-* screen that could show the run is alive sits perfectly still during the wait most likely to be
-* mistaken for a hang. Past a minute the seconds stop earning their precision and it rounds.
-*/
-function elapsedWords(startedAt, now = Date.now()) {
-	const seconds = Math.max(0, Math.round((now - startedAt) / 1e3));
-	if (seconds < 60) return `${seconds} second${seconds === 1 ? "" : "s"}`;
-	const minutes = Math.round(seconds / 60);
-	return `${minutes} minute${minutes === 1 ? "" : "s"}`;
-}
 /** Whether the change itself is known yet, which is what Files needs and nothing else does. */
 function changeIsRead(job) {
 	return job.completed.includes("context");
@@ -5510,9 +5493,15 @@ function renderProgressView(job, fileCount, cancellable = false, history) {
     ${renderElapsed(job, cancellable ? "Cancel review" : null, history)}`;
 }
 /**
-* How long it has run, with Cancel beside it when something can act on one.
+* The run's footer: how long it has run and how long earlier ones took, with Cancel beside them
+* when something can act on one. One block for a review and an update alike; only the button and
+* the history's words are the operation's own.
+*
+* Running time is the primary line, to the second (`48s`, `1m 08s`), so the one number on screen
+* that shows the run is alive keeps moving through the whole wait. History is secondary, under it,
+* and outside it, because the client rewrites that line's text every second. Cancel sits just
+* after the two rather than at the far edge: an action for this run, not a page control.
 *
-* Beside the line rather than inside it, because the client rewrites that line's text every second.
 * Offered only while a driver holds the job: with nothing running there is nothing to stop, and a
 * button that could not do what it says would be worse than none. Secondary on purpose: it is a way
 * out, not the point. A review has nothing to fall back on, so a cancelled one stops where it was
@@ -5521,16 +5510,34 @@ function renderProgressView(job, fileCount, cancellable = false, history) {
 function renderElapsed(job, cancel, history) {
 	const button = !cancel ? "" : cancel === "Cancel review" ? `<button type="button" class="prb-progress-cancel" data-review-cancel="${escapeHtml$1(job.id)}">Cancel review</button>` : `<button type="button" class="prb-progress-cancel" data-update-cancel="${escapeHtml$1(job.id)}">Cancel update</button>`;
 	return `<div class="prb-progress-foot">
-    <p class="prb-progress-elapsed" data-started-at="${job.startedAt}">Running for ${escapeHtml$1(elapsedWords(job.startedAt))}.</p>
+    <div class="prb-progress-runtime">
+      <p class="prb-progress-elapsed" data-started-at="${job.startedAt}">${escapeHtml$1(runningFor(job.startedAt))}</p>${renderHistory$1(history)}
+    </div>
     ${button}
-  </div>${renderHistory$1(history)}`;
+  </div>`;
+}
+/** The running time as the footer and the client's clock both say it. */
+function runningFor(startedAt, now = Date.now()) {
+	return `Running for ${clockWords(now - startedAt)}`;
 }
 /**
-* How long this took before, under the running time and outside it, because the client rewrites
-* that line every second. Nothing when no earlier run succeeded.
+* To the second at every length: `48s`, `1m 08s`, `1h 00m 01s`. Not `formatDuration`, which drops
+* the seconds past an hour because a finished run's length does not need them; a live count does,
+* or it stands still for a minute at a time on exactly the runs most likely to look hung.
+*/
+function clockWords(ms) {
+	const seconds = Math.max(0, Math.round(ms / 1e3));
+	const pad = (n) => String(n).padStart(2, "0");
+	if (seconds < 60) return `${seconds}s`;
+	if (seconds < 3600) return `${Math.floor(seconds / 60)}m ${pad(seconds % 60)}s`;
+	return `${Math.floor(seconds / 3600)}h ${pad(Math.floor(seconds / 60) % 60)}m ${pad(seconds % 60)}s`;
+}
+/**
+* How long this took before: a status line, so no full stop. Nothing when no earlier run
+* succeeded.
 */
 function renderHistory$1(history) {
-	return history ? `<p class="prb-progress-history" data-runtime-history>${escapeHtml$1(history)}.</p>` : "";
+	return history ? `<p class="prb-progress-history" data-runtime-history>${escapeHtml$1(history)}</p>` : "";
 }
 /**
 * An update running, in place of the review it is updating.
@@ -11373,7 +11380,7 @@ function buildPrompt$2(workspace, question, sectionId, cwd, context = {}) {
 	].join("\n");
 }
 /**
-* The simple explanation a follow-up was asked from, however long ago it was written.
+* The kept explanation a follow-up was asked from, however long ago it was written.
 *
 * Given in full rather than left to the recent turns above, which hold only the last four: a
 * reviewer who comes back a week later and presses Ask more is continuing from the explanation on
@@ -11383,22 +11390,24 @@ function continuingFrom(workspace, continuing) {
 	if (!continuing) return [];
 	return [
 		"---",
-		`THE REVIEWER IS CONTINUING FROM THIS PLAIN-ENGLISH EXPLANATION YOU GAVE EARLIER OF ${findingName(workspace, continuing)}:`,
+		`THE REVIEWER IS CONTINUING FROM THIS ${explanationActionOf(continuing) === "explain" ? "PLAIN-ENGLISH" : "TECHNICAL"} EXPLANATION YOU GAVE EARLIER OF ${findingName(workspace, continuing)}:`,
 		continuing.text
 	];
 }
 /**
-* Explain simply on a finding, whose answer is kept on the card and read again after the code has
-* moved on. So it explains the claim, what the finding says and why it was raised, rather than
-* asserting what the code does now: a later update or investigation says whether the claim still
-* holds, and an explanation written as a present-tense fact about the code would contradict it.
+* Explain (technical) or Explain (simple) on a finding, whose answer is kept on the card and read
+* again after the code has moved on. So either explains the claim, what the finding says and why it
+* was raised, rather than asserting what the code does now: a later update or investigation says
+* whether the claim still holds, and an explanation written as a present-tense fact about the code
+* would contradict it. They differ only in register, as the visible instruction asks.
 */
 function explainingFinding(workspace, explaining) {
 	if (!explaining) return [];
+	const simple = explaining.action === "explain";
 	return [
 		"---",
-		`THE REVIEWER ASKED FOR A SIMPLE EXPLANATION OF ${findingName(workspace, explaining)}. It is kept beside that finding and read again later, possibly after the code has changed.`,
-		"Explain in plain English what the finding is saying and why it was raised: the concern as the review stated it, in terms of what was reviewed (for example \"this was raised because...\"), not as a statement about how the code behaves now. Do not judge whether it still holds; the review tracks that separately."
+		`THE REVIEWER ASKED FOR ${simple ? "A SIMPLE" : "A TECHNICAL"} EXPLANATION OF ${findingName(workspace, explaining)}. It is kept beside that finding and read again later, possibly after the code has changed.`,
+		`Explain ${simple ? "in plain English" : "with the technical detail kept"} what the finding is saying and why it was raised: the concern as the review stated it, in terms of what was reviewed (for example "this was raised because..."), not as a statement about how the code behaves now. Do not judge whether it still holds; the review tracks that separately.`
 	];
 }
 /** `THE ISSUE "<title>"` or `THE QUESTION "<question>"`, or just the kind when the finding is gone. */
@@ -13221,11 +13230,13 @@ var MAX_KEPT_BRANCHES = 50;
 *
 * Only once the analysis is complete. The page is served while a review is still being made, and
 * recording that would file a row of zeroes — no risk, no themes, no findings — against a review
-* that has them, which is worse than recording nothing.
+* that has them, which is worse than recording nothing. `result` is that signal: a running or idle
+* session carries none. `completedAt` is not: it is the Reviewed status, and gating on it kept this
+* event off every first read.
 */
 function workspaceOpenedEvent(workspace) {
 	const result = workspace.session.result;
-	if (workspace.completedAt === void 0 || !result) return null;
+	if (!result) return null;
 	const severities = {
 		critical: 0,
 		major: 0,
@@ -13694,9 +13705,8 @@ async function handleAsk(req, res, workspace) {
 	const question = typeof body?.question === "string" ? body.question.trim() : "";
 	if (!question) return send(res, 400, { error: "No question given." });
 	const sectionId = typeof body?.sectionId === "string" ? body.sectionId : null;
-	const explains = findingNamed(workspace.session, body?.explains);
-	const continuesFrom = findingNamed(workspace.session, body?.continuesExplanation);
-	const continuing = continuesFrom ? simpleExplanationOf(workspace.session, continuesFrom.kind, continuesFrom.id) : void 0;
+	const explains = explanationRequested(workspace.session, body?.explains);
+	const continuing = explanationContinued(workspace, body?.continuesExplanation);
 	if (liveUpdateFor(workspace.id)) return send(res, 409, {
 		error: "This review is being updated. Ask again when the new review is ready.",
 		updating: true
@@ -13749,18 +13759,22 @@ async function handleAsk(req, res, workspace) {
 		});
 		let explanation;
 		mutateWorkspace(workspace.id, (fresh) => {
+			const askedAt = Date.now();
+			const kept = explains && findingNamed(fresh.session, explains) && answer.trim() ? explains : null;
 			fresh.transcript.push({
 				question,
 				answer,
 				sectionId,
-				askedAt: Date.now()
+				askedAt,
+				...kept ? { explains: kept } : {}
 			});
-			if (explains && findingNamed(fresh.session, explains) && answer.trim()) {
+			if (kept) {
 				explanation = {
-					...explains,
-					text: answer.trim()
+					...kept,
+					text: answer.trim(),
+					writtenAt: askedAt
 				};
-				fresh.session.simpleExplanations = withSimpleExplanation(fresh.session.simpleExplanations, explanation);
+				fresh.session.findingExplanations = withExplanation(fresh.session.findingExplanations, explanation);
 			}
 		});
 		sendIfOpen(res, 200, explanation ? {
@@ -13771,6 +13785,45 @@ async function handleAsk(req, res, workspace) {
 		endRun(run);
 	}
 }
+/**
+* The explanation a follow-up was staged from, named by its finding and `writtenAt`.
+*
+* The card's current one when it is still that explanation. When another has replaced it since, the
+* turn that wrote it, which the transcript keeps, so a question written against one explanation is
+* never sent with another's words. Nothing when neither is there. A follow-up without `writtenAt`
+* (an explanation kept before 0.72.0, or a page from then) continues from the card's current one.
+*/
+function explanationContinued(workspace, value) {
+	const finding = findingNamed(workspace.session, value);
+	if (!finding) return void 0;
+	const current = explanationOf(workspace.session, finding.kind, finding.id);
+	const writtenAt = value.writtenAt;
+	if (writtenAt === void 0 || current?.writtenAt === writtenAt) return current;
+	const turn = workspace.transcript.find((exchange) => exchange.askedAt === writtenAt && exchange.explains?.kind === finding.kind && exchange.explains.id === finding.id);
+	return turn?.explains && turn.answer.trim() ? {
+		...finding,
+		action: turn.explains.action,
+		text: turn.answer.trim(),
+		writtenAt: turn.askedAt
+	} : void 0;
+}
+/**
+* The finding a turn explains and the action explaining it, or null when its answer is not to be kept.
+*
+* A missing action is Explain (simple): the only one a page sent before there were two, and a tab
+* opened then still sends it that way. An action that is named but not an explanation is not
+* guessed at; the turn is answered like any other and nothing is kept.
+*/
+function explanationRequested(session, value) {
+	const finding = findingNamed(session, value);
+	if (!finding) return null;
+	const named = value.action;
+	const action = named === void 0 ? "explain" : named;
+	return isExplanationAction(action) ? {
+		...finding,
+		action
+	} : null;
+}
 /** An issue or question of this review, named by kind and id; null for anything else. */
 function findingNamed(session, value) {
 	const named = value;

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "prreviewbuddy",
-  "version": "0.71.2",
+  "version": "0.73.1",
   "description": "Understand code changes before you approve them. Local branch and pull request reviews with Claude Code, Codex or Gemini CLI.",
   "license": "UNLICENSED",
   "homepage": "https://prreviewbuddy.com",