@aarwitz/tapp 0.16.5 → 0.17.0-rc.10

package/mcp-server/src/ci-setup.js

@@ -15,7 +15,7 @@ export function targetSlug(value) {
   return slug || "target";
 }
 
-export function selectApplicationTarget(model, { platform = "", target = "" } = {}) {
+export function selectApplicationTarget(model, { platform = "", target = "", useDefault = false } = {}) {
   if (model?.kind !== "tapp-application-model" || !Array.isArray(model.targets)) {
     throw new Error("Expected a Tapp application model; run tapp init first");
   }
@@ -26,6 +26,14 @@ export function selectApplicationTarget(model, { platform = "", target = "" } =
     const normalized = requested.replaceAll("\\", "/").replace(/^\.\//, "");
     candidates = candidates.filter((item) => [item.id, item.name, item.sourcePath].some((value) => String(value || "").replaceAll("\\", "/") === normalized));
   }
+  // Target-resolution ladder (ADR-0005 §5): explicit narrowing (above) → exactly one candidate →
+  // the model's recorded default target (opt-in: `explore` uses it, but the gate/baseline stay
+  // strict so CI never silently picks a target) → otherwise list the choices and stop.
+  if (candidates.length > 1 && useDefault && !requested) {
+    const def = String(model.application?.defaultTargetId || "").trim();
+    const chosen = def && candidates.find((item) => item.id === def);
+    if (chosen) return chosen;
+  }
   if (candidates.length !== 1) {
     const summary = candidates.length ? candidates : model.targets.filter((item) => !selectedPlatform || item.platform === selectedPlatform);
     throw new Error(candidates.length
@@ -35,6 +43,22 @@ export function selectApplicationTarget(model, { platform = "", target = "" } =
   return candidates[0];
 }
 
+// For a bare `tapp explore` in a repo: if the model's default target is a web target with a recorded
+// owned URL, return that URL so exploration hits it directly (the `init --url X` → `explore` path).
+// Targets that must be built/started from source need the prepare pipeline (wired separately), so
+// this returns null and the caller falls back to its normal target resolution.
+export function defaultWebExploreUrl(model) {
+  let target;
+  try {
+    target = selectApplicationTarget(model, { useDefault: true });
+  } catch {
+    return null; // no model, or ambiguous with no recorded default
+  }
+  if (target?.platform !== "web") return null;
+  const url = String(target.runtime?.ownedUrl || "");
+  return /^https?:\/\//i.test(url) ? url : null;
+}
+
 export function baselinePathForTarget(projectDir, target) {
   const root = fs.realpathSync(path.resolve(projectDir));
   return path.join(root, ".tapp", "baselines", target.platform, `${targetSlug(target.id)}.json`);
@@ -52,9 +76,15 @@ export function validateBaselineReport(report, { platform, targetId } = {}) {
   const reportTarget = String(report.targetKey || report.baselineIdentity?.targetId || "").trim();
   if (!reportTarget) throw new Error("Baseline source is missing its targetKey; Tapp will not guess which same-platform application produced the evidence");
   if (reportTarget !== targetId) throw new Error(`Baseline target '${reportTarget}' does not match application-model target '${targetId}'`);
-  if (report.inconclusive === true) throw new Error("An inconclusive run cannot become a trusted baseline");
-  if (report.verdict === "blocked") throw new Error("A blocked run cannot become a trusted baseline");
-  if (!report.gate || report.gate.failed !== false) throw new Error("Baseline creation requires a successful portable gate report (gate.failed must be false)");
+  // A trusted baseline must be a clean PASS. The gate outcome is the single authoritative signal
+  // (ADR-0005) — reject fail, inconclusive, error, or a missing/unknown outcome, naming which.
+  const outcome = report.gate?.outcome;
+  if (outcome !== "pass") {
+    const why = outcome === "inconclusive" ? "An inconclusive run"
+      : outcome === "fail" ? "A failing run"
+      : `A non-passing run (${outcome || "no gate outcome"})`;
+    throw new Error(`${why} cannot become a trusted baseline; only a passing gate run can`);
+  }
   for (const collection of ["flows", "scenarios", "contracts"]) {
     const failed = (report[collection] || []).filter((item) => item.passed !== true);
     if (failed.length) throw new Error(`Baseline source contains ${failed.length} failed ${collection}`);
@@ -64,7 +94,7 @@ export function validateBaselineReport(report, { platform, targetId } = {}) {
     platform,
     targetId,
     conclusive: true,
-    verdict: report.verdict,
+    outcome: report.gate?.outcome ?? null,
     screensExplored: Number(report.screensExplored || report.screens.length),
     actionsPerformed: Number(report.actionsPerformed || 0),
     suite: {

package/mcp-server/src/enrich.js

@@ -40,7 +40,7 @@ export async function enrichFindings(findings, { backend, callModel, screens = [
         `Title: ${f.title}`;
       try {
         const res = await Promise.race([
-          callModel(backend, { system, userText, model: process.env.TAPP_FINDING_MODEL || process.env.AUTOTAP_FINDING_MODEL || "claude-haiku-4-5-20251001", maxTokens: 300 }),
+          callModel(backend, { system, userText, model: process.env.TAPP_FINDING_MODEL || "claude-haiku-4-5-20251001", maxTokens: 300 }),
           new Promise((r) => setTimeout(() => r({ error: "timeout" }), TIMEOUT_MS)),
         ]);
         const parsed = res && !res.error ? parseEnrichment(res.text) : null;

package/mcp-server/src/html-report.js

@@ -8,12 +8,28 @@
 
 import fs from "fs";
 import path from "path";
-import { buildQaReport, qaScoreLabel, verdictBadge } from "./report.js";
+import { buildQaReport, observationBadge, observationSummary } from "./report.js";
 
 const SEV_COLOR = { critical: "#cf222e", high: "#bc4c00", medium: "#9a6700", low: "#57606a" };
 
 const esc = (s) => String(s ?? "").replace(/[&<>"]/g, (c) => ({ "&": "&amp;", "<": "&lt;", ">": "&gt;", '"': "&quot;" }[c]));
 
+const VALID_PLATFORMS = new Set(["web", "android", "ios"]);
+
+// Recover a capture's platform when no in-memory report is available. Prefer the run's persisted
+// metadata (ui-map.json → app.platforms) — it is authoritative and survives a renamed folder — and
+// fall back to the capture-id prefix (web-*/android-*, ios unprefixed) only for legacy captures that
+// predate the map. Defaults to ios if nothing is resolvable, matching buildQaReport's own default.
+export function capturePlatform(captureDir) {
+  try {
+    const map = JSON.parse(fs.readFileSync(path.join(captureDir, "ui-map.json"), "utf8"));
+    const fromMap = (map?.app?.platforms || []).find((p) => VALID_PLATFORMS.has(p));
+    if (fromMap) return fromMap;
+  } catch { /* no map, unreadable, or no valid platform — fall back to the id prefix */ }
+  const base = path.basename(captureDir);
+  return base.startsWith("web-") ? "web" : base.startsWith("android-") ? "android" : "ios";
+}
+
 // Two capture layouts exist: web runs write state_*.png at the capture root; iOS runs
 // export XCUITest attachments into screenshots/ as UUID files with a manifest carrying
 // the human-readable state_N_<Screen> names.
@@ -49,8 +65,12 @@ function collectShots(captureDir) {
   }
 }
 
-export function writeHtmlReport(captureDir, { report, label = "" } = {}) {
-  const r = report || buildQaReport(path.join(captureDir, "ocqa-markers.txt"));
+export function writeHtmlReport(captureDir, { report, label = "", recordingWarning = "" } = {}) {
+  // Rebuild path (e.g. `tapp report`): no report object is passed, so buildQaReport would default to
+  // native — rendering a web/android page with the wrong "Checked / Not checked" scope, overclaiming
+  // native checks it never ran. Recover the run's real platform (metadata first, id prefix as a
+  // legacy fallback). When a report object IS passed (explore/init), it is already correct.
+  const r = report || buildQaReport(path.join(captureDir, "ocqa-markers.txt"), { platform: capturePlatform(captureDir) });
   if (!r) return null;
 
   const shots = collectShots(captureDir);
@@ -78,7 +98,16 @@ export function writeHtmlReport(captureDir, { report, label = "" } = {}) {
 
   const videoHtml = video
     ? `<h2>Recording — the full exploration</h2>\n<video controls preload="metadata" style="width:100%;border:1px solid #d0d7de;border-radius:8px" src="${esc(video)}"></video>`
-    : "";
+    : recordingWarning
+      ? `<h2>Recording</h2>\n<p class="warning">Unavailable: ${esc(recordingWarning)}. Screenshots were still captured.</p>`
+      : "";
+
+  const scopeList = (items) => (items || []).map((item) => `<li>${esc(item)}</li>`).join("\n");
+  const scopeHtml = `<section class="scope">
+<div><h2>Checked this run</h2><ul>${scopeList(r.checkedFor) || "<li class='dim'>No automated checks completed.</li>"}</ul></div>
+<div><h2>Not checked this run</h2><ul>${scopeList(r.notChecked) || "<li class='dim'>No additional limitations recorded.</li>"}</ul></div>
+${r.conditionsNotReached?.length ? `<div><h2>Conditions not reached</h2><ul>${scopeList(r.conditionsNotReached)}</ul></div>` : ""}
+</section>`;
 
   const html = `<!doctype html>
 <html lang="en">
@@ -90,6 +119,10 @@ export function writeHtmlReport(captureDir, { report, label = "" } = {}) {
   h1 { font-size: 1.6rem; margin-bottom: 0.2rem; }
   .meta { color: #57606a; margin-bottom: 1.2rem; }
   .headline { background: #f6f8fa; border-radius: 8px; padding: 0.9rem 1.1rem; margin: 1rem 0; }
+  .warning { background: #fff8c5; border: 1px solid #d4a72c; border-radius: 8px; padding: 0.75rem 1rem; }
+  .scope { display: grid; grid-template-columns: repeat(auto-fit, minmax(260px, 1fr)); gap: 0.8rem 1.5rem; }
+  .scope h2 { font-size: 1.05rem; margin-bottom: 0.35rem; }
+  .scope ul { margin-top: 0; padding-left: 1.2rem; }
   .sev { color: #fff; border-radius: 4px; padding: 0.05rem 0.45rem; font-size: 0.78rem; font-weight: 600; margin-right: 0.4rem; }
   ul.findings { padding-left: 1.1rem; } ul.findings li { margin-bottom: 0.6rem; }
   .ai { color: #57606a; font-size: 0.88rem; margin: 0.15rem 0 0 0.2rem; }
@@ -101,10 +134,11 @@ export function writeHtmlReport(captureDir, { report, label = "" } = {}) {
 </style>
 </head>
 <body>
-<h1>${esc(verdictBadge(r))} <span class="dim">· ${esc(qaScoreLabel(r))}</span></h1>
-<div class="meta">${esc(label)} · ${r.screensExplored} screens · ${r.actionsPerformed} actions · ${r.findingCounts.total} finding(s)</div>
-${r.platform === "web" ? `<div class="meta">Verdict basis: ${r.verdictFindingCounts?.total || 0} deterministic finding(s); ${r.sampledFindingCounts?.total || 0} sampled probe finding(s) are advisory.</div>` : ""}
+<h1>${esc(observationBadge(r))} <span class="dim">· ${esc(observationSummary(r))}</span></h1>
+<div class="meta">${esc(label)} · evidence page (observation, not a release decision)</div>
+${r.platform === "web" ? `<div class="meta">Deterministic basis: ${r.deterministicFindingCounts?.total || 0} deterministic finding(s); ${r.sampledFindingCounts?.total || 0} sampled probe finding(s) are advisory.</div>` : ""}
 <div class="headline">${esc(r.headline)}</div>
+${scopeHtml}
 <h2>Findings</h2>
 <ul class="findings">
 ${findingsHtml}