swarmdo 1.19.0 → 1.29.0

package/v3/@swarmdo/cli/dist/src/swarmvector/model-prices.js

@@ -35,14 +35,24 @@ export const MODEL_PRICES = {
     // Mid tier
     'anthropic/claude-haiku-4.5': { in: 1.00, out: 5.00 },
     'openai/gpt-4.1': { in: 2.00, out: 8.00 },
-    // Strong tier
+    // Strong tier — current-gen Claude (first-party list prices; OpenRouter
+    // mirrors them for anthropic/* slugs). Opus dropped to $5/$25 at 4.5 and
+    // held there through 4.8 — do NOT price these at the legacy $15/$75.
+    'anthropic/claude-sonnet-5': { in: 3.00, out: 15.00 },
     'anthropic/claude-sonnet-4-6': { in: 3.00, out: 15.00 },
+    'anthropic/claude-opus-4-8': { in: 5.00, out: 25.00 },
+    'anthropic/claude-opus-4-5': { in: 5.00, out: 25.00 },
+    'anthropic/claude-fable-5': { in: 10.00, out: 50.00 },
+    // Legacy Opus 4.0 kept — its $15/$75 is correct for THAT model, still a
+    // valid slug; the current opus tier is priced by claude-opus-4-8 above.
     'anthropic/claude-opus-4': { in: 15.00, out: 75.00 },
     // Tier-label fallbacks (when the trajectory only carries a coarse tier
     // and not a concrete modelId — happens before iter 13 wiring landed).
+    // "opus tier" today resolves to Opus 4.8-class ($5/$25); pricing it at the
+    // retired $15/$75 made the cost-optimal router avoid opus 3x too eagerly.
     haiku: { in: 1.00, out: 5.00 },
     sonnet: { in: 3.00, out: 15.00 },
-    opus: { in: 15.00, out: 75.00 },
+    opus: { in: 5.00, out: 25.00 },
     inherit: { in: 3.00, out: 15.00 },
 };
 /**

package/v3/@swarmdo/cli/dist/src/testreport/testreport.d.ts

@@ -28,6 +28,20 @@ export interface TestSummary {
     total: number;
     durationMs: number;
     failures: TestFailure[];
+    /**
+     * TAP `# TODO` tests — "not expected to succeed" per the TAP spec, so a
+     * `not ok … # TODO` is a KNOWN-incomplete stub, counted here as a success,
+     * never a failure. Absent (0) for JUnit and TODO-free TAP.
+     */
+    todo?: number;
+    /**
+     * True if a TAP `Bail out!` line aborted the run. The counts are then
+     * INCOMPLETE — the suite stopped early, so a "0 failed" here does NOT mean
+     * the suite passed. Callers/CI must treat this as a failure.
+     */
+    bailedOut?: boolean;
+    /** Optional reason text following `Bail out!`. */
+    bailReason?: string;
 }
 /** Pull a `file:line` out of a stack-trace / assertion message. Pure. */
 export declare function extractFileLine(text: string): {

package/v3/@swarmdo/cli/dist/src/testreport/testreport.js

@@ -11,16 +11,56 @@
  * No XML dependency — well-formed JUnit escapes <>& in attribute values, so a
  * focused tokenizer over <testcase>/<failure> is safe.
  */
-/** Decode the five predefined XML entities. Pure. */
+/** A numeric character reference → its char, or the original ref if the code
+ * point is out of the valid Unicode range (so a malformed entity never throws
+ * and leaks literally instead of aborting the parse). Pure. */
+function fromCodePointSafe(n, original) {
+    return Number.isInteger(n) && n >= 0 && n <= 0x10ffff ? String.fromCodePoint(n) : original;
+}
+/** Decode the five predefined XML entities plus decimal AND hex numeric
+ * character references (`&#65;`, `&#x1b;`). XML 1.0 allows both numeric forms;
+ * JUnit stack traces with ANSI colouring escape control chars as hex. Pure. */
 function decodeXml(s) {
     return s
         .replace(/&lt;/g, '<')
         .replace(/&gt;/g, '>')
         .replace(/&quot;/g, '"')
         .replace(/&apos;/g, "'")
-        .replace(/&#(\d+);/g, (_, d) => String.fromCharCode(parseInt(d, 10)))
+        .replace(/&#x([0-9a-fA-F]+);/g, (m, h) => fromCodePointSafe(parseInt(h, 16), m))
+        .replace(/&#(\d+);/g, (m, d) => fromCodePointSafe(parseInt(d, 10), m))
         .replace(/&amp;/g, '&'); // last, so &amp;lt; → &lt; not <
 }
+/**
+ * Decode an element's text content, honoring CDATA sections. Text OUTSIDE
+ * `<![CDATA[ … ]]>` is entity-decoded; text INSIDE is taken literally (per the
+ * XML spec, CDATA content is never entity-expanded). Maven Surefire/Failsafe
+ * wrap `<failure>`/`<error>` stack traces in CDATA, so without this the raw
+ * `<![CDATA[`/`]]>` markers leak into the failure message. Pure.
+ */
+function decodeXmlContent(s) {
+    const OPEN = '<![CDATA[';
+    const CLOSE = ']]>';
+    if (!s.includes(OPEN))
+        return decodeXml(s);
+    let out = '';
+    let i = 0;
+    while (i < s.length) {
+        const start = s.indexOf(OPEN, i);
+        if (start < 0) {
+            out += decodeXml(s.slice(i));
+            break;
+        }
+        out += decodeXml(s.slice(i, start));
+        const end = s.indexOf(CLOSE, start + OPEN.length);
+        if (end < 0) {
+            out += s.slice(start + OPEN.length);
+            break;
+        } // unterminated → rest is literal
+        out += s.slice(start + OPEN.length, end);
+        i = end + CLOSE.length;
+    }
+    return out;
+}
 /** Extract key="value" / key='value' attributes from a tag's attribute text. Pure. */
 function parseAttrs(attrText) {
     const out = {};
@@ -62,7 +102,7 @@ export function parseJUnit(xml) {
         if (fail) {
             failed++;
             const fAttrs = parseAttrs(fail[2]);
-            const inner = decodeXml((fail[3] ?? '').trim());
+            const inner = decodeXmlContent(fail[3] ?? '').trim();
             const message = fAttrs.message || inner.split('\n')[0] || undefined;
             // Prefer explicit file/line attrs (pytest, some emitters), else sniff the trace.
             const loc = attrs.file
@@ -82,10 +122,20 @@ export function parseJUnit(xml) {
 /** Parse TAP (Test Anything Protocol) output. Pure. */
 export function parseTAP(text) {
     const failures = [];
-    let passed = 0, failed = 0, skipped = 0;
+    let passed = 0, failed = 0, skipped = 0, todo = 0;
+    let bailedOut = false;
+    let bailReason;
     const lines = text.split('\n');
     for (let i = 0; i < lines.length; i++) {
         const line = lines[i];
+        // `Bail out!` (column 1, per the TAP spec) aborts the run — stop parsing;
+        // anything after it is not a valid result.
+        const bail = line.match(/^Bail out!(?:\s+(.*))?$/);
+        if (bail) {
+            bailedOut = true;
+            bailReason = bail[1]?.trim() || undefined;
+            break;
+        }
         const m = line.match(/^(ok|not ok)\b\s*(\d+)?\s*-?\s*(.*)$/);
         if (!m)
             continue;
@@ -98,6 +148,12 @@ export function parseTAP(text) {
             skipped++;
             continue;
         }
+        // TODO tests are "not expected to succeed" — a `not ok … # TODO` is a known
+        // stub, not a real failure (TAP spec). Count it as todo, never a failure.
+        if (directive && /todo/i.test(directive[1])) {
+            todo++;
+            continue;
+        }
         if (ok) {
             passed++;
         }
@@ -131,7 +187,7 @@ export function parseTAP(text) {
             failures.push({ suite: '', name: desc || '(unnamed)', message, file, line: lineNo });
         }
     }
-    return { passed, failed, skipped, total: passed + failed + skipped, durationMs: 0, failures };
+    return { passed, failed, skipped, total: passed + failed + skipped + todo, durationMs: 0, failures, ...(todo && { todo }), ...(bailedOut && { bailedOut, bailReason }) };
 }
 /** Sniff the format from a path extension, then content. Pure. */
 export function detectFormat(content, path) {
@@ -153,7 +209,7 @@ export function parseTestReport(content, format) {
 }
 /** Merge several summaries (multi-file globs). Pure. */
 export function mergeSummaries(list) {
-    return list.reduce((acc, s) => ({
+    const merged = list.reduce((acc, s) => ({
         passed: acc.passed + s.passed,
         failed: acc.failed + s.failed,
         skipped: acc.skipped + s.skipped,
@@ -161,14 +217,31 @@ export function mergeSummaries(list) {
         durationMs: acc.durationMs + s.durationMs,
         failures: acc.failures.concat(s.failures),
     }), { passed: 0, failed: 0, skipped: 0, total: 0, durationMs: 0, failures: [] });
+    const todoSum = list.reduce((n, s) => n + (s.todo ?? 0), 0);
+    if (todoSum > 0)
+        merged.todo = todoSum;
+    // Any bailed file taints the whole run; keep the first reason seen.
+    const bailed = list.find((s) => s.bailedOut);
+    if (bailed) {
+        merged.bailedOut = true;
+        merged.bailReason = bailed.bailReason;
+    }
+    return merged;
 }
 /** Human-readable digest. Pure. */
 export function formatSummary(s, opts = {}) {
-    const head = `${s.passed} passed · ${s.failed} failed · ${s.skipped} skipped (${s.total} total, ${s.durationMs}ms)`;
-    if (s.failures.length === 0)
+    const todoSeg = s.todo ? ` · ${s.todo} todo` : '';
+    const head = `${s.passed} passed · ${s.failed} failed · ${s.skipped} skipped${todoSeg} (${s.total} total, ${s.durationMs}ms)`;
+    const bailLine = s.bailedOut
+        ? `⚠ suite ABORTED (Bail out!${s.bailReason ? `: ${s.bailReason}` : ''}) — results incomplete`
+        : '';
+    if (s.failures.length === 0) {
+        if (s.bailedOut)
+            return `${bailLine}\n${head}`;
         return head + (s.failed === 0 ? ' ✓' : '');
+    }
     const shown = opts.top && opts.top > 0 ? s.failures.slice(0, opts.top) : s.failures;
-    const lines = [head, ''];
+    const lines = s.bailedOut ? [bailLine, head, ''] : [head, ''];
     for (const f of shown) {
         const where = f.file ? `${f.file}${f.line ? `:${f.line}` : ''}` : '';
         lines.push(`✗ ${f.suite ? f.suite + ' › ' : ''}${f.name}${where ? `  (${where})` : ''}`);

package/v3/@swarmdo/cli/dist/src/transcript/export.d.ts

@@ -45,6 +45,11 @@ export interface RenderOptions {
 export declare function cleanUserText(s: string): string;
 /** Flatten tool_result content (string | array of {text}) to text. */
 export declare function contentToText(content: unknown): string;
+/** Count the conversational turns in a rendered body by matching only the
+ * exact role-heading lines. Anchored to end-of-line so `### Foo` markdown
+ * headings written *inside* a user/assistant message don't inflate the count.
+ * Pure. */
+export declare function countRenderedTurns(markdown: string): number;
 /** Render an ordered list of parsed transcript lines into Markdown. */
 export declare function renderTranscriptMarkdown(lines: RawTranscriptLine[], opts?: RenderOptions): string;
 export interface SessionSummary {

package/v3/@swarmdo/cli/dist/src/transcript/export.js

@@ -92,6 +92,13 @@ function renderLine(role, content, opts) {
         sections.push(results.join('\n'));
     return sections.join('\n\n');
 }
+/** Count the conversational turns in a rendered body by matching only the
+ * exact role-heading lines. Anchored to end-of-line so `### Foo` markdown
+ * headings written *inside* a user/assistant message don't inflate the count.
+ * Pure. */
+export function countRenderedTurns(markdown) {
+    return (markdown.match(/^### (?:🤖 Assistant|👤 User)$/gm) || []).length;
+}
 /** Render an ordered list of parsed transcript lines into Markdown. */
 export function renderTranscriptMarkdown(lines, opts = {}) {
     const resolved = {
@@ -202,7 +209,7 @@ export function exportSession(idOrLatest, opts = {}) {
     const sessionId = sessionIdFromFile(file);
     const project = path.basename(path.dirname(file));
     const body = renderTranscriptMarkdown(lines, opts);
-    const turns = (body.match(/^### /gm) || []).length;
+    const turns = countRenderedTurns(body);
     const header = `# Claude Code session \`${sessionId}\`\n\n- Project: \`${project}\`\n- Turns: ${turns}\n\n---\n\n`;
     return { markdown: header + body, sessionId, project, turns, file };
 }

package/v3/@swarmdo/cli/dist/src/usage/claude-pricing.d.ts

@@ -16,7 +16,10 @@
 export interface TranscriptModelPrice {
     in: number;
     out: number;
+    /** 5-minute-TTL cache write (1.25× base input) */
     cacheWrite: number;
+    /** 1-hour-TTL cache write (2× base input) */
+    cacheWrite1h: number;
     cacheRead: number;
 }
 /**
@@ -26,13 +29,20 @@ export interface TranscriptModelPrice {
  *   'claude-sonnet-4-6@20260115'           → 'claude-sonnet-4-6'  (Vertex)
  */
 export declare function normalizeTranscriptModelId(raw: string): string;
-/** Longest-prefix price lookup; undefined = unpriced (report it, don't guess). */
-export declare function resolveTranscriptPrice(rawModelId: string): TranscriptModelPrice | undefined;
+/**
+ * Longest-prefix price lookup; undefined = unpriced (report it, don't guess).
+ * Pass the transcript entry's timestamp as `atMs` to get date-ranged promo
+ * rates (e.g. Sonnet 5 intro pricing); omitted → sticker price.
+ */
+export declare function resolveTranscriptPrice(rawModelId: string, atMs?: number): TranscriptModelPrice | undefined;
 /** Token counts for one billed API response. */
 export interface TokenBundle {
     inputTokens: number;
     outputTokens: number;
+    /** TOTAL cache-write tokens (5-min + 1-hour). */
     cacheWriteTokens: number;
+    /** The 1-hour-TTL SUBSET of cacheWriteTokens (≤ cacheWriteTokens). Default 0 → all writes priced at the 5-min rate. */
+    cacheWrite1hTokens?: number;
     cacheReadTokens: number;
 }
 /** USD for one response at the given rates. */

package/v3/@swarmdo/cli/dist/src/usage/claude-pricing.js

@@ -16,19 +16,37 @@
  * Price families, matched by LONGEST normalized-id prefix so dated ids like
  * `claude-sonnet-4-6-20260115` resolve without per-snapshot entries.
  *
- * The Claude 5 tier (fable/mythos) is intentionally absent until Anthropic
- * publishes rates — absent means "unpriced", never "guessed".
+ * Opus pricing split (#41): Opus 4.0/4.1 were $15/$75, but Opus 4.5 dropped
+ * the tier to $5/$25 and 4.6/4.7/4.8 kept it. The bare 'claude-opus-4' entry
+ * stays at legacy rates for 4.0/4.1; the longer per-version prefixes win for
+ * 4.5+ via longest-prefix matching.
  */
 const PRICE_FAMILIES = {
-    'claude-opus-4': { in: 15, out: 75, cacheWrite: 18.75, cacheRead: 1.5 },
-    'claude-sonnet-4': { in: 3, out: 15, cacheWrite: 3.75, cacheRead: 0.3 },
-    'claude-haiku-4': { in: 1, out: 5, cacheWrite: 1.25, cacheRead: 0.1 },
-    'claude-3-7-sonnet': { in: 3, out: 15, cacheWrite: 3.75, cacheRead: 0.3 },
-    'claude-3-5-sonnet': { in: 3, out: 15, cacheWrite: 3.75, cacheRead: 0.3 },
-    'claude-3-5-haiku': { in: 0.8, out: 4, cacheWrite: 1, cacheRead: 0.08 },
-    'claude-3-opus': { in: 15, out: 75, cacheWrite: 18.75, cacheRead: 1.5 },
-    'claude-3-haiku': { in: 0.25, out: 1.25, cacheWrite: 0.3, cacheRead: 0.03 },
+    'claude-fable-5': { in: 10, out: 50, cacheWrite: 12.5, cacheWrite1h: 20, cacheRead: 1 },
+    'claude-mythos-5': { in: 10, out: 50, cacheWrite: 12.5, cacheWrite1h: 20, cacheRead: 1 },
+    'claude-opus-4': { in: 15, out: 75, cacheWrite: 18.75, cacheWrite1h: 30, cacheRead: 1.5 },
+    'claude-opus-4-5': { in: 5, out: 25, cacheWrite: 6.25, cacheWrite1h: 10, cacheRead: 0.5 },
+    'claude-opus-4-6': { in: 5, out: 25, cacheWrite: 6.25, cacheWrite1h: 10, cacheRead: 0.5 },
+    'claude-opus-4-7': { in: 5, out: 25, cacheWrite: 6.25, cacheWrite1h: 10, cacheRead: 0.5 },
+    'claude-opus-4-8': { in: 5, out: 25, cacheWrite: 6.25, cacheWrite1h: 10, cacheRead: 0.5 },
+    'claude-sonnet-5': { in: 3, out: 15, cacheWrite: 3.75, cacheWrite1h: 6, cacheRead: 0.3 },
+    'claude-sonnet-4': { in: 3, out: 15, cacheWrite: 3.75, cacheWrite1h: 6, cacheRead: 0.3 },
+    'claude-haiku-4': { in: 1, out: 5, cacheWrite: 1.25, cacheWrite1h: 2, cacheRead: 0.1 },
+    'claude-3-7-sonnet': { in: 3, out: 15, cacheWrite: 3.75, cacheWrite1h: 6, cacheRead: 0.3 },
+    'claude-3-5-sonnet': { in: 3, out: 15, cacheWrite: 3.75, cacheWrite1h: 6, cacheRead: 0.3 },
+    'claude-3-5-haiku': { in: 0.8, out: 4, cacheWrite: 1, cacheWrite1h: 1.6, cacheRead: 0.08 },
+    'claude-3-opus': { in: 15, out: 75, cacheWrite: 18.75, cacheWrite1h: 30, cacheRead: 1.5 },
+    'claude-3-haiku': { in: 0.25, out: 1.25, cacheWrite: 0.3, cacheWrite1h: 0.5, cacheRead: 0.03 },
 };
+const PROMOS = [
+    {
+        // Sonnet 5 introductory pricing: $2/$10 per Mtok through 2026-08-31.
+        prefix: 'claude-sonnet-5',
+        fromMs: Date.UTC(2026, 5, 30), // launched 2026-06-30
+        untilMs: Date.UTC(2026, 7, 31, 23, 59, 59, 999),
+        price: { in: 2, out: 10, cacheWrite: 2.5, cacheWrite1h: 4, cacheRead: 0.2 },
+    },
+];
 /**
  * Reduce a transcript/gateway model id to the bare Anthropic id:
  *   'anthropic/claude-opus-4-8'            → 'claude-opus-4-8'
@@ -47,9 +65,20 @@ export function normalizeTranscriptModelId(raw) {
     id = id.replace(/-v\d+$/, ''); // Bedrock '-v2'
     return id;
 }
-/** Longest-prefix price lookup; undefined = unpriced (report it, don't guess). */
-export function resolveTranscriptPrice(rawModelId) {
+/**
+ * Longest-prefix price lookup; undefined = unpriced (report it, don't guess).
+ * Pass the transcript entry's timestamp as `atMs` to get date-ranged promo
+ * rates (e.g. Sonnet 5 intro pricing); omitted → sticker price.
+ */
+export function resolveTranscriptPrice(rawModelId, atMs) {
     const id = normalizeTranscriptModelId(rawModelId);
+    if (atMs !== undefined) {
+        for (const promo of PROMOS) {
+            if (id.startsWith(promo.prefix) && atMs >= promo.fromMs && atMs <= promo.untilMs) {
+                return promo.price;
+            }
+        }
+    }
     let best;
     let bestLen = 0;
     for (const [prefix, price] of Object.entries(PRICE_FAMILIES)) {
@@ -62,9 +91,12 @@ export function resolveTranscriptPrice(rawModelId) {
 }
 /** USD for one response at the given rates. */
 export function transcriptCostUsd(price, t) {
+    const cacheWrite1h = t.cacheWrite1hTokens ?? 0;
+    const cacheWrite5m = Math.max(0, t.cacheWriteTokens - cacheWrite1h);
     return ((t.inputTokens * price.in +
         t.outputTokens * price.out +
-        t.cacheWriteTokens * price.cacheWrite +
+        cacheWrite5m * price.cacheWrite +
+        cacheWrite1h * price.cacheWrite1h +
         t.cacheReadTokens * price.cacheRead) / 1_000_000);
 }
 //# sourceMappingURL=claude-pricing.js.map

package/v3/@swarmdo/cli/dist/src/usage/transcript-usage.js

@@ -177,15 +177,18 @@ function toUsageEvent(line, project, seen, unpriced) {
         cacheWriteTokens: usage.cache_creation_input_tokens ?? 0,
         cacheReadTokens: usage.cache_read_input_tokens ?? 0,
     };
+    // 1-hour-TTL cache writes cost 2× base input, not the 1.25× 5-min rate. Recent
+    // transcripts split them in `cache_creation`; older ones don't (→ all 5-min).
+    const cacheWrite1hTokens = usage.cache_creation?.ephemeral_1h_input_tokens ?? 0;
     let costUsd;
     let costSource;
-    const price = resolveTranscriptPrice(model);
+    const price = resolveTranscriptPrice(model, ts.getTime());
     if (typeof line.costUSD === 'number') {
         costUsd = line.costUSD;
         costSource = 'transcript';
     }
     else if (price) {
-        costUsd = transcriptCostUsd(price, tokens);
+        costUsd = transcriptCostUsd(price, { ...tokens, cacheWrite1hTokens });
         costSource = 'computed';
     }
     else {

package/v3/@swarmdo/cli/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@swarmdo/cli",
-  "version": "1.19.0",
+  "version": "1.29.0",
   "type": "module",
   "description": "Swarmdo CLI - Enterprise AI agent orchestration with 60+ specialized agents, swarm coordination, MCP server, self-learning hooks, and vector memory for Claude Code",
   "main": "dist/src/index.js",
@@ -83,7 +83,7 @@
     "test:pattern-store": "npx tsx src/transfer/store/tests/standalone-test.ts",
     "postinstall": "node ./scripts/postinstall.cjs",
     "build:standalone": "node scripts/build-standalone.mjs",
-    "prepublishOnly": "node -e \"if (process.env.npm_command === 'publish') { console.error('\\n@swarmdo/cli must be published from the self-contained staging dir (file:../ siblings do not ship from here):\\n\\n  npm run build && npm run build:standalone && npm publish dist-standalone/\\n'); process.exit(1); } // pnpm runs prepublishOnly when packing local directory deps during install \u2014 that path must not fail\""
+    "prepublishOnly": "node -e \"if (process.env.npm_command === 'publish') { console.error('\\n@swarmdo/cli must be published from the self-contained staging dir (file:../ siblings do not ship from here):\\n\\n  npm run build && npm run build:standalone && npm publish dist-standalone/\\n'); process.exit(1); } // pnpm runs prepublishOnly when packing local directory deps during install — that path must not fail\""
   },
   "devDependencies": {
     "@types/better-sqlite3": "^7.6.0",