@tekmidian/pai 0.47.0 → 0.48.0

package/dist/cli/index.mjs

@@ -7,7 +7,7 @@ import "../helpers-crDEr6S2.mjs";
 import "../sync-Btcb6xA4.mjs";
 import "../embeddings-DvRKcQyt.mjs";
 import "../search-B69cQwDB.mjs";
-import { n as drainStdio, t as buildProgram } from "../program-BxpeczCx.mjs";
+import { n as drainStdio, t as buildProgram } from "../program-Bgv1lFFA.mjs";
 import "../context-handover-cache-B9NKSjwR.mjs";
 import "../config-DCu8THKi.mjs";
 import "../config-DxMZjpvY.mjs";

package/dist/cli/program.mjs

@@ -6,7 +6,7 @@ import "../helpers-crDEr6S2.mjs";
 import "../sync-Btcb6xA4.mjs";
 import "../embeddings-DvRKcQyt.mjs";
 import "../search-B69cQwDB.mjs";
-import { t as buildProgram } from "../program-BxpeczCx.mjs";
+import { t as buildProgram } from "../program-Bgv1lFFA.mjs";
 import "../context-handover-cache-B9NKSjwR.mjs";
 import "../config-DCu8THKi.mjs";
 import "../config-DxMZjpvY.mjs";

package/dist/{program-BxpeczCx.mjs → program-Bgv1lFFA.mjs}

@@ -10076,6 +10076,24 @@ function parseCompactionEvent(line, turnIndex) {
 	};
 }
 /**
+* `type:"system"` `subtype:"model_refusal_fallback"` lines mark an
+* automatic model switch fired by an API safety refusal (e.g. "cyber"
+* category), not a user or config choice — worth flagging separately since
+* it can rebuild the whole prompt cache mid-session.
+*/
+function isModelFallback(line) {
+	return line.type === "system" && line.subtype === "model_refusal_fallback";
+}
+function parseFallbackEvent(line, turnIndex) {
+	return {
+		turnIndex,
+		from: line.originalModel ?? "unknown",
+		to: line.fallbackModel ?? "unknown",
+		category: line.apiRefusalCategory ?? "unknown",
+		scope: line.scope ?? "unknown"
+	};
+}
+/**
 * Fold one already-parsed JSONL line into a report being accumulated.
 * Exported separately so both the streaming file reader below and tests
 * (which build fixtures as arrays of objects, not files) share one path.
@@ -10098,13 +10116,24 @@ function foldAssistantLine(report, seenIds, line, threshold) {
 		if (typeof v === "number") report.totals[key] += v;
 	}
 	const context = (Number(usage.cache_read_input_tokens) || 0) + (Number(usage.cache_creation_input_tokens) || 0) + (Number(usage.input_tokens) || 0);
-	if (report.firstTurnContext === null) report.firstTurnContext = context;
+	if (report.firstTurnContext === null) {
+		report.firstTurnContext = context;
+		report.firstTurnAt = line.timestamp ?? null;
+	}
 	report.lastTurnContext = context;
 	report.maxContext = report.maxContext === null ? context : Math.max(report.maxContext, context);
 	if (context > threshold) report.turnsAboveThreshold++;
 	if ((Number(usage.cache_creation_input_tokens) || 0) > 2e4) report.cacheRebuildTurns++;
 	const model = message?.model ?? "unknown";
 	report.models[model] = (report.models[model] ?? 0) + 1;
+	if (report.lastModel !== null && report.lastModel !== model) report.modelSwitches.push({
+		turnIndex: report.turns,
+		from: report.lastModel,
+		to: model,
+		cacheRead: Number(usage.cache_read_input_tokens) || 0,
+		cacheCreation: Number(usage.cache_creation_input_tokens) || 0
+	});
+	report.lastModel = model;
 	const split = usage.cache_creation;
 	if (split) {
 		report.cacheCreationSplit.ephemeral5m += Number(split.ephemeral_5m_input_tokens) || 0;
@@ -10120,6 +10149,8 @@ function foldAssistantLine(report, seenIds, line, threshold) {
 */
 function isRealUserPrompt(line) {
 	if (line.type !== "user") return false;
+	if (line.isMeta) return false;
+	if (line.origin?.kind && line.origin.kind !== "human") return false;
 	const content = line.message?.content;
 	if (typeof content === "string") return true;
 	if (Array.isArray(content)) return !content.some((block) => block?.type === "tool_result");
@@ -10145,7 +10176,11 @@ async function parseSessionUsage(path, threshold = 2e5) {
 		cacheRebuildTurns: 0,
 		userPrompts: 0,
 		promptExposure: 0,
-		compactions: []
+		compactions: [],
+		firstTurnAt: null,
+		lastModel: null,
+		modelSwitches: [],
+		fallbacks: []
 	};
 	const seenIds = /* @__PURE__ */ new Set();
 	let contextSum = 0;
@@ -10170,6 +10205,10 @@ async function parseSessionUsage(path, threshold = 2e5) {
 			if (event) report.compactions.push(event);
 			continue;
 		}
+		if (isModelFallback(obj)) {
+			report.fallbacks.push(parseFallbackEvent(obj, report.turns));
+			continue;
+		}
 		const context = foldAssistantLine(report, seenIds, obj, threshold);
 		if (context !== null) {
 			contextSum += context;
@@ -10239,6 +10278,44 @@ async function auditSession(path, threshold) {
 		percentages
 	};
 }
+/**
+* First-turn context of every session transcript in this project's
+* `~/.claude/projects/<encoded-cwd>` directory, newest first — the readings
+* needed to see when a step change (model switch, hook growth) happened
+* across sessions without an ad-hoc script.
+*/
+async function sessionHistory(cwd, limit = 20) {
+	const dir = join(homedir(), ".claude", "projects", encodeDir(cwd));
+	let entries;
+	try {
+		entries = readdirSync(dir).filter((f) => f.endsWith(".jsonl"));
+	} catch {
+		return [];
+	}
+	const rows = [];
+	for (const entry of entries) {
+		const path = join(dir, entry);
+		const report = await parseSessionUsage(path);
+		if (report.turns === 0) continue;
+		rows.push({
+			path,
+			sessionId: entry.replace(/\.jsonl$/, ""),
+			firstTurnAt: report.firstTurnAt,
+			firstTurnContext: report.firstTurnContext,
+			turns: report.turns,
+			models: report.models,
+			switches: report.modelSwitches.length,
+			fallbacks: report.fallbacks.length
+		});
+	}
+	rows.sort((a, b) => {
+		if (a.firstTurnAt === null && b.firstTurnAt === null) return 0;
+		if (a.firstTurnAt === null) return 1;
+		if (b.firstTurnAt === null) return -1;
+		return b.firstTurnAt.localeCompare(a.firstTurnAt);
+	});
+	return rows.slice(0, limit);
+}
 
 //#endregion
 //#region src/audit/spawn.ts
@@ -11026,6 +11103,7 @@ const FIRST_TURN_CONTEXT_LIMIT = 3e4;
 const DAEMON_FAILURE_RATE_LIMIT = .05;
 const CONTEXT_GROWTH_AVG_RED = 1e5;
 const CONTEXT_GROWTH_MAX_AMBER = 15e4;
+const MODEL_SWITCH_REBUILD_LIMIT = 2e4;
 const SEVERITY_RANK = {
 	RED: 0,
 	AMBER: 1,
@@ -11152,6 +11230,22 @@ function buildFindings(input) {
 			severity: overshoots.length > 0 ? "RED" : "GREEN",
 			evidence: overshoots.length > 0 ? `override not honoured: ${overshoots.map((c) => `preTokens=${c.preTokens}`).join(", ")} vs configured ${input.ctxThreshold}` : compactions.length === 0 ? `0 compactions in ${input.session.turns} turns` : `${compactions.length} compaction(s), all within 25% of configured ${input.ctxThreshold}`
 		});
+		const switches = input.session.modelSwitches;
+		const fallbacks = input.session.fallbacks;
+		const rebuilds = switches.filter((s) => s.cacheCreation > MODEL_SWITCH_REBUILD_LIMIT);
+		const short = (model) => model.replace(/^claude-/, "");
+		let switchEvidence;
+		if (switches.length === 0) switchEvidence = `0 switches in ${input.session.turns} turns, models ${JSON.stringify(input.session.models)}`;
+		else {
+			const list = switches.map((s) => `${short(s.from)}->${short(s.to)} at turn ${s.turnIndex} (cache_read ${s.cacheRead}, cache_creation ${s.cacheCreation})`).join("; ");
+			switchEvidence = `${switches.length} switch(es): ${list}`;
+			if (fallbacks.length > 0) switchEvidence += `; ${fallbacks.length} safeguard fallback(s): ${fallbacks.map((f) => f.category).join(",")}`;
+		}
+		findings.push({
+			finding: "mid-session model switches",
+			severity: switches.length === 0 ? "GREEN" : rebuilds.length > 0 ? "RED" : "AMBER",
+			evidence: switchEvidence
+		});
 	}
 	return sortFindings(findings);
 }
@@ -11701,8 +11795,40 @@ function printSession(report) {
 	console.log(`avg context: ${report.avgContext ?? "n/a"}, max context: ${report.maxContext ?? "n/a"}, turns above threshold: ${report.turnsAboveThreshold}, cache-rebuild turns: ${report.cacheRebuildTurns}, user prompts: ${report.userPrompts}, prompt exposure: ${report.promptExposure}`);
 	console.log(`compactions: ${report.compactions.length}`);
 	for (const c of report.compactions) console.log(dim$1(`  ${c.trigger} at turn ${c.turnIndex}: preTokens=${c.preTokens}`));
+	console.log(`model switches: ${report.modelSwitches.length}`);
+	for (const s of report.modelSwitches) console.log(dim$1(`  ${s.from} -> ${s.to} at turn ${s.turnIndex}: cache_read=${s.cacheRead}, cache_creation=${s.cacheCreation}`));
+	if (report.fallbacks.length > 0) {
+		console.log(`safeguard fallbacks: ${report.fallbacks.length}`);
+		for (const f of report.fallbacks) console.log(dim$1(`  ${f.from} -> ${f.to} before turn ${f.turnIndex}: ${f.category} (${f.scope})`));
+	}
+}
+function printSessionHistory(rows) {
+	console.log(header("Session history (newest first)"));
+	const tableRows = rows.map((r) => [
+		r.firstTurnAt ?? "n/a",
+		r.sessionId,
+		r.firstTurnContext === null ? "n/a" : String(r.firstTurnContext),
+		String(r.turns),
+		String(r.switches),
+		Object.entries(r.models).map(([name, count]) => `${name}:${count}`).join(",")
+	]);
+	console.log(renderTable([
+		"first turn (UTC)",
+		"session",
+		"first-turn ctx",
+		"turns",
+		"switches",
+		"models"
+	], tableRows));
 }
 async function cmdSession(path, opts) {
+	if (opts.history !== void 0) {
+		const limit = typeof opts.history === "string" ? parseInt(opts.history, 10) : 20;
+		const rows = await sessionHistory(process.cwd(), limit);
+		if (opts.json) return printJson(rows);
+		printSessionHistory(rows);
+		return;
+	}
 	const report = await auditSession(path, opts.ctxThreshold ? parseInt(opts.ctxThreshold, 10) : void 0);
 	if (!report) {
 		console.error(err("  No session transcript found."));
@@ -12186,7 +12312,7 @@ function registerAuditCommands(auditCmd) {
 		await cmdFiles(paths, jsonOpt());
 	});
 	tokensCmd.command("hooks").description("Token cost of every SessionStart / UserPromptSubmit hook").action(() => cmdHooks(jsonOpt()));
-	tokensCmd.command("session").description("Cache/input/output token split for a session transcript (default: newest)").argument("[path]", "Session JSONL path (default: newest under ~/.claude/projects)").option("--ctx-threshold <n>", "Per-turn context size above which a turn counts as 'above threshold' (default 200000)").action(async (path, opts) => {
+	tokensCmd.command("session").description("Cache/input/output token split for a session transcript (default: newest)").argument("[path]", "Session JSONL path (default: newest under ~/.claude/projects)").option("--ctx-threshold <n>", "Per-turn context size above which a turn counts as 'above threshold' (default 200000)").option("--history [n]", "List first-turn context of the newest n sessions in this project's transcript directory (default 20)").action(async (path, opts) => {
 		await cmdSession(path, {
 			...jsonOpt(),
 			...opts
@@ -20064,4 +20190,4 @@ claude() {
 
 //#endregion
 export { drainStdio as n, buildProgram as t };
-//# sourceMappingURL=program-BxpeczCx.mjs.map
+//# sourceMappingURL=program-Bgv1lFFA.mjs.map