@post-print/agent-test 0.1.10 → 0.1.13

package/dist/html-report.js

@@ -0,0 +1,462 @@
+import { mkdtemp, writeFile } from "node:fs/promises";
+import { tmpdir } from "node:os";
+import { join } from "node:path";
+function escapeHtml(value) {
+    return value
+        .replaceAll("&", "&")
+        .replaceAll("<", "&lt;")
+        .replaceAll(">", "&gt;")
+        .replaceAll('"', "&quot;")
+        .replaceAll("'", "&#39;");
+}
+function formatDuration(ms) {
+    if (ms < 1000) {
+        return `${Math.round(ms)}ms`;
+    }
+    return `${(ms / 1000).toFixed(1)}s`;
+}
+function statusLabel(result) {
+    if (result.skipped) {
+        return "skipped";
+    }
+    return result.passed ? "passed" : "failed";
+}
+function statusClass(result) {
+    if (result.skipped) {
+        return "status-skipped";
+    }
+    return result.passed ? "status-passed" : "status-failed";
+}
+/** Human-readable label + one-line explanation for the matcher codes assertRubric/judge emit. */
+const MATCHER_LABELS = {
+    runAgent: { label: "Agent run", hint: "The agent session itself failed or was cut short." },
+    must: {
+        label: "Missing requirement",
+        hint: "A required behavior did not show up in the transcript.",
+    },
+    mustNot: { label: "Forbidden behavior", hint: "The agent did something it was told not to do." },
+    mustRun: { label: "Missing command", hint: "An expected command never ran." },
+    mustInvokeSkill: {
+        label: "Missing skill",
+        hint: "The agent didn't read a skill it was expected to use.",
+    },
+    mustNotInvokeSkill: {
+        label: "Unexpected skill",
+        hint: "The agent read a skill it should have avoided.",
+    },
+    routingBlock: { label: "Routing", hint: "The routing announcement didn't match expectations." },
+    workingTreeLeak: {
+        label: "Working tree leak",
+        hint: "The agent's edits leaked outside its isolated worktree.",
+    },
+    recordTrace: { label: "Recording failed", hint: "Saving the trace to disk failed." },
+    judge: { label: "Judge", hint: "The LLM judge flagged this scenario." },
+};
+function humanizeMatcher(matcher) {
+    const [base] = matcher.split(":");
+    const known = MATCHER_LABELS[base ?? matcher];
+    if (known) {
+        if (matcher.includes(":")) {
+            const suffix = matcher.slice(matcher.indexOf(":") + 1);
+            return { label: `${known.label} — ${suffix}`, hint: known.hint };
+        }
+        return known;
+    }
+    return { label: matcher };
+}
+const ROLE_META = {
+    user: { label: "User", side: "right" },
+    assistant: { label: "Agent", side: "left" },
+    system: { label: "System", side: "center" },
+    tool: { label: "Tool", side: "center" },
+};
+function renderMessageBubble(message) {
+    const meta = ROLE_META[message.role] ?? { label: message.role, side: "center" };
+    return `
+<div class="chat-row side-${meta.side}">
+  <div class="bubble role-${escapeHtml(message.role)}">
+    <div class="bubble-label">${escapeHtml(meta.label)}</div>
+    <div class="bubble-text">${escapeHtml(message.content)}</div>
+  </div>
+</div>`;
+}
+/** Truncate long single-line arg values so a stray file body doesn't blow up the card. */
+function formatArgValue(value) {
+    const text = typeof value === "string" ? value : JSON.stringify(value);
+    const oneLine = text.replaceAll("\n", " ↵ ");
+    return oneLine.length > 140 ? `${oneLine.slice(0, 140)}…` : oneLine;
+}
+function renderToolArgs(args) {
+    const entries = args ? Object.entries(args) : [];
+    if (entries.length === 0) {
+        return "";
+    }
+    const rows = entries
+        .map(([key, value]) => `<div class="tool-arg"><span class="tool-arg-key">${escapeHtml(key)}</span><code>${escapeHtml(formatArgValue(value))}</code></div>`)
+        .join("");
+    return `<div class="tool-args">${rows}</div>`;
+}
+function renderToolCallCard(call) {
+    return `
+<div class="chat-row side-left">
+  <div class="tool-card">
+    <div class="tool-card-head"><span class="tool-icon">&#9881;</span><span class="tool-name">${escapeHtml(call.name)}</span></div>
+    ${renderToolArgs(call.args)}
+  </div>
+</div>`;
+}
+/** Interleave messages and tool calls chronologically when the trace recorded emission order. */
+function buildOrderedTimeline(trace) {
+    const items = [
+        ...trace.messages.map((message) => message.seq === undefined
+            ? undefined
+            : { kind: "message", seq: message.seq, message }),
+        ...trace.toolCalls.map((call) => call.seq === undefined ? undefined : { kind: "tool", seq: call.seq, call }),
+    ].filter((item) => item !== undefined);
+    const totalItems = trace.messages.length + trace.toolCalls.length;
+    if (items.length !== totalItems || totalItems === 0) {
+        return undefined;
+    }
+    return items.sort((a, b) => a.seq - b.seq);
+}
+function renderChat(trace) {
+    if (!trace) {
+        return `<p class="empty">No transcript recorded for this scenario.</p>`;
+    }
+    const parts = [];
+    const timeline = buildOrderedTimeline(trace);
+    if (timeline) {
+        parts.push(`<div class="chat">`);
+        for (const item of timeline) {
+            parts.push(item.kind === "message" ? renderMessageBubble(item.message) : renderToolCallCard(item.call));
+        }
+        parts.push(`</div>`);
+    }
+    else if (trace.messages.length > 0 || trace.toolCalls.length > 0) {
+        parts.push(`<p class="empty note">Emission order wasn't recorded for this trace — messages and tool calls are shown in separate groups below.</p>`);
+        if (trace.messages.length > 0) {
+            parts.push(`<div class="chat">`);
+            for (const message of trace.messages) {
+                parts.push(renderMessageBubble(message));
+            }
+            parts.push(`</div>`);
+        }
+        if (trace.toolCalls.length > 0) {
+            parts.push(`<h4>Tool calls</h4><div class="chat">`);
+            for (const call of trace.toolCalls) {
+                parts.push(renderToolCallCard(call));
+            }
+            parts.push(`</div>`);
+        }
+    }
+    else {
+        parts.push(`<p class="empty">No messages in transcript.</p>`);
+    }
+    if (trace.shellCommands.length > 0) {
+        parts.push(`<h4>Shell commands</h4><ul class="shell-commands">`);
+        for (const command of trace.shellCommands) {
+            parts.push(`<li><code>${escapeHtml(command)}</code></li>`);
+        }
+        parts.push(`</ul>`);
+    }
+    return parts.join("\n");
+}
+function renderJudgeVerdicts(result) {
+    const verdicts = result.judgeVerdicts;
+    if (!verdicts || verdicts.length === 0) {
+        return "";
+    }
+    const items = verdicts
+        .map((verdict) => {
+        const badge = verdict.pass ? "pass" : "fail";
+        const icon = verdict.pass ? "✓" : "✗";
+        return `
+<article class="verdict verdict-${badge}">
+  <p class="question"><span class="verdict-icon">${icon}</span>${escapeHtml(verdict.question)}</p>
+  <p class="rationale">${escapeHtml(verdict.rationale)}</p>
+</article>`;
+    })
+        .join("\n");
+    return `<div class="verdicts">${items}</div>`;
+}
+function renderFailures(result) {
+    if (result.failures.length === 0) {
+        return "";
+    }
+    const items = result.failures
+        .map((failure) => {
+        const { label, hint } = humanizeMatcher(failure.matcher);
+        return `
+<li>
+  <p class="failure-label">${escapeHtml(label)}</p>
+  ${hint ? `<p class="failure-hint">${escapeHtml(hint)}</p>` : ""}
+  <p class="failure-message">${escapeHtml(failure.message)}</p>
+</li>`;
+    })
+        .join("\n");
+    return `<ul class="failures">${items}</ul>`;
+}
+function renderScenario(result) {
+    const open = result.passed || result.skipped ? "" : " open";
+    const failures = renderFailures(result);
+    const judgeVerdicts = renderJudgeVerdicts(result);
+    const diagnostics = failures || judgeVerdicts
+        ? `<div class="diagnostics">
+    ${failures ? `<section><h3>What went wrong</h3>${failures}</section>` : ""}
+    ${judgeVerdicts ? `<section><h3>Judge verdict</h3>${judgeVerdicts}</section>` : ""}
+  </div>`
+        : "";
+    return `
+<details class="scenario ${statusClass(result)}"${open}>
+  <summary>
+    <span class="badge ${statusClass(result)}">${statusLabel(result)}</span>
+    <span class="scenario-name">${escapeHtml(result.scenario)}</span>
+    <span class="duration">${escapeHtml(formatDuration(result.durationMs))}</span>
+  </summary>
+  <div class="scenario-body">
+  ${diagnostics}
+  <section class="conversation">
+    <h3>Conversation</h3>
+    ${renderChat(result.trace)}
+  </section>
+  </div>
+</details>`;
+}
+function renderSuite(report) {
+    const scenarios = report.results.map(renderScenario).join("\n");
+    return `
+<section class="suite">
+  <header class="suite-header">
+    <div><h2>${escapeHtml(report.suite)}</h2><span class="host">${escapeHtml(report.host)}</span></div>
+    <p class="suite-counts">${report.passed} passed · ${report.failed} failed · ${report.skipped} skipped</p>
+  </header>
+  ${scenarios}
+</section>`;
+}
+/** Build a self-contained HTML report for a completed suite run. */
+export function renderHtmlReport(reports, meta = {}) {
+    const generatedAt = meta.generatedAt ?? new Date();
+    const totalPassed = reports.reduce((sum, report) => sum + report.passed, 0);
+    const totalFailed = reports.reduce((sum, report) => sum + report.failed, 0);
+    const totalSkipped = reports.reduce((sum, report) => sum + report.skipped, 0);
+    const host = meta.host ?? reports[0]?.host ?? "unknown";
+    const suitesHtml = reports.map(renderSuite).join("\n");
+    return `<!DOCTYPE html>
+<html lang="en">
+<head>
+<meta charset="utf-8" />
+<meta name="viewport" content="width=device-width, initial-scale=1" />
+<title>agent-test report</title>
+<style>
+  :root {
+    --bg: #0f1419;
+    --panel: #1a2332;
+    --panel-2: #141c28;
+    --text: #e7ecf3;
+    --muted: #8b9bb4;
+    --border: #2a3548;
+    --pass: #3dd68c;
+    --fail: #f07178;
+    --skip: #ffcc66;
+    --user-bubble: #24406b;
+    --assistant-bubble: #1f2f26;
+    --system-bubble: #241f38;
+    --tool: #e0af68;
+    --tool-bubble: #2a2214;
+  }
+  * { box-sizing: border-box; }
+  body {
+    margin: 0;
+    font-family: ui-sans-serif, system-ui, -apple-system, Segoe UI, sans-serif;
+    background: var(--bg);
+    color: var(--text);
+    line-height: 1.45;
+  }
+  main { max-width: 1180px; margin: 0 auto; padding: 1.25rem 1.25rem 3rem; }
+  h1 { font-size: 1.45rem; margin: 0; letter-spacing: -0.02em; }
+  h2 { font-size: 1.1rem; margin: 0; }
+  h3 { font-size: 0.78rem; margin: 0 0 0.5rem; color: var(--muted); text-transform: uppercase; letter-spacing: 0.07em; }
+  h4 { font-size: 0.72rem; margin: 0.9rem 0 0.35rem; color: var(--muted); text-transform: uppercase; letter-spacing: 0.05em; }
+  .muted { color: var(--muted); }
+  .empty { color: var(--muted); font-size: 0.85rem; font-style: italic; margin: 0; }
+  .empty.note { margin-bottom: 0.5rem; }
+  .report-header { display: flex; align-items: center; justify-content: space-between; margin-bottom: 0.65rem; }
+  .report-kicker { color: var(--muted); font-size: 0.78rem; }
+  .summary {
+    background: var(--panel);
+    border: 1px solid var(--border);
+    border-radius: 10px;
+    padding: 0.7rem 0.9rem;
+    display: flex;
+    align-items: center;
+    justify-content: space-between;
+    gap: 1rem;
+  }
+  .stats { display: flex; gap: 0.45rem; align-items: center; }
+  .stat { padding: 0.3rem 0.55rem; border-radius: 6px; background: #111923; font-size: 0.8rem; }
+  .stat strong { font-size: 1rem; margin-right: 0.25rem; }
+  .stat-pass strong { color: var(--pass); }
+  .stat-fail strong { color: var(--fail); }
+  .stat-skip strong { color: var(--skip); }
+  .summary dl {
+    display: flex;
+    flex-wrap: wrap;
+    justify-content: flex-end;
+    gap: 0.15rem 1rem;
+    margin: 0;
+    font-size: 0.78rem;
+  }
+  .summary dt { color: var(--muted); margin-right: -0.7rem; }
+  .summary dd { margin: 0; }
+  .suite { margin-top: 1.25rem; }
+  .suite-header { display: flex; align-items: end; justify-content: space-between; margin: 0 0 0.5rem; padding: 0 0.2rem; }
+  .suite-header > div { display: flex; align-items: center; gap: 0.5rem; }
+  .host { color: var(--muted); font-size: 0.72rem; background: var(--panel); border: 1px solid var(--border); border-radius: 999px; padding: 0.1rem 0.4rem; }
+  .suite-counts { color: var(--muted); margin: 0; font-size: 0.78rem; }
+  details.scenario {
+    background: var(--panel);
+    border: 1px solid var(--border);
+    border-radius: 10px;
+    margin: 0.45rem 0;
+    overflow: hidden;
+  }
+  details.scenario summary {
+    cursor: pointer;
+    display: flex;
+    flex-wrap: wrap;
+    gap: 0.45rem 0.65rem;
+    align-items: center;
+    list-style: none;
+    padding: 0.55rem 0.7rem;
+  }
+  details.scenario[open] summary { border-bottom: 1px solid var(--border); background: #17202d; }
+  details.scenario summary::-webkit-details-marker { display: none; }
+  .scenario-name { font-weight: 600; font-size: 0.9rem; }
+  .duration { color: var(--muted); margin-left: auto; font-variant-numeric: tabular-nums; }
+  .badge {
+    display: inline-block;
+    font-size: 0.66rem;
+    font-weight: 700;
+    text-transform: uppercase;
+    letter-spacing: 0.04em;
+    padding: 0.1rem 0.4rem;
+    border-radius: 999px;
+    border: 1px solid transparent;
+  }
+  .status-passed { color: var(--pass); border-color: color-mix(in srgb, var(--pass) 40%, transparent); }
+  .status-failed { color: var(--fail); border-color: color-mix(in srgb, var(--fail) 40%, transparent); }
+  .status-skipped { color: var(--skip); border-color: color-mix(in srgb, var(--skip) 40%, transparent); }
+  .scenario-body { padding: 0.7rem; }
+  .diagnostics { display: grid; grid-template-columns: repeat(auto-fit, minmax(320px, 1fr)); gap: 0.7rem; margin-bottom: 0.9rem; }
+  .diagnostics section { background: var(--panel-2); border: 1px solid var(--border); border-radius: 8px; padding: 0.65rem 0.7rem; min-width: 0; }
+  .conversation { min-width: 0; }
+
+  .failures { list-style: none; margin: 0; padding: 0; display: flex; flex-direction: column; gap: 0.55rem; }
+  .failures li { border-left: 3px solid var(--fail); padding-left: 0.6rem; }
+  .failure-label { margin: 0; font-weight: 600; font-size: 0.85rem; color: var(--fail); }
+  .failure-hint { margin: 0.1rem 0 0; font-size: 0.78rem; color: var(--muted); }
+  .failure-message { margin: 0.3rem 0 0; font-size: 0.82rem; font-family: ui-monospace, SFMono-Regular, Menlo, monospace; white-space: pre-wrap; word-break: break-word; background: #0b1017; border: 1px solid var(--border); border-radius: 6px; padding: 0.45rem 0.55rem; }
+
+  .verdicts { display: flex; flex-direction: column; gap: 0.55rem; }
+  .verdict { border-left: 3px solid var(--border); padding-left: 0.6rem; }
+  .verdict-pass { border-left-color: var(--pass); }
+  .verdict-fail { border-left-color: var(--fail); }
+  .verdict-icon { display: inline-block; width: 1.1rem; font-weight: 700; }
+  .verdict-pass .verdict-icon { color: var(--pass); }
+  .verdict-fail .verdict-icon { color: var(--fail); }
+  .question { margin: 0; font-size: 0.85rem; font-weight: 600; }
+  .rationale { margin: 0.3rem 0 0; font-size: 0.82rem; color: var(--muted); }
+
+  .chat { display: flex; flex-direction: column; gap: 0.5rem; }
+  .chat-row { display: flex; }
+  .chat-row.side-left { justify-content: flex-start; }
+  .chat-row.side-right { justify-content: flex-end; }
+  .chat-row.side-center { justify-content: center; }
+  .bubble { max-width: 78%; border-radius: 12px; padding: 0.5rem 0.7rem; border: 1px solid var(--border); }
+  .bubble-label { font-size: 0.65rem; font-weight: 700; text-transform: uppercase; letter-spacing: 0.05em; color: var(--muted); margin-bottom: 0.25rem; }
+  .bubble-text { white-space: pre-wrap; word-break: break-word; font-size: 0.85rem; }
+  .bubble.role-user { background: var(--user-bubble); border-top-right-radius: 3px; }
+  .bubble.role-assistant { background: var(--assistant-bubble); border-top-left-radius: 3px; }
+  .bubble.role-system, .bubble.role-tool { background: var(--system-bubble); font-size: 0.8rem; max-width: 90%; }
+
+  .tool-card {
+    max-width: 82%;
+    border-radius: 10px;
+    padding: 0.5rem 0.65rem;
+    border: 1px solid color-mix(in srgb, var(--tool) 35%, var(--border));
+    background: var(--tool-bubble);
+  }
+  .tool-card-head {
+    display: flex;
+    align-items: center;
+    gap: 0.4rem;
+    font-size: 0.8rem;
+    font-weight: 700;
+    color: var(--tool);
+  }
+  .tool-icon { font-size: 0.85rem; }
+  .tool-name { font-weight: 700; }
+  .tool-args {
+    margin-top: 0.4rem;
+    display: flex;
+    flex-direction: column;
+    gap: 0.25rem;
+    padding-top: 0.4rem;
+    border-top: 1px solid color-mix(in srgb, var(--tool) 20%, var(--border));
+  }
+  .tool-arg { display: flex; gap: 0.5rem; align-items: baseline; font-size: 0.78rem; }
+  .tool-arg-key { color: var(--muted); flex-shrink: 0; }
+  .tool-arg code {
+    color: var(--text);
+    background: #0b1017;
+    border-radius: 4px;
+    padding: 0.05rem 0.35rem;
+    word-break: break-word;
+  }
+
+  .shell-commands { margin: 0; padding-left: 0; list-style: none; display: flex; flex-direction: column; gap: 0.3rem; }
+  .shell-commands li { font-size: 0.8rem; background: #0b1017; border: 1px solid var(--border); border-radius: 6px; padding: 0.3rem 0.5rem; }
+  code { font-family: ui-monospace, SFMono-Regular, Menlo, monospace; font-size: 0.85em; }
+  @media (max-width: 720px) {
+    main { padding: 0.75rem; }
+    .summary { align-items: flex-start; flex-direction: column; }
+    .summary dl { justify-content: flex-start; }
+    .suite-header { align-items: flex-start; flex-direction: column; gap: 0.2rem; }
+    .diagnostics { grid-template-columns: 1fr; }
+    .bubble, .tool-card { max-width: 92%; }
+  }
+</style>
+</head>
+<body>
+<main>
+  <header class="report-header">
+    <h1>agent-test report</h1>
+    <span class="report-kicker">${reports.length} suite${reports.length === 1 ? "" : "s"}</span>
+  </header>
+  <div class="summary">
+    <div class="stats">
+      <span class="stat stat-pass"><strong>${totalPassed}</strong> passed</span>
+      <span class="stat stat-fail"><strong>${totalFailed}</strong> failed</span>
+      <span class="stat stat-skip"><strong>${totalSkipped}</strong> skipped</span>
+    </div>
+    <dl>
+      <dt>Host</dt><dd>${escapeHtml(String(host))}</dd>
+      <dt>Generated</dt><dd>${escapeHtml(generatedAt.toISOString())}</dd>
+      ${meta.suitesDir ? `<dt>Suites</dt><dd>${escapeHtml(meta.suitesDir)}</dd>` : ""}
+    </dl>
+  </div>
+  ${suitesHtml}
+</main>
+</body>
+</html>
+`;
+}
+/** Write an HTML report under a fresh temp directory; returns the report file path. */
+export async function writeHtmlReport(reports, meta = {}) {
+    const dir = await mkdtemp(join(tmpdir(), "agent-test-report-"));
+    const path = join(dir, "report.html");
+    const html = renderHtmlReport(reports, meta);
+    await writeFile(path, html, "utf8");
+    return path;
+}
+//# sourceMappingURL=html-report.js.map

package/dist/html-report.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"html-report.js","sourceRoot":"","sources":["../src/html-report.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,OAAO,EAAE,SAAS,EAAE,MAAM,kBAAkB,CAAC;AACtD,OAAO,EAAE,MAAM,EAAE,MAAM,SAAS,CAAC;AACjC,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AAYjC,SAAS,UAAU,CAAC,KAAa;IAChC,OAAO,KAAK;SACV,UAAU,CAAC,GAAG,EAAE,OAAO,CAAC;SACxB,UAAU,CAAC,GAAG,EAAE,MAAM,CAAC;SACvB,UAAU,CAAC,GAAG,EAAE,MAAM,CAAC;SACvB,UAAU,CAAC,GAAG,EAAE,QAAQ,CAAC;SACzB,UAAU,CAAC,GAAG,EAAE,OAAO,CAAC,CAAC;AAC5B,CAAC;AAED,SAAS,cAAc,CAAC,EAAU;IACjC,IAAI,EAAE,GAAG,IAAI,EAAE,CAAC;QACf,OAAO,GAAG,IAAI,CAAC,KAAK,CAAC,EAAE,CAAC,IAAI,CAAC;IAC9B,CAAC;IACD,OAAO,GAAG,CAAC,EAAE,GAAG,IAAI,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG,CAAC;AACrC,CAAC;AAED,SAAS,WAAW,CAAC,MAAsB;IAC1C,IAAI,MAAM,CAAC,OAAO,EAAE,CAAC;QACpB,OAAO,SAAS,CAAC;IAClB,CAAC;IACD,OAAO,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,QAAQ,CAAC;AAC5C,CAAC;AAED,SAAS,WAAW,CAAC,MAAsB;IAC1C,IAAI,MAAM,CAAC,OAAO,EAAE,CAAC;QACpB,OAAO,gBAAgB,CAAC;IACzB,CAAC;IACD,OAAO,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,eAAe,CAAC,CAAC,CAAC,eAAe,CAAC;AAC1D,CAAC;AAED,iGAAiG;AACjG,MAAM,cAAc,GAAqD;IACxE,QAAQ,EAAE,EAAE,KAAK,EAAE,WAAW,EAAE,IAAI,EAAE,mDAAmD,EAAE;IAC3F,IAAI,EAAE;QACL,KAAK,EAAE,qBAAqB;QAC5B,IAAI,EAAE,wDAAwD;KAC9D;IACD,OAAO,EAAE,EAAE,KAAK,EAAE,oBAAoB,EAAE,IAAI,EAAE,gDAAgD,EAAE;IAChG,OAAO,EAAE,EAAE,KAAK,EAAE,iBAAiB,EAAE,IAAI,EAAE,gCAAgC,EAAE;IAC7E,eAAe,EAAE;QAChB,KAAK,EAAE,eAAe;QACtB,IAAI,EAAE,uDAAuD;KAC7D;IACD,kBAAkB,EAAE;QACnB,KAAK,EAAE,kBAAkB;QACzB,IAAI,EAAE,gDAAgD;KACtD;IACD,YAAY,EAAE,EAAE,KAAK,EAAE,SAAS,EAAE,IAAI,EAAE,qDAAqD,EAAE;IAC/F,eAAe,EAAE;QAChB,KAAK,EAAE,mBAAmB;QAC1B,IAAI,EAAE,yDAAyD;KAC/D;IACD,WAAW,EAAE,EAAE,KAAK,EAAE,kBAAkB,EAAE,IAAI,EAAE,kCAAkC,EAAE;IACpF,KAAK,EAAE,EAAE,KAAK,EAAE,OAAO,EAAE,IAAI,EAAE,sCAAsC,EAAE;CACvE,CAAC;AAEF,SAAS,eAAe,CAAC,OAAe;IACvC,MAAM,CAAC,IAAI,CAAC,GAAG,OAAO,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC;IAClC,MAAM,KAAK,GAAG,cAAc,CAAC,IAAI,IAAI,OAAO,CAAC,CAAC;IAC9C,IAAI,KAAK,EAAE,CAAC;QACX,IAAI,OAAO,CAAC,QAAQ,CAAC,GAAG,CAAC,EAAE,CAAC;YAC3B,MAAM,MAAM,GAAG,OAAO,CAAC,KAAK,CAAC,OAAO,CAAC,OAAO,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC,CAAC;YACvD,OAAO,EAAE,KAAK,EAAE,GAAG,KAAK,CAAC,KAAK,MAAM,MAAM,EAAE,EAAE,IAAI,EAAE,KAAK,CAAC,IAAI,EAAE,CAAC;QAClE,CAAC;QACD,OAAO,KAAK,CAAC;IACd,CAAC;IACD,OAAO,EAAE,KAAK,EAAE,OAAO,EAAE,CAAC;AAC3B,CAAC;AAED,MAAM,SAAS,GAAyE;IACvF,IAAI,EAAE,EAAE,KAAK,EAAE,MAAM,EAAE,IAAI,EAAE,OAAO,EAAE;IACtC,SAAS,EAAE,EAAE,KAAK,EAAE,OAAO,EAAE,IAAI,EAAE,MAAM,EAAE;IAC3C,MAAM,EAAE,EAAE,KAAK,EAAE,QAAQ,EAAE,IAAI,EAAE,QAAQ,EAAE;IAC3C,IAAI,EAAE,EAAE,KAAK,EAAE,MAAM,EAAE,IAAI,EAAE,QAAQ,EAAE;CACvC,CAAC;AAEF,SAAS,mBAAmB,CAAC,OAAqB;IACjD,MAAM,IAAI,GAAG,SAAS,CAAC,OAAO,CAAC,IAAI,CAAC,IAAI,EAAE,KAAK,EAAE,OAAO,CAAC,IAAI,EAAE,IAAI,EAAE,QAAQ,EAAE,CAAC;IAChF,OAAO;4BACoB,IAAI,CAAC,IAAI;4BACT,UAAU,CAAC,OAAO,CAAC,IAAI,CAAC;gCACpB,UAAU,CAAC,IAAI,CAAC,KAAK,CAAC;+BACvB,UAAU,CAAC,OAAO,CAAC,OAAO,CAAC;;OAEnD,CAAC;AACR,CAAC;AAED,0FAA0F;AAC1F,SAAS,cAAc,CAAC,KAAc;IACrC,MAAM,IAAI,GAAG,OAAO,KAAK,KAAK,QAAQ,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,IAAI,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC;IACvE,MAAM,OAAO,GAAG,IAAI,CAAC,UAAU,CAAC,IAAI,EAAE,KAAK,CAAC,CAAC;IAC7C,OAAO,OAAO,CAAC,MAAM,GAAG,GAAG,CAAC,CAAC,CAAC,GAAG,OAAO,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,GAAG,CAAC,CAAC,CAAC,OAAO,CAAC;AACrE,CAAC;AAED,SAAS,cAAc,CAAC,IAAyC;IAChE,MAAM,OAAO,GAAG,IAAI,CAAC,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC;IACjD,IAAI,OAAO,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QAC1B,OAAO,EAAE,CAAC;IACX,CAAC;IACD,MAAM,IAAI,GAAG,OAAO;SAClB,GAAG,CACH,CAAC,CAAC,GAAG,EAAE,KAAK,CAAC,EAAE,EAAE,CAChB,oDAAoD,UAAU,CAAC,GAAG,CAAC,gBAAgB,UAAU,CAAC,cAAc,CAAC,KAAK,CAAC,CAAC,eAAe,CACpI;SACA,IAAI,CAAC,EAAE,CAAC,CAAC;IACX,OAAO,0BAA0B,IAAI,QAAQ,CAAC;AAC/C,CAAC;AAED,SAAS,kBAAkB,CAAC,IAAmB;IAC9C,OAAO;;;gGAGwF,UAAU,CAAC,IAAI,CAAC,IAAI,CAAC;MAC/G,cAAc,CAAC,IAAI,CAAC,IAAI,CAAC;;OAExB,CAAC;AACR,CAAC;AAMD,iGAAiG;AACjG,SAAS,oBAAoB,CAAC,KAAiB;IAC9C,MAAM,KAAK,GAAmB;QAC7B,GAAG,KAAK,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,OAAO,EAAE,EAAE,CACjC,OAAO,CAAC,GAAG,KAAK,SAAS;YACxB,CAAC,CAAC,SAAS;YACX,CAAC,CAAE,EAAE,IAAI,EAAE,SAAS,EAAE,GAAG,EAAE,OAAO,CAAC,GAAG,EAAE,OAAO,EAAY,CAC5D;QACD,GAAG,KAAK,CAAC,SAAS,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAC/B,IAAI,CAAC,GAAG,KAAK,SAAS,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAE,EAAE,IAAI,EAAE,MAAM,EAAE,GAAG,EAAE,IAAI,CAAC,GAAG,EAAE,IAAI,EAAY,CACrF;KACD,CAAC,MAAM,CAAC,CAAC,IAAI,EAAwB,EAAE,CAAC,IAAI,KAAK,SAAS,CAAC,CAAC;IAE7D,MAAM,UAAU,GAAG,KAAK,CAAC,QAAQ,CAAC,MAAM,GAAG,KAAK,CAAC,SAAS,CAAC,MAAM,CAAC;IAClE,IAAI,KAAK,CAAC,MAAM,KAAK,UAAU,IAAI,UAAU,KAAK,CAAC,EAAE,CAAC;QACrD,OAAO,SAAS,CAAC;IAClB,CAAC;IAED,OAAO,KAAK,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,GAAG,GAAG,CAAC,CAAC,GAAG,CAAC,CAAC;AAC5C,CAAC;AAED,SAAS,UAAU,CAAC,KAA6B;IAChD,IAAI,CAAC,KAAK,EAAE,CAAC;QACZ,OAAO,gEAAgE,CAAC;IACzE,CAAC;IAED,MAAM,KAAK,GAAa,EAAE,CAAC;IAC3B,MAAM,QAAQ,GAAG,oBAAoB,CAAC,KAAK,CAAC,CAAC;IAE7C,IAAI,QAAQ,EAAE,CAAC;QACd,KAAK,CAAC,IAAI,CAAC,oBAAoB,CAAC,CAAC;QACjC,KAAK,MAAM,IAAI,IAAI,QAAQ,EAAE,CAAC;YAC7B,KAAK,CAAC,IAAI,CACT,IAAI,CAAC,IAAI,KAAK,SAAS,CAAC,CAAC,CAAC,mBAAmB,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,kBAAkB,CAAC,IAAI,CAAC,IAAI,CAAC,CAC3F,CAAC;QACH,CAAC;QACD,KAAK,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;IACtB,CAAC;SAAM,IAAI,KAAK,CAAC,QAAQ,CAAC,MAAM,GAAG,CAAC,IAAI,KAAK,CAAC,SAAS,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACpE,KAAK,CAAC,IAAI,CACT,uIAAuI,CACvI,CAAC;QACF,IAAI,KAAK,CAAC,QAAQ,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YAC/B,KAAK,CAAC,IAAI,CAAC,oBAAoB,CAAC,CAAC;YACjC,KAAK,MAAM,OAAO,IAAI,KAAK,CAAC,QAAQ,EAAE,CAAC;gBACtC,KAAK,CAAC,IAAI,CAAC,mBAAmB,CAAC,OAAO,CAAC,CAAC,CAAC;YAC1C,CAAC;YACD,KAAK,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;QACtB,CAAC;QACD,IAAI,KAAK,CAAC,SAAS,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YAChC,KAAK,CAAC,IAAI,CAAC,uCAAuC,CAAC,CAAC;YACpD,KAAK,MAAM,IAAI,IAAI,KAAK,CAAC,SAAS,EAAE,CAAC;gBACpC,KAAK,CAAC,IAAI,CAAC,kBAAkB,CAAC,IAAI,CAAC,CAAC,CAAC;YACtC,CAAC;YACD,KAAK,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;QACtB,CAAC;IACF,CAAC;SAAM,CAAC;QACP,KAAK,CAAC,IAAI,CAAC,iDAAiD,CAAC,CAAC;IAC/D,CAAC;IAED,IAAI,KAAK,CAAC,aAAa,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACpC,KAAK,CAAC,IAAI,CAAC,oDAAoD,CAAC,CAAC;QACjE,KAAK,MAAM,OAAO,IAAI,KAAK,CAAC,aAAa,EAAE,CAAC;YAC3C,KAAK,CAAC,IAAI,CAAC,aAAa,UAAU,CAAC,OAAO,CAAC,cAAc,CAAC,CAAC;QAC5D,CAAC;QACD,KAAK,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC;IACrB,CAAC;IAED,OAAO,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AACzB,CAAC;AAED,SAAS,mBAAmB,CAAC,MAAsB;IAClD,MAAM,QAAQ,GAAG,MAAM,CAAC,aAAa,CAAC;IACtC,IAAI,CAAC,QAAQ,IAAI,QAAQ,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACxC,OAAO,EAAE,CAAC;IACX,CAAC;IAED,MAAM,KAAK,GAAG,QAAQ;SACpB,GAAG,CAAC,CAAC,OAAO,EAAE,EAAE;QAChB,MAAM,KAAK,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,CAAC;QAC7C,MAAM,IAAI,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,GAAG,CAAC;QACtC,OAAO;kCACwB,KAAK;mDACY,IAAI,UAAU,UAAU,CAAC,OAAO,CAAC,QAAQ,CAAC;yBACpE,UAAU,CAAC,OAAO,CAAC,SAAS,CAAC;WAC3C,CAAC;IACV,CAAC,CAAC;SACD,IAAI,CAAC,IAAI,CAAC,CAAC;IAEb,OAAO,yBAAyB,KAAK,QAAQ,CAAC;AAC/C,CAAC;AAED,SAAS,cAAc,CAAC,MAAsB;IAC7C,IAAI,MAAM,CAAC,QAAQ,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QAClC,OAAO,EAAE,CAAC;IACX,CAAC;IAED,MAAM,KAAK,GAAG,MAAM,CAAC,QAAQ;SAC3B,GAAG,CAAC,CAAC,OAAO,EAAE,EAAE;QAChB,MAAM,EAAE,KAAK,EAAE,IAAI,EAAE,GAAG,eAAe,CAAC,OAAO,CAAC,OAAO,CAAC,CAAC;QACzD,OAAO;;6BAEmB,UAAU,CAAC,KAAK,CAAC;IAC1C,IAAI,CAAC,CAAC,CAAC,2BAA2B,UAAU,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE;+BAClC,UAAU,CAAC,OAAO,CAAC,OAAO,CAAC;MACpD,CAAC;IACL,CAAC,CAAC;SACD,IAAI,CAAC,IAAI,CAAC,CAAC;IAEb,OAAO,wBAAwB,KAAK,OAAO,CAAC;AAC7C,CAAC;AAED,SAAS,cAAc,CAAC,MAAsB;IAC7C,MAAM,IAAI,GAAG,MAAM,CAAC,MAAM,IAAI,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,OAAO,CAAC;IAC5D,MAAM,QAAQ,GAAG,cAAc,CAAC,MAAM,CAAC,CAAC;IACxC,MAAM,aAAa,GAAG,mBAAmB,CAAC,MAAM,CAAC,CAAC;IAClD,MAAM,WAAW,GAChB,QAAQ,IAAI,aAAa;QACxB,CAAC,CAAC;MACC,QAAQ,CAAC,CAAC,CAAC,oCAAoC,QAAQ,YAAY,CAAC,CAAC,CAAC,EAAE;MACxE,aAAa,CAAC,CAAC,CAAC,kCAAkC,aAAa,YAAY,CAAC,CAAC,CAAC,EAAE;SAC7E;QACN,CAAC,CAAC,EAAE,CAAC;IACP,OAAO;2BACmB,WAAW,CAAC,MAAM,CAAC,IAAI,IAAI;;yBAE7B,WAAW,CAAC,MAAM,CAAC,KAAK,WAAW,CAAC,MAAM,CAAC;kCAClC,UAAU,CAAC,MAAM,CAAC,QAAQ,CAAC;6BAChC,UAAU,CAAC,cAAc,CAAC,MAAM,CAAC,UAAU,CAAC,CAAC;;;IAGtE,WAAW;;;MAGT,UAAU,CAAC,MAAM,CAAC,KAAK,CAAC;;;WAGnB,CAAC;AACZ,CAAC;AAED,SAAS,WAAW,CAAC,MAAsB;IAC1C,MAAM,SAAS,GAAG,MAAM,CAAC,OAAO,CAAC,GAAG,CAAC,cAAc,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IAChE,OAAO;;;eAGO,UAAU,CAAC,MAAM,CAAC,KAAK,CAAC,2BAA2B,UAAU,CAAC,MAAM,CAAC,IAAI,CAAC;8BAC3D,MAAM,CAAC,MAAM,aAAa,MAAM,CAAC,MAAM,aAAa,MAAM,CAAC,OAAO;;IAE5F,SAAS;WACF,CAAC;AACZ,CAAC;AAED,oEAAoE;AACpE,MAAM,UAAU,gBAAgB,CAAC,OAAyB,EAAE,IAAI,GAAmB,EAAE;IACpF,MAAM,WAAW,GAAG,IAAI,CAAC,WAAW,IAAI,IAAI,IAAI,EAAE,CAAC;IACnD,MAAM,WAAW,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,MAAM,EAAE,EAAE,CAAC,GAAG,GAAG,MAAM,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC;IAC5E,MAAM,WAAW,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,MAAM,EAAE,EAAE,CAAC,GAAG,GAAG,MAAM,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC;IAC5E,MAAM,YAAY,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,MAAM,EAAE,EAAE,CAAC,GAAG,GAAG,MAAM,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC;IAC9E,MAAM,IAAI,GAAG,IAAI,CAAC,IAAI,IAAI,OAAO,CAAC,CAAC,CAAC,EAAE,IAAI,IAAI,SAAS,CAAC;IAExD,MAAM,UAAU,GAAG,OAAO,CAAC,GAAG,CAAC,WAAW,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IAEvD,OAAO;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;kCA8L0B,OAAO,CAAC,MAAM,SAAS,OAAO,CAAC,MAAM,KAAK,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,GAAG;;;;6CAI3C,WAAW;6CACX,WAAW;6CACX,YAAY;;;yBAGhC,UAAU,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC;8BACnB,UAAU,CAAC,WAAW,CAAC,WAAW,EAAE,CAAC;QAC3D,IAAI,CAAC,SAAS,CAAC,CAAC,CAAC,sBAAsB,UAAU,CAAC,IAAI,CAAC,SAAS,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE;;;IAGjF,UAAU;;;;CAIb,CAAC;AACF,CAAC;AAED,uFAAuF;AACvF,MAAM,CAAC,KAAK,UAAU,eAAe,CACpC,OAAyB,EACzB,IAAI,GAAmB,EAAE;IAEzB,MAAM,GAAG,GAAG,MAAM,OAAO,CAAC,IAAI,CAAC,MAAM,EAAE,EAAE,oBAAoB,CAAC,CAAC,CAAC;IAChE,MAAM,IAAI,GAAG,IAAI,CAAC,GAAG,EAAE,aAAa,CAAC,CAAC;IACtC,MAAM,IAAI,GAAG,gBAAgB,CAAC,OAAO,EAAE,IAAI,CAAC,CAAC;IAC7C,MAAM,SAAS,CAAC,IAAI,EAAE,IAAI,EAAE,MAAM,CAAC,CAAC;IACpC,OAAO,IAAI,CAAC;AACb,CAAC"}

package/dist/index.d.ts

@@ -1,7 +1,8 @@
 export { assertRubric, expectTrace, TraceAssertion } from "./expect.js";
+export { type HtmlReportMeta, renderHtmlReport, writeHtmlReport } from "./html-report.js";
 export { loadSuiteFile } from "./load-suite.js";
 export { assertLiveDogfoodPreflight } from "./preflight.js";
-export { cleanupLegacyRepoRecordings, cleanupStagingSession, createLiveStagingSessionId, getLiveStagingRoot, getLiveStagingSessionRoot, type RecordingPathKind, type ResolvedRecordingPath, recordTrace, resolveRecordingPath, } from "./record-trace.js";
+export { cleanupLegacyRepoRecordings, cleanupStagingSession, createLiveStagingSessionId, getLiveStagingRoot, getLiveStagingSessionRoot, getStagingResultPath, type LiveScenarioResultSidecar, loadStagingResult, type RecordingPathKind, type ResolvedRecordingPath, recordTrace, resolveRecordingPath, writeStagingResult, } from "./record-trace.js";
 export { discoverSuites, type RunSuiteOptions, runAllSuites, runSuite, } from "./run-suite.js";
 export type { AgentScenario, AgentSuiteFile, AssertionFailure, JudgeRubricItem, JudgeVerdictResult, ScenarioResult, ScenarioRubric, SuiteRunReport, } from "./types.js";
 //# sourceMappingURL=index.d.ts.map

package/dist/index.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,YAAY,EAAE,WAAW,EAAE,cAAc,EAAE,MAAM,aAAa,CAAC;AAExE,OAAO,EAAE,aAAa,EAAE,MAAM,iBAAiB,CAAC;AAChD,OAAO,EAAE,0BAA0B,EAAE,MAAM,gBAAgB,CAAC;AAC5D,OAAO,EACN,2BAA2B,EAC3B,qBAAqB,EACrB,0BAA0B,EAC1B,kBAAkB,EAClB,yBAAyB,EACzB,KAAK,iBAAiB,EACtB,KAAK,qBAAqB,EAC1B,WAAW,EACX,oBAAoB,GACpB,MAAM,mBAAmB,CAAC;AAC3B,OAAO,EACN,cAAc,EACd,KAAK,eAAe,EACpB,YAAY,EACZ,QAAQ,GACR,MAAM,gBAAgB,CAAC;AACxB,YAAY,EACX,aAAa,EACb,cAAc,EACd,gBAAgB,EAChB,eAAe,EACf,kBAAkB,EAClB,cAAc,EACd,cAAc,EACd,cAAc,GACd,MAAM,YAAY,CAAC"}
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,YAAY,EAAE,WAAW,EAAE,cAAc,EAAE,MAAM,aAAa,CAAC;AACxE,OAAO,EAAE,KAAK,cAAc,EAAE,gBAAgB,EAAE,eAAe,EAAE,MAAM,kBAAkB,CAAC;AAC1F,OAAO,EAAE,aAAa,EAAE,MAAM,iBAAiB,CAAC;AAChD,OAAO,EAAE,0BAA0B,EAAE,MAAM,gBAAgB,CAAC;AAC5D,OAAO,EACN,2BAA2B,EAC3B,qBAAqB,EACrB,0BAA0B,EAC1B,kBAAkB,EAClB,yBAAyB,EACzB,oBAAoB,EACpB,KAAK,yBAAyB,EAC9B,iBAAiB,EACjB,KAAK,iBAAiB,EACtB,KAAK,qBAAqB,EAC1B,WAAW,EACX,oBAAoB,EACpB,kBAAkB,GAClB,MAAM,mBAAmB,CAAC;AAC3B,OAAO,EACN,cAAc,EACd,KAAK,eAAe,EACpB,YAAY,EACZ,QAAQ,GACR,MAAM,gBAAgB,CAAC;AACxB,YAAY,EACX,aAAa,EACb,cAAc,EACd,gBAAgB,EAChB,eAAe,EACf,kBAAkB,EAClB,cAAc,EACd,cAAc,EACd,cAAc,GACd,MAAM,YAAY,CAAC"}

package/dist/index.js

@@ -1,6 +1,7 @@
 export { assertRubric, expectTrace, TraceAssertion } from "./expect.js";
+export { renderHtmlReport, writeHtmlReport } from "./html-report.js";
 export { loadSuiteFile } from "./load-suite.js";
 export { assertLiveDogfoodPreflight } from "./preflight.js";
-export { cleanupLegacyRepoRecordings, cleanupStagingSession, createLiveStagingSessionId, getLiveStagingRoot, getLiveStagingSessionRoot, recordTrace, resolveRecordingPath, } from "./record-trace.js";
+export { cleanupLegacyRepoRecordings, cleanupStagingSession, createLiveStagingSessionId, getLiveStagingRoot, getLiveStagingSessionRoot, getStagingResultPath, loadStagingResult, recordTrace, resolveRecordingPath, writeStagingResult, } from "./record-trace.js";
 export { discoverSuites, runAllSuites, runSuite, } from "./run-suite.js";
 //# sourceMappingURL=index.js.map

package/dist/index.js.map

@@ -1 +1 @@
-{"version":3,"file":"index.js","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,YAAY,EAAE,WAAW,EAAE,cAAc,EAAE,MAAM,aAAa,CAAC;AAExE,OAAO,EAAE,aAAa,EAAE,MAAM,iBAAiB,CAAC;AAChD,OAAO,EAAE,0BAA0B,EAAE,MAAM,gBAAgB,CAAC;AAC5D,OAAO,EACN,2BAA2B,EAC3B,qBAAqB,EACrB,0BAA0B,EAC1B,kBAAkB,EAClB,yBAAyB,EAGzB,WAAW,EACX,oBAAoB,GACpB,MAAM,mBAAmB,CAAC;AAC3B,OAAO,EACN,cAAc,EAEd,YAAY,EACZ,QAAQ,GACR,MAAM,gBAAgB,CAAC"}
+{"version":3,"file":"index.js","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,YAAY,EAAE,WAAW,EAAE,cAAc,EAAE,MAAM,aAAa,CAAC;AACxE,OAAO,EAAuB,gBAAgB,EAAE,eAAe,EAAE,MAAM,kBAAkB,CAAC;AAC1F,OAAO,EAAE,aAAa,EAAE,MAAM,iBAAiB,CAAC;AAChD,OAAO,EAAE,0BAA0B,EAAE,MAAM,gBAAgB,CAAC;AAC5D,OAAO,EACN,2BAA2B,EAC3B,qBAAqB,EACrB,0BAA0B,EAC1B,kBAAkB,EAClB,yBAAyB,EACzB,oBAAoB,EAEpB,iBAAiB,EAGjB,WAAW,EACX,oBAAoB,EACpB,kBAAkB,GAClB,MAAM,mBAAmB,CAAC;AAC3B,OAAO,EACN,cAAc,EAEd,YAAY,EACZ,QAAQ,GACR,MAAM,gBAAgB,CAAC"}

package/dist/live-isolation.d.ts

@@ -16,6 +16,12 @@ export interface SpawnLiveScenarioOptions {
     scenarioIndex?: number;
     /** Total scenarios in the full suite (for child CLI counters). */
     scenarioTotal?: number;
+    /** In-process harness timeout (parent adds a kill backstop). */
+    timeoutMs?: number;
+    /** Disable harness deadline in the child (forwards --no-timeout). */
+    noTimeout?: boolean;
+    /** Allow AskQuestion-style tools (default false — live is single-shot). */
+    allowUserInput?: boolean;
 }
 export interface LiveScenarioCommand {
     command: string;
@@ -24,9 +30,26 @@ export interface LiveScenarioCommand {
 }
 /** Build the Node subprocess command for one live scenario (same CLI entry as the parent). */
 export declare function buildLiveScenarioCommand(options: SpawnLiveScenarioOptions): LiveScenarioCommand;
+/** Delay from agent-start marker until parent SIGTERM (harness deadline + cleanup slack). */
+export declare function subprocessKillDelayMs(agentStartMs: number, agentTimeoutMs: number): number;
 /** Run one live scenario in a fresh Node subprocess; inherit stdio for live progress. */
 export declare function spawnLiveScenario(options: SpawnLiveScenarioOptions): Promise<number>;
 export declare function subprocessFailureMessage(exitCode: number): string;
+export interface LiveSubprocessStagingResult {
+    passed: boolean;
+    failures: Array<{
+        matcher: string;
+        message: string;
+    }>;
+}
+/**
+ * Map isolated child exit + optional staging sidecar to parent failures.
+ * A persisted pass sidecar wins over a late timeout kill (exit 124).
+ */
+export declare function failuresForLiveSubprocessExit(exitCode: number, childResult: LiveSubprocessStagingResult | undefined): Array<{
+    matcher: string;
+    message: string;
+}>;
 /** Parent-provided counters for isolated child runs (1-based index). */
 export declare function parentScenarioCounters(): {
     index: number;

package/dist/live-isolation.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"live-isolation.d.ts","sourceRoot":"","sources":["../src/live-isolation.ts"],"names":[],"mappings":"AAKA,mGAAmG;AACnG,wBAAgB,4BAA4B,IAAI,OAAO,CAKtD;AAED,wBAAgB,gBAAgB,IAAI,MAAM,CAQzC;AAED,MAAM,WAAW,wBAAwB;IACxC,GAAG,EAAE,MAAM,CAAC;IACZ,SAAS,EAAE,MAAM,CAAC;IAClB,YAAY,EAAE,MAAM,CAAC;IACrB,SAAS,EAAE,MAAM,CAAC;IAClB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,gBAAgB,CAAC,EAAE,MAAM,CAAC;IAC1B,cAAc,CAAC,EAAE,OAAO,CAAC;IACzB,cAAc,CAAC,EAAE,OAAO,CAAC;IACzB,QAAQ,CAAC,EAAE,OAAO,CAAC;IACnB,KAAK,CAAC,EAAE,OAAO,CAAC;IAChB,gEAAgE;IAChE,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB,kEAAkE;IAClE,aAAa,CAAC,EAAE,MAAM,CAAC;CACvB;AAED,MAAM,WAAW,mBAAmB;IACnC,OAAO,EAAE,MAAM,CAAC;IAChB,IAAI,EAAE,MAAM,EAAE,CAAC;IACf,QAAQ,EAAE,MAAM,EAAE,CAAC;CACnB;AAED,8FAA8F;AAC9F,wBAAgB,wBAAwB,CACvC,OAAO,EAAE,wBAAwB,GAC/B,mBAAmB,CA+BrB;AAQD,yFAAyF;AACzF,wBAAsB,iBAAiB,CACtC,OAAO,EAAE,wBAAwB,GAC/B,OAAO,CAAC,MAAM,CAAC,CAmCjB;AAED,wBAAgB,wBAAwB,CAAC,QAAQ,EAAE,MAAM,GAAG,MAAM,CAKjE;AAED,wEAAwE;AACxE,wBAAgB,sBAAsB,IACnC;IAAE,KAAK,EAAE,MAAM,CAAC;IAAC,KAAK,EAAE,MAAM,CAAA;CAAE,GAChC,SAAS,CAYX"}
+{"version":3,"file":"live-isolation.d.ts","sourceRoot":"","sources":["../src/live-isolation.ts"],"names":[],"mappings":"AAYA,mGAAmG;AACnG,wBAAgB,4BAA4B,IAAI,OAAO,CAEtD;AAED,wBAAgB,gBAAgB,IAAI,MAAM,CAIzC;AAED,MAAM,WAAW,wBAAwB;IACxC,GAAG,EAAE,MAAM,CAAC;IACZ,SAAS,EAAE,MAAM,CAAC;IAClB,YAAY,EAAE,MAAM,CAAC;IACrB,SAAS,EAAE,MAAM,CAAC;IAClB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,gBAAgB,CAAC,EAAE,MAAM,CAAC;IAC1B,cAAc,CAAC,EAAE,OAAO,CAAC;IACzB,cAAc,CAAC,EAAE,OAAO,CAAC;IACzB,QAAQ,CAAC,EAAE,OAAO,CAAC;IACnB,KAAK,CAAC,EAAE,OAAO,CAAC;IAChB,gEAAgE;IAChE,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB,kEAAkE;IAClE,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB,gEAAgE;IAChE,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,qEAAqE;IACrE,SAAS,CAAC,EAAE,OAAO,CAAC;IACpB,2EAA2E;IAC3E,cAAc,CAAC,EAAE,OAAO,CAAC;CACzB;AAED,MAAM,WAAW,mBAAmB;IACnC,OAAO,EAAE,MAAM,CAAC;IAChB,IAAI,EAAE,MAAM,EAAE,CAAC;IACf,QAAQ,EAAE,MAAM,EAAE,CAAC;CACnB;AAED,8FAA8F;AAC9F,wBAAgB,wBAAwB,CAAC,OAAO,EAAE,wBAAwB,GAAG,mBAAmB,CAsC/F;AA4BD,6FAA6F;AAC7F,wBAAgB,qBAAqB,CAAC,YAAY,EAAE,MAAM,EAAE,cAAc,EAAE,MAAM,GAAG,MAAM,CAE1F;AAED,yFAAyF;AACzF,wBAAsB,iBAAiB,CAAC,OAAO,EAAE,wBAAwB,GAAG,OAAO,CAAC,MAAM,CAAC,CAgH1F;AAED,wBAAgB,wBAAwB,CAAC,QAAQ,EAAE,MAAM,GAAG,MAAM,CAQjE;AAED,MAAM,WAAW,2BAA2B;IAC3C,MAAM,EAAE,OAAO,CAAC;IAChB,QAAQ,EAAE,KAAK,CAAC;QAAE,OAAO,EAAE,MAAM,CAAC;QAAC,OAAO,EAAE,MAAM,CAAA;KAAE,CAAC,CAAC;CACtD;AAED;;;GAGG;AACH,wBAAgB,6BAA6B,CAC5C,QAAQ,EAAE,MAAM,EAChB,WAAW,EAAE,2BAA2B,GAAG,SAAS,GAClD,KAAK,CAAC;IAAE,OAAO,EAAE,MAAM,CAAC;IAAC,OAAO,EAAE,MAAM,CAAA;CAAE,CAAC,CAgB7C;AAED,wEAAwE;AACxE,wBAAgB,sBAAsB,IAAI;IAAE,KAAK,EAAE,MAAM,CAAC;IAAC,KAAK,EAAE,MAAM,CAAA;CAAE,GAAG,SAAS,CAOrF"}