ccqa 1.20.1 → 1.22.0

package/dist/bin/ccqa.mjs

@@ -4,7 +4,7 @@ import { t as EVIDENCE_DIR_ENV } from "../evidence-constants-C425F7ZG.mjs";
 import { a as formatAgentBrowserUnavailableMessage, i as assertAgentBrowserAvailable, n as spawnAB, o as pathWithAgentBrowserShim, r as AgentBrowserUnavailableError, s as resolveAgentBrowserBin$1, t as sleepSync } from "../spawn-ab-CRIVfWpw.mjs";
 import { createRequire } from "node:module";
 import { Command } from "commander";
-import { accessSync, createWriteStream, existsSync, readFileSync, rmSync } from "node:fs";
+import { accessSync, appendFileSync, createWriteStream, existsSync, readFileSync, rmSync } from "node:fs";
 import { fileURLToPath } from "node:url";
 import { createCipheriv, createDecipheriv, createHash, randomBytes, randomUUID, timingSafeEqual } from "node:crypto";
 import { access, chmod, cp, mkdir, mkdtemp, open, readFile, readdir, rename, rm, stat, unlink, writeFile } from "node:fs/promises";
@@ -771,13 +771,21 @@ function waitExit(child) {
 //#endregion
 //#region src/runtime/live-artifacts.ts
 /**
-* Build a sortable run id from the current wall-clock time. ISO8601 with
-* `:` / `.` replaced so it's filename-safe. Caller is expected to mkdir the
-* directory once and pass `runDir = <baseDir>/<runId>` to the path helpers
-* below.
+* Build a sortable, unique run id. ISO8601 with `:` / `.` replaced so it's
+* filename-safe, timestamp first so run directories still sort by time, and a
+* random suffix because the timestamp alone does not separate two specs.
+*
+* The pool launches specs back-to-back, so at `--concurrency > 1` two of them
+* land in the same millisecond. A spec that puts `${CCQA_RUN_ID}` in the name
+* of something it creates would then share that name with its neighbour, and
+* each would find — and delete — the other's row. Nothing fails; the
+* assertions just read the wrong state.
+*
+* Caller is expected to mkdir the directory once and pass
+* `runDir = <baseDir>/<runId>` to the path helpers below.
 */
 function buildRunId() {
-	return (/* @__PURE__ */ new Date()).toISOString().replace(/[:.]/g, "-");
+	return `${(/* @__PURE__ */ new Date()).toISOString().replace(/[:.]/g, "-")}-${randomUUID().slice(0, 8)}`;
 }
 /**
 * Per-step artifact paths under a run directory. `<runDir>/steps/<stepId>.*`.
@@ -1361,8 +1369,8 @@ async function withCostTally(fn) {
 * The active scope's total so far, or null outside one.
 *
 * Read rather than pushed at the caller because commands end in
-* `process.exit`, which never reaches a `finally`. Each one reports the
-* total itself, at the point it knows it is done.
+* `process.exit`, which never reaches a `finally`; whoever opened the scope
+* reads the total on the way out (see `withCostReporting`).
 *
 * Fields stay `null` when no invocation reported them, so a caller can tell
 * "nothing was billed" from "the SDK didn't say" (mock runs, SDK errors).
@@ -1425,6 +1433,17 @@ function resolveEndpointEnv() {
 	}
 	return endpointEnv;
 }
+/**
+* Drop endpoint variables that are present but empty, so an empty value never
+* reaches the Claude Code process as an override. "Set to nothing" is how a
+* caller that cannot omit the key says "use the default" — a CI job wiring
+* `ANTHROPIC_BASE_URL` from an unset repository variable, most of all.
+*/
+function withoutEmptyEndpointVars(env) {
+	const out = { ...env };
+	for (const key of ENDPOINT_ENV_KEYS) if (out[key] === "") delete out[key];
+	return out;
+}
 let nativeBinaryWarned = false;
 /**
 * Warn once per process when the SDK's per-platform native binary is missing:
@@ -1441,10 +1460,10 @@ async function invokeClaudeStreaming(options, onEvent) {
 	const { prompt, systemPrompt, allowedTools, disableBuiltinTools = false, disableThinking = false, mcpServers, maxTurns, env, model, cwd, onAbAction, onAbActionFailed, silenceBashLog = false, relaxAbConstraints = false } = options;
 	const resolvedModel = resolveModel(model);
 	const hasEndpointEnv = Object.keys(resolveEndpointEnv()).length > 0;
-	const mergedEnv = env || hasEndpointEnv ? {
+	const mergedEnv = env || hasEndpointEnv ? withoutEmptyEndpointVars({
 		...process.env,
 		...env
-	} : void 0;
+	}) : void 0;
 	let lastAbToolUseId = null;
 	const claimAbToolUse = (toolUseId) => {
 		if (toolUseId !== lastAbToolUseId) return false;
@@ -3227,20 +3246,6 @@ const ReportArtifactSchema = z.object({
 	sizeBytes: z.number()
 });
 /**
-* Per-step row for a live-mode run (spec.yaml `mode: live`). Mirrors the
-* structure produced by `src/runtime/live-executor.ts:LiveStepResult` but
-* encoded against the report schema so the HTML renderer can carry both
-* deterministic (`evidence`) and live (`liveRun`) sources of step-boundary
-* screenshots.
-*
-* `beforePng` / `afterPng` are RELATIVE to the report directory, same
-* convention as `ReportEvidenceSchema.pngPath` above. The caller copies the
-* PNG files into `<reportDir>/evidence/<feature>/<spec>/` and computes the
-* relative path with `node:path`'s `relative()`, so the report directory is
-* self-contained: it can be archived and shipped on its own (e.g. a hub
-* push) without also bundling the `.ccqa` runs dir.
-*/
-/**
 * Per-step / per-run cost+usage record, pulled from the SDK's `result` message.
 * Every numeric field is nullable so the report can carry partial telemetry
 * (e.g. when the SDK omits a field, or when a step was skipped).
@@ -3248,7 +3253,7 @@ const ReportArtifactSchema = z.object({
 * `models` is the union of model ids the SDK reported using; usually a
 * single element, but the SDK can fan out across models in some modes.
 */
-const LiveReportCostSchema = z.object({
+const ReportCostSchema = z.object({
 	totalCostUsd: z.number().nullable(),
 	durationApiMs: z.number().nullable(),
 	numTurns: z.number().nullable(),
@@ -3258,6 +3263,20 @@ const LiveReportCostSchema = z.object({
 	outputTokens: z.number().nullable(),
 	models: z.array(z.string())
 });
+/**
+* Per-step row for a live-mode run (spec.yaml `mode: live`). Mirrors the
+* structure produced by `src/runtime/live-executor.ts:LiveStepResult` but
+* encoded against the report schema so the HTML renderer can carry both
+* deterministic (`evidence`) and live (`liveRun`) sources of step-boundary
+* screenshots.
+*
+* `beforePng` / `afterPng` are RELATIVE to the report directory, same
+* convention as `ReportEvidenceSchema.pngPath` above. The caller copies the
+* PNG files into `<reportDir>/evidence/<feature>/<spec>/` and computes the
+* relative path with `node:path`'s `relative()`, so the report directory is
+* self-contained: it can be archived and shipped on its own (e.g. a hub
+* push) without also bundling the `.ccqa` runs dir.
+*/
 const LiveReportStepSchema = z.object({
 	stepId: z.string(),
 	source: z.string(),
@@ -3272,7 +3291,7 @@ const LiveReportStepSchema = z.object({
 	beforePng: z.string().nullable(),
 	afterPng: z.string().nullable(),
 	durationMs: z.number(),
-	cost: LiveReportCostSchema,
+	cost: ReportCostSchema,
 	commands: z.array(z.string()).optional()
 });
 const LiveReportRunSchema = z.object({
@@ -3281,7 +3300,7 @@ const LiveReportRunSchema = z.object({
 	startedAt: z.string(),
 	durationMs: z.number(),
 	steps: z.array(LiveReportStepSchema),
-	cost: LiveReportCostSchema
+	cost: ReportCostSchema
 });
 const ReportSpecResultSchema = z.object({
 	feature: z.string(),
@@ -3351,6 +3370,7 @@ const RunReportDataSchema = z.object({
 	customPromptVersion: z.string().nullable().default(null),
 	triageUserPromptHash: z.string().optional(),
 	deployedSha: z.string().optional(),
+	cost: ReportCostSchema.nullable().default(null),
 	results: z.array(ReportSpecResultSchema)
 });
 /** Shape of the "export labels" download produced by the report's client-side JS. */
@@ -4581,9 +4601,137 @@ const DraftNamingSchema = z.object({
 	reason: z.string().optional()
 });
 //#endregion
+//#region src/runtime/live-cost-format.ts
+/**
+* Compact one-line cost summary. Format:
+*   "$0.1234 · 4 turns · 42 in / 6,511 out · 2.0M cached · sonnet"
+* Returns null only when the invocation reported nothing at all (a mock run,
+* an SDK error, or a command that never called a model).
+*
+* The price is one segment among several, not a precondition. An endpoint the
+* SDK has no pricing table for — any Anthropic-compatible gateway in front of
+* a third-party model — reports usage but no `total_cost_usd`, and dropping
+* the whole line there would hide real consumption behind silence. Tokens come
+* from the API response rather than a price list, so they survive that case
+* and become the signal to read.
+*
+* `compact: false` (default for CLI logs) keeps raw numbers and adds a
+* `model=...` segment. `compact: true` (HTML chip) thousand-separates fresh
+* tokens, abbreviates cache-read with K/M, drops the `model=` prefix.
+*/
+function formatLiveCost(cost, options) {
+	const compact = options.compact;
+	const sep = compact ? " · " : " / ";
+	const parts = [];
+	if (cost.totalCostUsd !== null) parts.push(`$${cost.totalCostUsd.toFixed(4)}`);
+	if (cost.numTurns !== null) parts.push(`${cost.numTurns} turns`);
+	if (cost.inputTokens !== null || cost.outputTokens !== null) {
+		const i = cost.inputTokens ?? 0;
+		const o = cost.outputTokens ?? 0;
+		parts.push(compact ? `${formatNumber(i)} in / ${formatNumber(o)} out` : `${i}+${o} tokens`);
+	}
+	if (cost.cacheReadInputTokens !== null && cost.cacheReadInputTokens > 0) parts.push(compact ? `${formatTokenK(cost.cacheReadInputTokens)} cached` : `${cost.cacheReadInputTokens} cache-read`);
+	if (!compact && cost.models.length > 0) parts.push(`model=${cost.models.join(",")}`);
+	return parts.length > 0 ? parts.join(sep) : null;
+}
+/** Thousand-separated count for token figures. */
+function formatNumber(n) {
+	return n.toLocaleString("en-US");
+}
+/** Compact token count: 9,043,456 → "9.0M", 12000 → "12K", small → plain. */
+function formatTokenK(n) {
+	if (n >= 1e6) return `${(n / 1e6).toFixed(1)}M`;
+	if (n >= 1e4) return `${Math.round(n / 1e3)}K`;
+	if (n >= 1e3) return `${(n / 1e3).toFixed(1)}K`;
+	return n.toLocaleString("en-US");
+}
+//#endregion
+//#region src/cli/cost-line.ts
+/**
+* Report what this command spent on Claude: a human line on stderr, and — when
+* `CCQA_COST_FILE` is set — one JSON line appended to that file.
+*
+* stderr rather than stdout because `audit --report-format json` and
+* `select-specs` put machine-readable output on stdout, and a cost line mixed
+* into it breaks the consumer. Cost is diagnostics about the command, not its
+* output.
+*
+* No-op outside a `withCostTally` scope: a command that never opened one has no
+* number to report, which is not the same as having spent nothing.
+*
+* Reached through `withCostReporting`; exported for its own test.
+*/
+function reportCost(command) {
+	const cost = readCostTally();
+	if (cost === null) return;
+	const summary = formatLiveCost(cost, { compact: false });
+	if (summary) process.stderr.write(`[cost] ${summary}\n`);
+	appendCostRecord(command, cost);
+}
+/**
+* Run a command's action inside a cost tally and report the total once, however
+* the command ends.
+*
+* Neither half covers the other. Commands that end deep inside themselves with
+* `process.exit` never unwind, so only the `exit` listener sees them — and both
+* it and `reportCost` are synchronous, so the report still lands. Commands that
+* return normally are past `withCostTally`'s scope by the time `exit` fires, so
+* only the `finally` can still read the tally.
+*
+* Exactly one of the two runs: the `finally` detaches the listener before
+* reporting, and a `process.exit` terminates before the `finally` is reached.
+*
+* A signal handler only reports if it was REGISTERED inside this scope. Node
+* creates the SIGINT/SIGTERM handle when its first listener is added and binds
+* the async context then, so a handler installed at module load would read an
+* empty tally and drop the whole report on interrupt. `run.ts` installs its
+* teardown handlers inside the callback for that reason.
+*/
+async function withCostReporting(command, fn) {
+	return withCostTally(async () => {
+		const report = () => reportCost(command);
+		process.once("exit", report);
+		try {
+			return await fn();
+		} finally {
+			process.off("exit", report);
+			report();
+		}
+	});
+}
+/**
+* Append one JSONL record to `$CCQA_COST_FILE`. Appended, never truncated: a CI
+* job invokes ccqa several times (select-specs, audit, run) and the point of the
+* file is that one `jq -s 'map(.totalCostUsd)|add'` covers the whole job.
+*
+* Written even when `totalCostUsd` is null — that the command ran and was not
+* billed is itself the answer. Synchronous because every caller is about to
+* `process.exit`, which would drop a pending async write.
+*/
+function appendCostRecord(command, cost) {
+	const path = process.env.CCQA_COST_FILE;
+	if (!path) return;
+	const record = {
+		command,
+		at: (/* @__PURE__ */ new Date()).toISOString(),
+		totalCostUsd: cost.totalCostUsd,
+		numTurns: cost.numTurns,
+		inputTokens: cost.inputTokens,
+		outputTokens: cost.outputTokens,
+		cacheReadInputTokens: cost.cacheReadInputTokens,
+		models: cost.models
+	};
+	try {
+		appendFileSync(path, `${JSON.stringify(record)}\n`, "utf8");
+	} catch {}
+}
+//#endregion
 //#region src/cli/draft.ts
 const CATEGORY_LABEL = DRAFT_CATEGORY_LABEL;
 const draftCommand = addLanguageOption(new Command("draft").argument("[feature/spec]", "Optional spec path (e.g. tasks/create-and-complete). If omitted, Claude proposes one from your intent.").description("Interactively draft and refine a spec.yaml with Claude Code").option("--instruction <text>", "Non-interactive single-shot instruction (skips the interactive loop)").option("-y, --yes", "Apply each generated patch without asking [y/N]", false)).action(withUsageErrors(async (specPath, opts) => {
+	await withCostReporting("draft", () => runDraftCli(specPath, opts));
+}));
+async function runDraftCli(specPath, opts) {
 	await ensureCcqaDir();
 	let featureName;
 	let specName;
@@ -4596,7 +4744,7 @@ const draftCommand = addLanguageOption(new Command("draft").argument("[feature/s
 		prefilledIntent = intent;
 	}
 	await runDraft(featureName, specName, opts, prefilledIntent);
-}));
+}
 async function runDraft(featureName, specName, opts, prefilledIntent) {
 	header("draft", `${featureName}/${specName}`);
 	const ja = useJapanesePrompts(opts.language);
@@ -6611,6 +6759,7 @@ z.object({
 	}),
 	gitHead: z.string().nullable(),
 	promptVersion: z.string(),
+	costUsd: z.number().nullable().optional(),
 	ciRunId: z.string().nullable(),
 	runUrl: z.string().nullable().optional(),
 	reportCreatedAt: z.string(),
@@ -6693,10 +6842,26 @@ const SpecLedgerEntrySchema = z.object({
 	deployedSha: z.string().nullable().optional(),
 	deployedShaAmbiguous: z.boolean().optional()
 });
+/**
+* A red-bucket entry: where the failure happened, plus what it was. The cause
+* is copied off the run report's `analysis` so a reader learns why a spec is
+* red without fetching the report of every red spec.
+*
+* Only the red bucket carries it. A pass has no cause, so putting these on
+* `SpecLedgerEntry` would invite writing them where they cannot exist.
+*
+* Both fields are optional and mean the same thing by their absence — nothing
+* is on record. Failure analysis is opt-in (`--on-fail-explain`), and entries
+* written before this release have neither field.
+*/
+const SpecRedLedgerEntrySchema = SpecLedgerEntrySchema.extend({
+	label: PredictedLabelSchema.optional(),
+	headline: z.string().optional()
+});
 z.object({
 	green: z.record(z.string(), SpecLedgerEntrySchema).default({}),
 	run: z.record(z.string(), SpecLedgerEntrySchema).default({}),
-	red: z.record(z.string(), SpecLedgerEntrySchema).default({})
+	red: z.record(z.string(), SpecRedLedgerEntrySchema).default({})
 });
 /** One deploy, as the consumer's deploy job reported it (ADR-0010). */
 const DeployEntrySchema = z.object({
@@ -6876,7 +7041,7 @@ const SpecRerunSchema = z.object({
 	heldBy: SpecLockSchema.nullable(),
 	lastRun: SpecLedgerEntrySchema.nullable(),
 	lastGreen: SpecLedgerEntrySchema.nullable(),
-	lastRed: SpecLedgerEntrySchema.nullable(),
+	lastRed: SpecRedLedgerEntrySchema.nullable(),
 	touchedBy: z.array(z.string()).optional(),
 	touchedByDeploy: DeployRefSchema.nullable().optional()
 });
@@ -6926,7 +7091,7 @@ z.object({
 z.object({
 	entries: z.record(z.string(), SpecLedgerEntrySchema),
 	lastRun: z.record(z.string(), SpecLedgerEntrySchema).default({}),
-	lastRed: z.record(z.string(), SpecLedgerEntrySchema).default({})
+	lastRed: z.record(z.string(), SpecRedLedgerEntrySchema).default({})
 });
 /**
 * One spec's last drift audit, as recorded by `ccqa audit --report-to-hub`. Unlike the
@@ -7136,6 +7301,35 @@ function emitGithubAnnotations(data) {
 	return lines;
 }
 //#endregion
+//#region src/claude/to-report-cost.ts
+/**
+* Narrow an invocation's cost to the shape the report carries.
+*
+* The two differ by exactly one field — the report keeps only the API-time
+* figure, not `durationMs` — so this is a rest-spread rather than a
+* field-by-field copy: a field added to `ClaudeInvocationCost` and to
+* `ReportCost` then flows through without an edit here, and one added to only
+* the former fails to compile. Written out by hand, the same growth would
+* silently drop the new field, and per-step and run-level cost would stop
+* describing the same thing.
+*/
+function toReportCost(cost) {
+	const { durationMs: _durationMs, ...rest } = cost;
+	return rest;
+}
+//#endregion
+//#region src/report/run-cost.ts
+/**
+* What the running command has spent on Claude so far, in the report's shape.
+*
+* Null outside a `withCostTally` scope — a library caller of `executeRun`, or a
+* unit test — which is not the same as "spent nothing".
+*/
+function currentReportCost() {
+	const cost = readCostTally();
+	return cost === null ? null : toReportCost(cost);
+}
+//#endregion
 //#region src/run/spec-catalog.ts
 async function readSpecs(refs, cwd) {
 	const entries = await Promise.all(refs.map(async (ref) => {
@@ -8681,6 +8875,9 @@ const promptRm = new Command("rm").description("Delete a prompt from the hub.").
 }));
 const promptCommand = new Command("prompt").description("Manage prompt assets (per-flow user/agent guidance, triage/audit user guidance, learned calibration prompts) stored on the hub (fetched automatically by `ccqa run` / `ccqa audit` at run time).").addCommand(promptPush).addCommand(promptLs).addCommand(promptRm);
 const deployRecord = new Command("record").description("Tell the hub what a deploy shipped, so it can answer which specs need a re-run (`ccqa run --only-hub-rerun-needed`). Run this from the deploy job, after the deploy succeeds. The changed paths are computed locally with a two-dot `git diff <previous> <sha>`; a job that has only curl and git can POST the same body directly (see docs/hub.md).").requiredOption("--profile <name>", "Environment this deploy landed in (e.g. 'stg'). Required: dev and stg sit at different commits, so the deploy log is per-profile.").requiredOption("--sha <sha>", "Commit that was deployed.").option("--previous <sha>", "Commit this deploy replaced. Omit it and the hub's current log head is used — the normal case, recording no discontinuity. Pass a sha that differs from the head and the hub records one (gapBefore) in the chain: use this for a first record with a real baseline, or to re-anchor a head that no longer matches reality. With no head and nothing passed, there's nothing to diff against: changedPaths is unset and the spec selection is skipped.").option("--ref <ref>", "Ref that was deployed (branch or tag). Recorded for display only.").option("--no-select-specs", "Record the deploy without deciding which specs it reaches. The entry then becomes a hole in the range — every spec behind it is assumed reached rather than being cleared, and nothing can fill it in later, since the hub has no checkout to diff. Only pass this when no Claude credential is available; it costs one model call to leave the range clearable.").option("-m, --model <name>", "Model for the spec selection. Claude alias ('sonnet'|'opus'|'haiku') or full ID. Overrides CCQA_MODEL.").option(...hubUrlOption).option(...hubTokenOption).option("--project <name>", "Project whose deploy log this entry joins. Defaults to the current directory's name.").option("--cwd <path>", "Directory the git diff and the default --project name are resolved against.").action(withHubErrors(async (opts) => {
+	await withCostReporting("hub deploy record", () => runDeployRecord(opts));
+}));
+async function runDeployRecord(opts) {
 	const cwd = resolveCwd(opts.cwd);
 	const project = resolveProject(opts);
 	const hub = connect(opts);
@@ -8709,7 +8906,7 @@ const deployRecord = new Command("record").description("Tell the hub what a depl
 		process.exit(1);
 	}
 	info(`recorded deploy #${entry.index}`);
-}));
+}
 /**
 * Decide which specs this deploy reaches, in the shape the hub stores.
 *
@@ -8804,67 +9001,6 @@ function isStorageStateShape(state) {
 	return typeof state === "object" && state !== null && Array.isArray(state.cookies) && Array.isArray(state.origins);
 }
 //#endregion
-//#region src/runtime/live-cost-format.ts
-/**
-* Compact one-line cost summary. Format:
-*   "$0.1234 · 4 turns · 42 in / 6,511 out · 2.0M cached · sonnet"
-* Returns null when no cost data is available (mock runs / SDK errors).
-*
-* `compact: false` (default for CLI logs) keeps raw numbers and adds a
-* `model=...` segment. `compact: true` (HTML chip) thousand-separates fresh
-* tokens, abbreviates cache-read with K/M, drops the `model=` prefix.
-*/
-function formatLiveCost(cost, options) {
-	if (cost.totalCostUsd === null) return null;
-	const compact = options.compact;
-	const sep = compact ? " · " : " / ";
-	const parts = [`$${cost.totalCostUsd.toFixed(4)}`];
-	if (cost.numTurns !== null) parts.push(`${cost.numTurns} turns`);
-	if (cost.inputTokens !== null || cost.outputTokens !== null) {
-		const i = cost.inputTokens ?? 0;
-		const o = cost.outputTokens ?? 0;
-		parts.push(compact ? `${formatNumber(i)} in / ${formatNumber(o)} out` : `${i}+${o} tokens`);
-	}
-	if (cost.cacheReadInputTokens !== null && cost.cacheReadInputTokens > 0) parts.push(compact ? `${formatTokenK(cost.cacheReadInputTokens)} cached` : `${cost.cacheReadInputTokens} cache-read`);
-	if (!compact && cost.models.length > 0) parts.push(`model=${cost.models.join(",")}`);
-	return parts.join(sep);
-}
-/**
-* Sum of per-spec costs for a batch. Used only by the CLI batch summary.
-* Returns null when no spec has cost data.
-*/
-function formatLiveBatchCost(costs) {
-	let totalUsd = 0;
-	let seen = false;
-	let totalIn = 0;
-	let totalOut = 0;
-	let totalCacheRead = 0;
-	for (const c of costs) {
-		if (c.totalCostUsd !== null) {
-			totalUsd += c.totalCostUsd;
-			seen = true;
-		}
-		totalIn += c.inputTokens ?? 0;
-		totalOut += c.outputTokens ?? 0;
-		totalCacheRead += c.cacheReadInputTokens ?? 0;
-	}
-	if (!seen) return null;
-	const parts = [`$${totalUsd.toFixed(4)}`, `${totalIn}+${totalOut} tokens`];
-	if (totalCacheRead > 0) parts.push(`${totalCacheRead} cache-read`);
-	return parts.join(" / ");
-}
-/** Thousand-separated count for token figures. */
-function formatNumber(n) {
-	return n.toLocaleString("en-US");
-}
-/** Compact token count: 9,043,456 → "9.0M", 12000 → "12K", small → plain. */
-function formatTokenK(n) {
-	if (n >= 1e6) return `${(n / 1e6).toFixed(1)}M`;
-	if (n >= 1e4) return `${Math.round(n / 1e3)}K`;
-	if (n >= 1e3) return `${(n / 1e3).toFixed(1)}K`;
-	return n.toLocaleString("en-US");
-}
-//#endregion
 //#region src/claude/agent-browser-invoke.ts
 function agentBrowserInvokeBase(input) {
 	return {
@@ -8883,14 +9019,9 @@ function agentBrowserInvokeBase(input) {
 }
 //#endregion
 //#region src/prompts/live.ts
-/**
-* Unique agent-browser session name. The runId is millisecond-precision wall
-* clock, so under `--concurrency > 1` two specs can start in the same
-* millisecond and collide; a random suffix guarantees each spec gets its own
-* Chrome session and state never bleeds across parallel runs.
-*/
+/** Unique agent-browser session name, so parallel specs never share a Chrome. */
 function generateLiveSessionName() {
-	return `ccqa-live-${buildRunId()}-${randomUUID().slice(0, 8)}`;
+	return `ccqa-live-${buildRunId()}`;
 }
 /**
 * Static prefix of the `ccqa run` (live spec) system prompt. Built once per
@@ -9179,16 +9310,7 @@ async function runLiveExecutor(input) {
 			});
 			isError = result.isError;
 			errorDetail = result.errorDetail;
-			cost = {
-				totalCostUsd: result.cost.totalCostUsd,
-				durationApiMs: result.cost.durationApiMs,
-				numTurns: result.cost.numTurns,
-				inputTokens: result.cost.inputTokens,
-				cacheCreationInputTokens: result.cost.cacheCreationInputTokens,
-				cacheReadInputTokens: result.cost.cacheReadInputTokens,
-				outputTokens: result.cost.outputTokens,
-				models: result.cost.models
-			};
+			cost = toReportCost(result.cost);
 		} catch (err) {
 			isError = true;
 			errorDetail = err instanceof Error ? err.message : String(err);
@@ -9576,7 +9698,6 @@ async function runLiveSpecs(specs, opts) {
 	const failedCount = runs.filter((r) => r.kind === "error" || r.kind === "run" && r.result.status === "failed").length;
 	blank();
 	meta("live-summary", `${runs.length - failedCount} passed / ${failedCount} failed`);
-	logBatchCost(runs);
 	return {
 		failedCount,
 		reportResults: built.flatMap((b) => b.row ? [b.row] : [])
@@ -9779,10 +9900,6 @@ async function runOneSpec(args) {
 		await closeSession(sessionName);
 	}
 }
-function logBatchCost(runs) {
-	const line = formatLiveBatchCost(runs.flatMap((r) => r.kind === "run" ? [r.result.cost] : []));
-	if (line) meta("total-cost", line);
-}
 /**
 * Classify one failed live run via `analyzeFailure` — same prompt as the
 * deterministic path (Issue #47), fed the live transcript instead of the
@@ -11660,13 +11777,14 @@ async function runExternalSpecs(dispatch, ctx) {
 }
 //#endregion
 //#region src/run/incremental-report.ts
-function createIncrementalReport(reportDir, envelope, sink) {
+function createIncrementalReport(reportDir, envelope, sink, costNow) {
 	const byKey = /* @__PURE__ */ new Map();
 	const reportPath = join(reportDir, "report.json");
 	let queue = Promise.resolve();
 	const key = (r) => `${r.feature}/${r.spec}`;
 	const buildData = () => ({
 		...envelope,
+		...costNow ? { cost: costNow() } : {},
 		results: [...byKey.values()]
 	});
 	const doFlush = async () => {
@@ -12414,7 +12532,8 @@ async function executeRun(targets, opts) {
 				const evidence = await readRowFilesBase64(row, reportDir);
 				await hubCtx.hub.patchRun(runId, {
 					rows: [row],
-					evidence
+					evidence,
+					reportMeta: { cost: currentReportCost() }
 				});
 				hubPatchEverSucceeded = true;
 			} catch (err) {
@@ -12433,7 +12552,7 @@ async function executeRun(targets, opts) {
 		triageUserPromptHash,
 		deployedSha,
 		opts
-	}), hubSink);
+	}), hubSink, currentReportCost);
 	let completedNormally = false;
 	opts.teardown?.onFinalize(async () => {
 		if (completedNormally) return;
@@ -12442,7 +12561,8 @@ async function executeRun(targets, opts) {
 			await hubCtx.hub.patchRun(hubRunId, {
 				rows: incrementalReport.rows(),
 				done: true,
-				finalStatus: "failed"
+				finalStatus: "failed",
+				reportMeta: { cost: currentReportCost() }
 			});
 		} catch (err) {
 			warn(`hub: could not finalize interrupted run ${hubRunId}: ${errMessage(err)}`);
@@ -12803,7 +12923,8 @@ function buildReportEnvelope(args) {
 		promptVersion: "13",
 		customPromptVersion,
 		...triageUserPromptHash !== null ? { triageUserPromptHash } : {},
-		...deployedSha !== null ? { deployedSha } : {}
+		...deployedSha !== null ? { deployedSha } : {},
+		cost: currentReportCost()
 	};
 }
 /** Write the unified JSON (+ optional GitHub-annotation) report for one run. Returns the report data. */
@@ -13100,23 +13221,24 @@ function headerTarget(targets, opts) {
 async function runCliAction(targets, opts) {
 	header("run", headerTarget(targets, opts));
 	const cwd = resolveCwd(opts.cwd);
-	const teardown = createRunTeardown();
-	const disposeSignalHandlers = installTeardownSignalHandlers(teardown);
-	let exitCode;
-	try {
-		exitCode = (await executeRun(targets, {
-			...opts,
-			cwd,
-			teardown
-		})).exitCode;
-	} catch (err) {
-		if (!(err instanceof RunUsageError)) throw err;
-		error(err.message);
-		exitCode = err.exitCode;
-	} finally {
-		await teardown.run();
-		disposeSignalHandlers();
-	}
+	const exitCode = await withCostReporting("run", async () => {
+		const teardown = createRunTeardown();
+		const disposeSignalHandlers = installTeardownSignalHandlers(teardown);
+		try {
+			return (await executeRun(targets, {
+				...opts,
+				cwd,
+				teardown
+			})).exitCode;
+		} catch (err) {
+			if (!(err instanceof RunUsageError)) throw err;
+			error(err.message);
+			return err.exitCode;
+		} finally {
+			await teardown.run();
+			disposeSignalHandlers();
+		}
+	});
 	process.exit(exitCode);
 }
 //#endregion
@@ -14688,6 +14810,9 @@ async function confirmOverwrite(path) {
 	}
 }
 const generateCommand = addHubOptions(addProfileOption(addLanguageOption(new Command("generate").argument("<feature/spec>", "Spec id in '<feature>/<spec>' form (resolves to .ccqa/features/<feature>/test-cases/<spec>/)").description("Generate test code from a spec via its target plugin. Recording-backed targets compile the existing ir.json (run `ccqa record` first); spec-input targets generate directly from the spec.").optionsGroup("How to generate:").option("-m, --model <name>", "Claude model alias ('sonnet'|'opus'|'haiku') or full ID. Overrides CCQA_MODEL.").option("--target <id>", "Generate through this target instead of the spec's own — e.g. emit a Playwright spec from an agent-browser recording. The spec's `target:` stays the default for `ccqa run`.").option("--auto-fix <mode>", "Auto-fix behaviour during script generation: 'interactive' (default, prompt y/N; declines on non-TTY), 'auto' (apply without prompt, for CI), 'skip' (agent-browser: apply only high-confidence fixes; external targets like playwright/runn: no fix pass at all).", parseAutoFixFlag, "interactive").option("--auto-fix-max-retries <n>", "Maximum number of auto-fix retries", "3").option("--no-session-pin", "Don't pin AGENT_BROWSER_SESSION / capture page snapshots after a failure (debug toggle)").optionsGroup("What to do with the result:").option("--overwrite", "Replace previously generated test code without warning").optionsGroup("Learning:").option("--learn-hub-codegen-prompt", "After generation, ask Claude to refresh the target's \"<target>.agent\" learning prompt on the hub from a summary of the run. LLM-generating targets (playwright, runn) only; requires a hub connection.").optionsGroup("Environment and connection:").option("--cwd <path>", "Working directory containing the .ccqa/ tree (monorepo support). Defaults to the current directory.").option("--project <name>", "Project name for the hub. Defaults to the current directory's name.")))).action(withUsageErrors(async (specPath, opts) => {
+	await withCostReporting("generate", () => runGenerateCli(specPath, opts));
+}));
+async function runGenerateCli(specPath, opts) {
 	const { featureName, specName } = parseSpecPath(specPath);
 	const language = opts.language ?? "auto";
 	const cwd = resolveCwd(opts.cwd);
@@ -14738,21 +14863,6 @@ const generateCommand = addHubOptions(addProfileOption(addLanguageOption(new Com
 		}
 		throw e;
 	}
-}));
-//#endregion
-//#region src/cli/cost-line.ts
-/**
-* Write what this command spent on Claude to stderr.
-*
-* stderr rather than stdout because `drift --format json` and `select-specs`
-* put machine-readable output on stdout, and a cost line mixed into it breaks
-* the consumer. Cost is diagnostics about the command, not its output.
-*/
-function reportCost() {
-	const cost = readCostTally();
-	if (cost === null) return;
-	const summary = formatLiveCost(cost, { compact: false });
-	if (summary) process.stderr.write(`[cost] ${summary}\n`);
 }
 //#endregion
 //#region src/cli/record.ts
@@ -14761,13 +14871,7 @@ const recordCommand = addHubOptions(addProfileOption(addLanguageOption(new Comma
 	if (VALIDATION_MODES.includes(raw)) return raw;
 	throw new Error(`--trace-validation must be one of ${VALIDATION_MODES.join(" | ")}`);
 }, "lenient").option("--auto-fix <mode>", "Auto-fix behaviour during script generation: 'interactive' (default, prompt y/N; declines on non-TTY), 'auto' (apply without prompt, for CI), 'skip' (agent-browser: apply only high-confidence fixes; external targets like playwright/runn: no fix pass at all).", parseAutoFixFlag, "interactive").option("--auto-fix-max-retries <n>", "Maximum number of auto-fix retries", "3").option("--trace-only", "Stop after the trace step; do not generate test code").option("--no-session-pin", "Don't pin AGENT_BROWSER_SESSION / capture page snapshots after a failure (debug toggle)").optionsGroup("What to do with the result:").option("--overwrite", "Replace an existing test.spec.ts without warning").optionsGroup("Learning:").option("--learn-hub-trace-prompt", "After the trace finishes, ask Claude to refresh the \"record.agent\" prompt on the hub from a summary of the run. Requires a hub connection.").optionsGroup("Environment and connection:").option("--cwd <path>", "Working directory containing the .ccqa/ tree (monorepo support). Defaults to the current directory.").option("--project <name>", "Project name for the hub. Defaults to the current directory's name.")))).action(withUsageErrors(async (specPath, opts) => {
-	await withCostTally(async () => {
-		try {
-			await runRecord(specPath, opts);
-		} finally {
-			reportCost();
-		}
-	});
+	await withCostReporting("record", () => runRecord(specPath, opts));
 }));
 async function runRecord(specPath, opts) {
 	const { featureName, specName } = parseSpecPath(specPath);
@@ -15451,6 +15555,7 @@ function driftResultsToReport(results, meta) {
 		promptVersion: meta.promptVersion ?? DRIFT_REPORT_PROMPT_VERSION,
 		customPromptVersion: meta.customPromptVersion ?? null,
 		...meta.triageUserPromptHash ? { triageUserPromptHash: meta.triageUserPromptHash } : {},
+		cost: currentReportCost(),
 		results: specResults
 	};
 }
@@ -15640,7 +15745,7 @@ function selectSpecsNeedingAudit(targets, report) {
 //#region src/cli/audit.ts
 const DEFAULT_CONCURRENCY = 3;
 const auditCommand = addProfileOption(addLanguageOption(new Command("audit").argument("[feature/spec]", "Optional spec id. If omitted, every spec under .ccqa/features/ is checked.").description("Read each spec against the code it describes and report where the two have drifted. Static: no browser is run, so this is the cheap check to put in front of `ccqa run`.").optionsGroup("Which specs to audit:").option("--only-affected-by <ref>", "Only specs `ccqa select-specs` judges reached by the diff against <ref> (e.g. origin/main). In pull_request CI, pass $GITHUB_BASE_REF. Costs one model call; specs it cannot decide are audited rather than skipped.").option("--only-hub-audit-needed", "Only specs the hub says a deploy has landed on since the audit last read them. A spec that was never audited is always included, and one the hub cannot answer for is audited rather than skipped. No git diff involved. Requires a hub connection and --hub-profile.").optionsGroup("How to run it:").option("--concurrency <n>", `Parallel spec checks (default: ${DEFAULT_CONCURRENCY})`).option("-m, --model <name>", "Claude model alias ('sonnet'|'opus'|'haiku') or full ID. Overrides CCQA_MODEL.").optionsGroup("What to do with the results:").option("--report-format <fmt>", "Output format: text | json | github", "text").option("--report-to-hub", "Push the result to a ccqa hub as a run (kind: drift), which is what updates the drift ledger. A spec it finds drifted answers `needsRepair` to `ccqa run --only-hub-rerun-needed`, and is not run until a person repairs it.").option("--exit-on <level>", "Exit non-zero on this severity or higher: warn | error", "error").optionsGroup("Environment and connection:").option("--cwd <path>", "Working directory used as both the .ccqa root and the codebase Claude reads. Useful for monorepos. Defaults to process.cwd().").option("--project <name>", "Logical project name for the pushed run. Defaults to the current directory's name.").option(...hubUrlOption).option(...hubTokenOption).option(...hubHeaderOption))).action(withUsageErrors(async (specPath, opts) => {
-	await withCostTally(() => runAudit(specPath, opts));
+	await withCostReporting("audit", () => runAudit(specPath, opts));
 }));
 async function runAudit(specPath, opts) {
 	const format = parseFormat$1(opts.reportFormat);
@@ -15745,7 +15850,6 @@ async function runAudit(specPath, opts) {
 		baseRef,
 		promptCtx
 	});
-	reportCost();
 	process.exit(determineExitCode(results, threshold));
 }
 /** Longer than the audit's own timeout, so a sweep cannot outlive its claim. */
@@ -15896,7 +16000,6 @@ function asHubReadError(err) {
 	throw new RunUsageError(`could not read from the hub: ${errMessage(err)}`);
 }
 function exitWithNoSpecs(format, reason, message) {
-	reportCost();
 	if (format === "json") process.stdout.write(`${JSON.stringify({
 		specs: [],
 		skipped: reason
@@ -16039,8 +16142,7 @@ For each test case above, write a 1–2 sentence factual \`summary\` of what it
 //#endregion
 //#region src/cli/perspectives.ts
 const perspectivesCommand = addHubOptions(addLanguageOption(new Command("perspectives").description("Generate/update the project's perspectives document on the hub — a factual inventory of existing test coverage (no severity, no gap analysis)").option("--instruction <text>", "Hint to steer how summaries are written").option("-y, --yes", "Apply without asking [y/N]", false).option("--verify", "Check the hub document against the local specs (mechanical fields only) and exit 1 when it is stale. No Claude calls — cheap enough for CI.", false).option("-m, --model <name>", "Claude model alias ('sonnet'|'opus'|'haiku') or full ID").option("--project <name>", "Hub project to store the document under (default: cwd directory name)"))).action(withHubErrors(async (opts) => {
-	if (opts.verify) await runPerspectivesCheck(opts);
-	else await runPerspectives(opts);
+	await withCostReporting("perspectives", () => opts.verify ? runPerspectivesCheck(opts) : runPerspectives(opts));
 }));
 /**
 * `--check`: compare the hub document against a freshly-built local skeleton
@@ -16415,7 +16517,7 @@ function parseSummaries(json) {
 //#region src/cli/select-specs.ts
 const DEFAULT_HEAD = "HEAD";
 const selectSpecsCommand = new Command("select-specs").description("Decide which specs a range of commits reaches. Reads the diff and the spec inventory and returns one verdict per spec: needed | notNeeded | unknown.").requiredOption("--base <ref>", "Commit the range starts at — typically what is currently deployed, or the previous commit on the branch.").option("--head <ref>", `Commit the range ends at (default: ${DEFAULT_HEAD})`).option("--cwd <path>", "Working directory used as both the .ccqa root and the codebase Claude reads. Changes outside it are reported but never attributed to a spec. Defaults to process.cwd().").option("-m, --model <name>", "Claude model alias ('sonnet'|'opus'|'haiku') or full ID. Overrides CCQA_MODEL.").option("--format <fmt>", "Output format: text | json", "text").action(async (opts) => {
-	await withCostTally(() => runSelectSpecs(opts));
+	await withCostReporting("select-specs", () => runSelectSpecs(opts));
 });
 async function runSelectSpecs(opts) {
 	const format = parseFormat(opts.format);
@@ -16463,7 +16565,6 @@ async function runSelectSpecs(opts) {
 		...opts.model ? { model: opts.model } : {}
 	});
 	process.stdout.write(format === "json" ? `${JSON.stringify(report, null, 2)}\n` : renderText(report));
-	reportCost();
 	process.exit(0);
 }
 const VERDICT_ORDER = [
@@ -16670,12 +16771,17 @@ function toLedger(raw) {
 }
 /** Fold `from` into `into` in place, per bucket and key, newest `at` winning. */
 function mergeLedgerInto(into, from) {
-	for (const name of BUCKET_NAMES) for (const [key, entry] of Object.entries(from[name])) {
-		const prev = into[name][key];
-		if (!prev || prev.at <= entry.at) into[name][key] = entry;
-	}
+	mergeBucket(into.green, from.green);
+	mergeBucket(into.run, from.run);
+	mergeBucket(into.red, from.red);
 	return into;
 }
+function mergeBucket(into, from) {
+	for (const [key, entry] of Object.entries(from)) {
+		const prev = into[key];
+		if (!prev || prev.at <= entry.at) into[key] = entry;
+	}
+}
 //#endregion
 //#region src/hub/api/validate.ts
 /**
@@ -16755,6 +16861,7 @@ function createPushRunHandler(config) {
 				},
 				gitHead: report.git.head,
 				promptVersion: report.promptVersion,
+				costUsd: report.cost?.totalCostUsd ?? null,
 				ciRunId: report.runId,
 				runUrl: report.runUrl ?? null,
 				reportCreatedAt: report.createdAt,
@@ -16805,6 +16912,7 @@ function createOpenRunHandler(config) {
 			},
 			gitHead: gitHead || null,
 			promptVersion: "",
+			costUsd: null,
 			ciRunId: ciRunId || null,
 			runUrl: runUrl || null,
 			reportCreatedAt: now,
@@ -16828,7 +16936,8 @@ const PatchRunRequestSchema = z.object({
 		promptVersion: z.string().optional(),
 		customPromptVersion: z.string().nullable().optional(),
 		runUrl: z.string().nullable().optional(),
-		triageUserPromptHash: z.string().optional()
+		triageUserPromptHash: z.string().optional(),
+		cost: ReportCostSchema.nullable().optional()
 	}).partial().optional()
 });
 /** Insert or replace `rows` into `results`, upserting by feature/spec identity. */
@@ -16883,7 +16992,7 @@ async function updateSpecLedger(storage, run, results) {
 		const key = `${row.feature}/${row.spec}`;
 		ledger.run[key] = entry;
 		if (row.status === "passed") ledger.green[key] = entry;
-		else ledger.red[key] = entry;
+		else ledger.red[key] = redEntry(entry, row);
 	}
 	if (Object.keys(ledger.run).length === 0) return;
 	try {
@@ -16893,6 +17002,23 @@ async function updateSpecLedger(storage, run, results) {
 	}
 }
 /**
+* The red bucket's entry: the coordinate every bucket carries, plus what the
+* failure analysis concluded, so a reader of the ledger learns why a spec is
+* red without fetching its report.
+*
+* A field with nothing behind it is left out rather than written empty:
+* `analysis` is null when the run did not ask for one (`--on-fail-explain` is
+* opt-in), and a model that answered with no headline leaves `headline` "".
+*/
+function redEntry(entry, row) {
+	if (!row.analysis) return entry;
+	return {
+		...entry,
+		label: row.analysis.label,
+		...row.analysis.headline ? { headline: row.analysis.headline } : {}
+	};
+}
+/**
 * Advance the drift ledger from a terminal `kind: "drift"` run: each row's
 * `analysis` (the diagnosis `driftResultsToReport` put there) becomes that
 * spec's newest audit entry. No profile — drift asks whether the spec still
@@ -16988,6 +17114,7 @@ function createPatchRunHandler(config) {
 		const { rows, evidence, done, finalStatus, reportMeta } = await readJsonBody(ctx.req, maxPushBytes, PatchRunRequestSchema, "request body");
 		let specs = run.specs;
 		let mergedResults = [];
+		let costUsd = run.costUsd ?? null;
 		await config.storage.artifacts.updateJsonFile(id, "report.json", (current) => {
 			const base = current ?? {
 				schemaVersion: 1,
@@ -17001,7 +17128,8 @@ function createPatchRunHandler(config) {
 				model: null,
 				language: null,
 				promptVersion: "",
-				customPromptVersion: null
+				customPromptVersion: null,
+				cost: null
 			};
 			const envelope = {
 				...base,
@@ -17016,11 +17144,13 @@ function createPatchRunHandler(config) {
 				...reportMeta?.promptVersion !== void 0 ? { promptVersion: reportMeta.promptVersion } : {},
 				...reportMeta?.customPromptVersion !== void 0 ? { customPromptVersion: reportMeta.customPromptVersion } : {},
 				...reportMeta?.runUrl !== void 0 ? { runUrl: reportMeta.runUrl } : {},
-				...reportMeta?.triageUserPromptHash !== void 0 ? { triageUserPromptHash: reportMeta.triageUserPromptHash } : {}
+				...reportMeta?.triageUserPromptHash !== void 0 ? { triageUserPromptHash: reportMeta.triageUserPromptHash } : {},
+				...reportMeta?.cost !== void 0 ? { cost: reportMeta.cost } : {}
 			};
 			const merged = mergeResults(current?.results ?? [], rows);
 			specs = countSpecs(merged);
 			mergedResults = merged;
+			costUsd = envelope.cost?.totalCostUsd ?? null;
 			return {
 				...envelope,
 				results: merged
@@ -17030,11 +17160,15 @@ function createPatchRunHandler(config) {
 		const patch = done ? {
 			status: finalStatus ?? (specs.failed > 0 ? "failed" : "passed"),
 			specs,
+			costUsd,
 			...run.kind === "drift" ? { drift: summarizeDrift(mergedResults) } : {},
 			...reportMeta?.git?.head ? { gitHead: reportMeta.git.head } : {},
 			...reportMeta?.promptVersion ? { promptVersion: reportMeta.promptVersion } : {},
 			...await deployHeadMovedDuringRun(config.storage, run) ? { deployedShaAmbiguous: true } : {}
-		} : { specs };
+		} : {
+			specs,
+			costUsd
+		};
 		const updated = await config.storage.runs.update(id, patch);
 		if (done) {
 			await updateSpecLedger(config.storage, updated, mergedResults);
@@ -18585,6 +18719,7 @@ const HTML_BODY = `
     <section id="view-runs">
       <div class="page-bar">
         <h1 data-i18n="runs.title">Runs</h1>
+        <span class="total" id="runs-total-cost" hidden></span>
         <div class="spacer"></div>
         ${refreshButton("runs-refresh")}
       </div>
@@ -18592,7 +18727,7 @@ const HTML_BODY = `
         <div class="card" id="runs-card">
           <div class="table-wrap">
             <table>
-              <thead><tr><th data-i18n="runs.col.run">Run</th><th data-i18n="runs.col.branch">Branch</th><th data-i18n="meta.profile">Profile</th><th data-i18n="runs.col.status">Status</th><th data-i18n="runs.col.specs">Specs</th><th data-i18n="runs.col.created">Created</th></tr></thead>
+              <thead><tr><th data-i18n="runs.col.run">Run</th><th data-i18n="runs.col.branch">Branch</th><th data-i18n="meta.profile">Profile</th><th data-i18n="runs.col.status">Status</th><th data-i18n="runs.col.specs">Specs</th><th data-i18n="runs.col.cost">Cost</th><th data-i18n="runs.col.created">Created</th></tr></thead>
               <tbody id="runs-tbody"></tbody>
             </table>
           </div>
@@ -18948,6 +19083,7 @@ const CSS = `
   .page-bar .back:hover { color: var(--fg); }
   .page-bar .back svg { width: 15px; height: 15px; }
   .page-bar .filters { display: flex; gap: 8px; margin-left: 8px; }
+  .page-bar .total { font-size: 12px; color: var(--muted); font-variant-numeric: tabular-nums; white-space: nowrap; }
   .page-bar .spacer { flex: 1; }
   .content { padding: 18px 24px 48px; }
 
@@ -19489,11 +19625,12 @@ const CLIENT_JS = `
       "runs.title": "Runs", "runs.empty": "Select a project to see its runs.",
       "runs.none": "No runs yet for this project.", "projects.none": "No projects yet. Create one to get started.", "projects.noneShort": "No projects yet",
       "runs.col.run": "Run", "runs.col.branch": "Branch", "runs.col.status": "Status",
-      "runs.col.specs": "Specs", "runs.col.created": "Created",
+      "runs.col.specs": "Specs", "runs.col.cost": "Cost", "runs.col.created": "Created",
+      "runs.totalCost": "Cost of these {n}:",
       "detail.back": "Runs", "detail.specs": "Specs",
       "detail.download": "Download artifacts",
       "detail.triage": "Triage",
-      "meta.branch": "Branch", "meta.specs": "Specs",
+      "meta.branch": "Branch", "meta.specs": "Specs", "meta.cost": "Cost",
       "meta.created": "Created", "meta.passed": "passed", "meta.profile": "Profile",
       "meta.drift": "Drift",
       "diag.cause": "Cause", "diag.fix": "Fix",
@@ -19646,11 +19783,12 @@ const CLIENT_JS = `
       "runs.title": "実行", "runs.empty": "プロジェクトを選択すると実行一覧が表示されます。",
       "runs.none": "このプロジェクトにはまだ実行がありません。", "projects.none": "まだプロジェクトがありません。作成して始めましょう。", "projects.noneShort": "プロジェクトなし",
       "runs.col.run": "実行", "runs.col.branch": "ブランチ", "runs.col.status": "ステータス",
-      "runs.col.specs": "スペック", "runs.col.created": "作成",
+      "runs.col.specs": "スペック", "runs.col.cost": "コスト", "runs.col.created": "作成",
+      "runs.totalCost": "この {n} 件のコスト:",
       "detail.back": "実行", "detail.specs": "スペック",
       "detail.download": "アーティファクトをダウンロード",
       "detail.triage": "トリアージ",
-      "meta.branch": "ブランチ", "meta.specs": "スペック",
+      "meta.branch": "ブランチ", "meta.specs": "スペック", "meta.cost": "コスト",
       "meta.created": "作成", "meta.passed": "合格", "meta.profile": "プロファイル",
       "meta.drift": "ドリフト",
       "diag.cause": "原因", "diag.fix": "対処",
@@ -20042,6 +20180,11 @@ const CLIENT_JS = `
     return run.kind === "drift" ? driftFoundBadge(driftRunState(run), "drift.run.") : statusBadge(run.status);
   }
 
+  // A run's Claude spend, in the same $x.xxxx form as the per-step badge.
+  // A run that billed nothing, and one stored before costs were recorded, both
+  // arrive as a non-number — printing $0.0000 would claim a measured zero.
+  function costText(usd) { return typeof usd === "number" ? "$" + usd.toFixed(4) : "—"; }
+
   // One chip per non-zero drift label, worded via labelText — the same
   // vocabulary the diagnosis card uses — so a run's summary never disagrees
   // with its own spec cards. A zero-count label is omitted, same reasoning as
@@ -20260,9 +20403,38 @@ const CLIENT_JS = `
 
   // ── runs list ────────────────────────────────────────────────────────
 
+  // What the listed runs cost together — the accumulating number an operator
+  // reads to decide how often CI should run, so it follows whatever filter
+  // produced the list. Hidden when no listed run carries a cost at all, since
+  // a "$0.0000" total would read as "CI is free" rather than "nothing measured".
+  //
+  // The label names the run count on purpose. The list is capped (limit=50), so
+  // an unqualified "total" would quietly under-report a project's spend the
+  // moment it has more runs than that — the one number this feature exists to
+  // get right.
+  function renderRunsTotalCost(runs) {
+    var span = document.getElementById("runs-total-cost");
+    // null until the first measured run, which is what distinguishes "nothing
+    // was measured" from "the measured total happens to be zero".
+    var total = null;
+    // The runs that actually contributed, not the runs on screen. Runs stored
+    // before costs were recorded carry nothing, so counting the list would
+    // credit the sum to rows that gave it nothing — the exact misreading the
+    // count is here to prevent.
+    var measured = 0;
+    runs.forEach(function (r) {
+      if (typeof r.costUsd === "number") { total = (total || 0) + r.costUsd; measured++; }
+    });
+    span.hidden = total === null;
+    if (total !== null) {
+      span.textContent = t("runs.totalCost").replace("{n}", measured) + " " + costText(total);
+    }
+  }
+
   function renderRunsList(runs) {
     var tbody = document.getElementById("runs-tbody");
     clear(tbody);
+    renderRunsTotalCost(runs);
     var empty = document.getElementById("runs-empty");
     if (runs.length === 0) {
       empty.hidden = false;
@@ -20325,6 +20497,7 @@ const CLIENT_JS = `
       specsCell.appendChild(specsWrap);
       tr.appendChild(specsCell);
 
+      tr.appendChild(el("td", "muted num", costText(r.costUsd)));
       tr.appendChild(el("td", "muted num", relTime(r.createdAt)));
       tbody.appendChild(tr);
     });
@@ -20337,6 +20510,7 @@ const CLIENT_JS = `
       .then(function (data) { renderRunsList(data.runs); })
       .catch(function (err) {
         clear(document.getElementById("runs-tbody"));
+        renderRunsTotalCost([]);
         empty.hidden = false;
         empty.textContent = "Error loading runs: " + err.message;
       });
@@ -20396,6 +20570,9 @@ const CLIENT_JS = `
     } else {
       metaItem(t("meta.specs"), run.specs.passed + " / " + run.specs.total + " " + t("meta.passed"));
     }
+    // Everything this run spent on Claude — live browsing, triage, the audit a
+    // failure triggers, spec selection — not the sum of the per-step badges.
+    metaItem(t("meta.cost"), costText(run.costUsd));
     metaItem(t("meta.created"), relTime(run.createdAt));
     head.appendChild(meta);
 
@@ -20679,7 +20856,7 @@ const CLIENT_JS = `
     steps.forEach(function (s, i) {
       var built = stepCard(s.status, "#" + (i + 1), s.instruction, s.expected, s.reasoning);
       if (s.cost && s.cost.totalCostUsd != null) {
-        built.head.appendChild(el("span", "cost", "$" + s.cost.totalCostUsd.toFixed(4)));
+        built.head.appendChild(el("span", "cost", costText(s.cost.totalCostUsd)));
       }
       if (s.beforePng || s.afterPng) {
         var frames = el("div", "step-frames");
@@ -22066,6 +22243,17 @@ const CLIENT_JS = `
     badge.appendChild(document.createTextNode(" " + t("perspectives.run.state." + runState)));
     td.appendChild(badge);
 
+    // What the failure was, as the run's analysis called it — the same place
+    // the audit column names its drift label. Absent on a red entry the run
+    // never analyzed, and on entries written before the ledger carried it.
+    if (runState === "failed" && rr.lastRed && rr.lastRed.label) {
+      var cause = el("span", "cellsub", labelText(rr.lastRed.label));
+      // The one-line conclusion, for a pointer. The detail panel shows it in
+      // full, so nothing here depends on finding it.
+      if (rr.lastRed.headline) cause.title = rr.lastRed.headline;
+      td.appendChild(cause);
+    }
+
     // The sub-line is the coordinate of the result being reported — the same
     // "when is this from" evidence the audit column carries above, so the two
     // axes read as siblings. Why the currency matters belongs to the verdict
@@ -22284,6 +22472,21 @@ const CLIENT_JS = `
     return wrap;
   }
 
+  // The failure: which run it was, then what the analysis concluded. The
+  // headline is model output, already localized server-side, so it is shown as
+  // written. A run made without failure analysis carries neither field, and the
+  // row is then the coordinate alone — what it has always been.
+  function rerunFailureValue(entry) {
+    var wrap = el("div");
+    wrap.appendChild(ledgerLine(entry));
+    if (entry.label) {
+      var line = el("div", "d-prose", labelText(entry.label));
+      if (entry.headline) line.appendChild(document.createTextNode(" · " + entry.headline));
+      wrap.appendChild(line);
+    }
+    return wrap;
+  }
+
   // Detail row: a definition list of the case's fields plus the note editor.
   // Built with createElement/textContent throughout — every field here is
   // API-derived, so none of it may go through innerHTML.
@@ -22317,7 +22520,7 @@ const CLIENT_JS = `
     var rr = ledgerEntryFor(perspState.rerun, feature, spec);
     if (rr) {
       row(rerunEvidenceLabelKey(rr), rerunEvidenceValue(rr));
-      if (rerunHasFailure(rr)) row("perspectives.d.lastRed", ledgerLine(rr.lastRed));
+      if (rerunHasFailure(rr)) row("perspectives.d.lastRed", rerunFailureValue(rr.lastRed));
     }
     frag.appendChild(dl);
 

package/dist/hub-client/index.d.mts

@@ -59,6 +59,7 @@ declare const RunSchema: z.ZodObject<{
   }, z.core.$strip>;
   gitHead: z.ZodNullable<z.ZodString>;
   promptVersion: z.ZodString;
+  costUsd: z.ZodOptional<z.ZodNullable<z.ZodNumber>>;
   ciRunId: z.ZodNullable<z.ZodString>;
   runUrl: z.ZodOptional<z.ZodNullable<z.ZodString>>;
   reportCreatedAt: z.ZodString;
@@ -313,6 +314,14 @@ declare const RerunReportSchema: z.ZodObject<{
       at: z.ZodString;
       deployedSha: z.ZodOptional<z.ZodNullable<z.ZodString>>;
       deployedShaAmbiguous: z.ZodOptional<z.ZodBoolean>;
+      label: z.ZodOptional<z.ZodEnum<{
+        TEST_DRIFT: "TEST_DRIFT";
+        SPEC_CHANGE: "SPEC_CHANGE";
+        PRODUCT_BUG: "PRODUCT_BUG";
+        ENVIRONMENT: "ENVIRONMENT";
+        UNKNOWN: "UNKNOWN";
+      }>>;
+      headline: z.ZodOptional<z.ZodString>;
     }, z.core.$strip>>;
     touchedBy: z.ZodOptional<z.ZodArray<z.ZodString>>;
     touchedByDeploy: z.ZodOptional<z.ZodNullable<z.ZodObject<{
@@ -575,6 +584,16 @@ declare const RunReportDataSchema: z.ZodObject<{
   customPromptVersion: z.ZodDefault<z.ZodNullable<z.ZodString>>;
   triageUserPromptHash: z.ZodOptional<z.ZodString>;
   deployedSha: z.ZodOptional<z.ZodString>;
+  cost: z.ZodDefault<z.ZodNullable<z.ZodObject<{
+    totalCostUsd: z.ZodNullable<z.ZodNumber>;
+    durationApiMs: z.ZodNullable<z.ZodNumber>;
+    numTurns: z.ZodNullable<z.ZodNumber>;
+    inputTokens: z.ZodNullable<z.ZodNumber>;
+    cacheCreationInputTokens: z.ZodNullable<z.ZodNumber>;
+    cacheReadInputTokens: z.ZodNullable<z.ZodNumber>;
+    outputTokens: z.ZodNullable<z.ZodNumber>;
+    models: z.ZodArray<z.ZodString>;
+  }, z.core.$strip>>>;
   results: z.ZodArray<z.ZodObject<{
     feature: z.ZodString;
     spec: z.ZodString;

package/dist/package.json

@@ -1,6 +1,6 @@
 {
   "name": "ccqa",
-  "version": "1.20.1",
+  "version": "1.22.0",
   "type": "module",
   "description": "Browser test recorder powered by Claude Code and agent-browser",
   "repository": {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "ccqa",
-  "version": "1.20.1",
+  "version": "1.22.0",
   "type": "module",
   "description": "Browser test recorder powered by Claude Code and agent-browser",
   "repository": {