nodebench-mcp 2.22.0 → 2.26.0

package/dist/tools/localDashboardTools.js

@@ -0,0 +1,332 @@
+/**
+ * Local Dashboard MCP Tools
+ *
+ * 5 tools for operating the local Daily Brief dashboard via Claude Code.
+ * All tools read from local SQLite — zero network dependency.
+ */
+import { getDb, genId } from "../db.js";
+import { getBriefDashboardUrl, startBriefDashboardServer } from "../dashboard/briefServer.js";
+function safeParseJson(s) {
+    if (!s)
+        return null;
+    try {
+        return JSON.parse(s);
+    }
+    catch {
+        return s;
+    }
+}
+export const localDashboardTools = [
+    // ── sync_daily_brief ────────────────────────────────────────────────
+    {
+        name: "sync_daily_brief",
+        description: "Sync daily brief + narrative data from Convex to local SQLite. Requires CONVEX_SITE_URL and MCP_SECRET environment variables. Returns sync summary with row counts and timing.",
+        inputSchema: {
+            type: "object",
+            properties: {
+                days: {
+                    type: "number",
+                    description: "Number of days to sync (default: 7)",
+                },
+                force: {
+                    type: "boolean",
+                    description: "Force re-sync even if data already exists (default: false)",
+                },
+            },
+        },
+        handler: async (args) => {
+            const siteUrl = process.env.CONVEX_SITE_URL || process.env.VITE_CONVEX_URL;
+            if (!siteUrl) {
+                return {
+                    error: true,
+                    message: "Missing CONVEX_SITE_URL or VITE_CONVEX_URL environment variable",
+                    setupInstructions: "Set CONVEX_SITE_URL in .env.local or as an environment variable",
+                };
+            }
+            const secret = process.env.MCP_SECRET;
+            if (!secret) {
+                return {
+                    error: true,
+                    message: "Missing MCP_SECRET environment variable",
+                    setupInstructions: "Set MCP_SECRET via: npx convex env set MCP_SECRET <value>",
+                };
+            }
+            const db = getDb();
+            const syncId = genId("sync");
+            const startMs = Date.now();
+            db.prepare("INSERT INTO sync_runs (id, status) VALUES (?, 'running')").run(syncId);
+            const counts = {};
+            try {
+                // Fetch latest dashboard snapshot
+                const snapRes = await fetch(`${siteUrl.replace(/\/$/, "")}/api/mcpGateway`, {
+                    method: "POST",
+                    headers: { "Content-Type": "application/json", "x-mcp-secret": secret },
+                    body: JSON.stringify({ fn: "getLatestDashboardSnapshot", args: {} }),
+                });
+                const snapData = await snapRes.json();
+                if (snapData.success && snapData.data) {
+                    const s = snapData.data;
+                    db.prepare(`
+            INSERT OR REPLACE INTO brief_snapshots (id, date_string, generated_at, dashboard_metrics, source_summary, version)
+            VALUES (?, ?, ?, ?, ?, 1)
+          `).run(s._id ?? genId("snap"), s.dateString ?? new Date().toISOString().slice(0, 10), s.generatedAt ?? Date.now(), JSON.stringify(s.dashboardMetrics ?? {}), JSON.stringify(s.sourceSummary ?? null));
+                    counts.brief_snapshots = 1;
+                }
+                // Fetch narrative threads
+                const threadRes = await fetch(`${siteUrl.replace(/\/$/, "")}/api/mcpGateway`, {
+                    method: "POST",
+                    headers: { "Content-Type": "application/json", "x-mcp-secret": secret },
+                    body: JSON.stringify({ fn: "getPublicThreads", args: { limit: 100 } }),
+                });
+                const threadData = await threadRes.json();
+                if (threadData.success && Array.isArray(threadData.data)) {
+                    const upsert = db.prepare(`
+            INSERT OR REPLACE INTO narrative_threads_local
+            (id, thread_id, name, slug, thesis, counter_thesis, entity_keys, topic_tags,
+             current_phase, first_event_at, latest_event_at, event_count, plot_twist_count, quality)
+            VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)
+          `);
+                    const tx = db.transaction((items) => {
+                        for (const t of items) {
+                            upsert.run(t._id, t.threadId ?? t._id, t.name, t.slug ?? "", t.thesis, t.counterThesis ?? null, JSON.stringify(t.entityKeys ?? []), JSON.stringify(t.topicTags ?? []), t.currentPhase, t.firstEventAt ?? null, t.latestEventAt ?? null, t.eventCount ?? 0, t.plotTwistCount ?? 0, JSON.stringify(t.quality ?? null));
+                        }
+                    });
+                    tx(threadData.data);
+                    counts.narrative_threads = threadData.data.length;
+                }
+                const durationMs = Date.now() - startMs;
+                db.prepare(`
+          UPDATE sync_runs SET status = 'success', completed_at = datetime('now'),
+            tables_synced = ?, duration_ms = ?
+          WHERE id = ?
+        `).run(JSON.stringify(counts), durationMs, syncId);
+                return {
+                    success: true,
+                    syncId,
+                    durationMs,
+                    counts,
+                    dashboardUrl: getBriefDashboardUrl(),
+                };
+            }
+            catch (err) {
+                const durationMs = Date.now() - startMs;
+                db.prepare(`
+          UPDATE sync_runs SET status = 'error', completed_at = datetime('now'),
+            error = ?, duration_ms = ?
+          WHERE id = ?
+        `).run(err.message, durationMs, syncId);
+                return { error: true, message: err.message, syncId, durationMs };
+            }
+        },
+    },
+    // ── get_daily_brief_summary ─────────────────────────────────────────
+    {
+        name: "get_daily_brief_summary",
+        description: "Get the latest daily brief summary from local SQLite. Returns dashboard metrics, features, and source summary. No network needed — reads cached data.",
+        inputSchema: {
+            type: "object",
+            properties: {
+                date: {
+                    type: "string",
+                    description: "Date string (YYYY-MM-DD). Omit for latest.",
+                },
+            },
+        },
+        handler: async (args) => {
+            const db = getDb();
+            const row = args.date
+                ? db.prepare("SELECT * FROM brief_snapshots WHERE date_string = ? ORDER BY version DESC LIMIT 1").get(args.date)
+                : db.prepare("SELECT * FROM brief_snapshots ORDER BY generated_at DESC LIMIT 1").get();
+            if (!row) {
+                return {
+                    empty: true,
+                    message: args.date
+                        ? `No brief for ${args.date}. Run sync_daily_brief to fetch data.`
+                        : "No briefs synced yet. Run sync_daily_brief to fetch data.",
+                    tip: "Use sync_daily_brief tool or run: npm run local:sync",
+                };
+            }
+            return {
+                dateString: row.date_string,
+                generatedAt: row.generated_at,
+                version: row.version,
+                dashboardMetrics: safeParseJson(row.dashboard_metrics),
+                sourceSummary: safeParseJson(row.source_summary),
+                syncedAt: row.synced_at,
+                dashboardUrl: getBriefDashboardUrl(),
+            };
+        },
+    },
+    // ── get_narrative_status ────────────────────────────────────────────
+    {
+        name: "get_narrative_status",
+        description: "Get narrative thread status from local SQLite. Returns threads grouped by phase (emerging, escalating, climax, resolution, dormant) with event counts. No network needed.",
+        inputSchema: {
+            type: "object",
+            properties: {
+                phase: {
+                    type: "string",
+                    enum: ["emerging", "escalating", "climax", "resolution", "dormant"],
+                    description: "Filter by phase (optional)",
+                },
+                limit: {
+                    type: "number",
+                    description: "Max threads to return (default: 20)",
+                },
+            },
+        },
+        handler: async (args) => {
+            const db = getDb();
+            const limit = Math.min(args.limit ?? 20, 100);
+            const query = args.phase
+                ? "SELECT * FROM narrative_threads_local WHERE current_phase = ? ORDER BY latest_event_at DESC LIMIT ?"
+                : "SELECT * FROM narrative_threads_local ORDER BY latest_event_at DESC LIMIT ?";
+            const rows = args.phase
+                ? db.prepare(query).all(args.phase, limit)
+                : db.prepare(query).all(limit);
+            if (rows.length === 0) {
+                return {
+                    empty: true,
+                    message: "No narrative threads synced. Run sync_daily_brief to fetch data.",
+                };
+            }
+            // Group by phase
+            const grouped = {};
+            for (const r of rows) {
+                const phase = r.current_phase || "dormant";
+                if (!grouped[phase])
+                    grouped[phase] = [];
+                grouped[phase].push({
+                    id: r.id,
+                    name: r.name,
+                    thesis: r.thesis,
+                    eventCount: r.event_count,
+                    plotTwistCount: r.plot_twist_count,
+                    latestEventAt: r.latest_event_at,
+                    entityKeys: safeParseJson(r.entity_keys),
+                    topicTags: safeParseJson(r.topic_tags),
+                    quality: safeParseJson(r.quality),
+                });
+            }
+            // Phase distribution
+            const distribution = {};
+            try {
+                const dist = db.prepare("SELECT current_phase, COUNT(*) as count FROM narrative_threads_local GROUP BY current_phase").all();
+                for (const d of dist)
+                    distribution[d.current_phase] = d.count;
+            }
+            catch { /* table may not exist */ }
+            return {
+                totalThreads: rows.length,
+                phaseDistribution: distribution,
+                threads: grouped,
+                dashboardUrl: getBriefDashboardUrl(),
+            };
+        },
+    },
+    // ── get_ops_dashboard ───────────────────────────────────────────────
+    {
+        name: "get_ops_dashboard",
+        description: "Get operational dashboard status from local SQLite. Returns last sync info, tool call frequency, active verification cycles, and data counts. No network needed.",
+        inputSchema: {
+            type: "object",
+            properties: {},
+        },
+        handler: async () => {
+            const db = getDb();
+            // Last sync
+            let lastSync = null;
+            try {
+                const row = db.prepare("SELECT * FROM sync_runs ORDER BY started_at DESC LIMIT 1").get();
+                if (row) {
+                    lastSync = {
+                        ...row,
+                        tables_synced: safeParseJson(row.tables_synced),
+                    };
+                }
+            }
+            catch { /* table may not exist */ }
+            // Tool call frequency (last 24h)
+            let toolStats = [];
+            try {
+                toolStats = db.prepare(`
+          SELECT tool_name, COUNT(*) as count, AVG(duration_ms) as avg_duration,
+            SUM(CASE WHEN status = 'error' THEN 1 ELSE 0 END) as errors
+          FROM tool_call_log
+          WHERE created_at > datetime('now', '-1 day')
+          GROUP BY tool_name ORDER BY count DESC LIMIT 15
+        `).all();
+            }
+            catch { /* table may not exist */ }
+            // Active verification cycles
+            let activeCycles = [];
+            try {
+                activeCycles = db.prepare(`
+          SELECT id, title, status, created_at FROM verification_cycles
+          WHERE status NOT IN ('completed', 'abandoned')
+          ORDER BY created_at DESC LIMIT 5
+        `).all();
+            }
+            catch { /* table may not exist */ }
+            // Data counts
+            let briefCount = 0, threadCount = 0, eventCount = 0;
+            try {
+                briefCount = db.prepare("SELECT COUNT(*) as c FROM brief_snapshots").get()?.c ?? 0;
+                threadCount = db.prepare("SELECT COUNT(*) as c FROM narrative_threads_local").get()?.c ?? 0;
+                eventCount = db.prepare("SELECT COUNT(*) as c FROM narrative_events_local").get()?.c ?? 0;
+            }
+            catch { /* tables may not exist */ }
+            // Privacy/audience stats
+            let privacyMode = null;
+            try {
+                const todayEvents = db.prepare(`
+          SELECT COUNT(*) as count,
+            SUM(CASE WHEN is_public = 1 THEN 1 ELSE 0 END) as public_triggers
+          FROM audience_events WHERE created_at > datetime('now', '-1 day')
+        `).get();
+                if (todayEvents && todayEvents.count > 0) {
+                    privacyMode = {
+                        triggeredToday: todayEvents.public_triggers ?? 0,
+                        totalEvents: todayEvents.count,
+                    };
+                }
+            }
+            catch { /* table may not exist */ }
+            return {
+                lastSync,
+                toolCallFrequency: toolStats,
+                activeCycles,
+                dataCounts: { briefs: briefCount, threads: threadCount, events: eventCount },
+                privacyMode,
+                dashboardUrl: getBriefDashboardUrl(),
+            };
+        },
+    },
+    // ── open_local_dashboard ────────────────────────────────────────────
+    {
+        name: "open_local_dashboard",
+        description: "Start the local Daily Brief dashboard server if needed, and return the URL. The dashboard shows Brief metrics, Narrative thread lanes, and Ops status — all from local SQLite.",
+        inputSchema: {
+            type: "object",
+            properties: {},
+        },
+        handler: async () => {
+            let url = getBriefDashboardUrl();
+            if (!url) {
+                try {
+                    const port = await startBriefDashboardServer(getDb(), 6275);
+                    url = `http://127.0.0.1:${port}`;
+                }
+                catch (err) {
+                    return { error: true, message: `Failed to start dashboard: ${err.message}` };
+                }
+            }
+            return {
+                url,
+                views: ["Brief (metrics, features, sources)", "Narrative (thread lanes by phase)", "Ops (sync status, tool frequency)"],
+                tip: "Open the URL in a browser to see the dashboard. Data auto-refreshes every 30s.",
+            };
+        },
+    },
+];
+//# sourceMappingURL=localDashboardTools.js.map

package/dist/tools/localDashboardTools.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"localDashboardTools.js","sourceRoot":"","sources":["../../src/tools/localDashboardTools.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAGH,OAAO,EAAE,KAAK,EAAE,KAAK,EAAE,MAAM,UAAU,CAAC;AACxC,OAAO,EAAE,oBAAoB,EAAE,yBAAyB,EAAE,MAAM,6BAA6B,CAAC;AAE9F,SAAS,aAAa,CAAC,CAA4B;IACjD,IAAI,CAAC,CAAC;QAAE,OAAO,IAAI,CAAC;IACpB,IAAI,CAAC;QAAC,OAAO,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC;IAAC,CAAC;IAAC,MAAM,CAAC;QAAC,OAAO,CAAC,CAAC;IAAC,CAAC;AACnD,CAAC;AAED,MAAM,CAAC,MAAM,mBAAmB,GAAc;IAC5C,uEAAuE;IACvE;QACE,IAAI,EAAE,kBAAkB;QACxB,WAAW,EACT,gLAAgL;QAClL,WAAW,EAAE;YACX,IAAI,EAAE,QAAQ;YACd,UAAU,EAAE;gBACV,IAAI,EAAE;oBACJ,IAAI,EAAE,QAAQ;oBACd,WAAW,EAAE,qCAAqC;iBACnD;gBACD,KAAK,EAAE;oBACL,IAAI,EAAE,SAAS;oBACf,WAAW,EAAE,4DAA4D;iBAC1E;aACF;SACF;QACD,OAAO,EAAE,KAAK,EAAE,IAAI,EAAE,EAAE;YACtB,MAAM,OAAO,GAAG,OAAO,CAAC,GAAG,CAAC,eAAe,IAAI,OAAO,CAAC,GAAG,CAAC,eAAe,CAAC;YAC3E,IAAI,CAAC,OAAO,EAAE,CAAC;gBACb,OAAO;oBACL,KAAK,EAAE,IAAI;oBACX,OAAO,EAAE,iEAAiE;oBAC1E,iBAAiB,EAAE,iEAAiE;iBACrF,CAAC;YACJ,CAAC;YAED,MAAM,MAAM,GAAG,OAAO,CAAC,GAAG,CAAC,UAAU,CAAC;YACtC,IAAI,CAAC,MAAM,EAAE,CAAC;gBACZ,OAAO;oBACL,KAAK,EAAE,IAAI;oBACX,OAAO,EAAE,yCAAyC;oBAClD,iBAAiB,EAAE,2DAA2D;iBAC/E,CAAC;YACJ,CAAC;YAED,MAAM,EAAE,GAAG,KAAK,EAAE,CAAC;YACnB,MAAM,MAAM,GAAG,KAAK,CAAC,MAAM,CAAC,CAAC;YAC7B,MAAM,OAAO,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;YAC3B,EAAE,CAAC,OAAO,CAAC,0DAA0D,CAAC,CAAC,GAAG,CAAC,MAAM,CAAC,CAAC;YAEnF,MAAM,MAAM,GAA2B,EAAE,CAAC;YAE1C,IAAI,CAAC;gBACH,kCAAkC;gBAClC,MAAM,OAAO,GAAG,MAAM,KAAK,CAAC,GAAG,OAAO,CAAC,OAAO,CAAC,KAAK,EAAE,EAAE,CAAC,iBAAiB,EAAE;oBAC1E,MAAM,EAAE,MAAM;oBACd,OAAO,EAAE,EAAE,cAAc,EAAE,kBAAkB,EAAE,cAAc,EAAE,MAAM,EAAE;oBACvE,IAAI,EAAE,IAAI,CAAC,SAAS,CAAC,EAAE,EAAE,EAAE,4BAA4B,EAAE,IAAI,EAAE,EAAE,EAAE,CAAC;iBACrE,CAAC,CAAC;gBACH,MAAM,QAAQ,GAAQ,MAAM,OAAO,CAAC,IAAI,EAAE,CAAC;gBAC3C,IAAI,QAAQ,CAAC,OAAO,IAAI,QAAQ,CAAC,IAAI,EAAE,CAAC;oBACtC,MAAM,CAAC,GAAG,QAAQ,CAAC,IAAI,CAAC;oBACxB,EAAE,CAAC,OAAO,CAAC;;;WAGV,CAAC,CAAC,GAAG,CACJ,CAAC,CAAC,GAAG,IAAI,KAAK,CAAC,MAAM,CAAC,EACtB,CAAC,CAAC,UAAU,IAAI,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC,KAAK,CAAC,CAAC,EAAE,EAAE,CAAC,EACrD,CAAC,CAAC,WAAW,IAAI,IAAI,CAAC,GAAG,EAAE,EAC3B,IAAI,CAAC,SAAS,CAAC,CAAC,CAAC,gBAAgB,IAAI,EAAE,CAAC,EACxC,IAAI,CAAC,SAAS,CAAC,CAAC,CAAC,aAAa,IAAI,IAAI,CAAC,CACxC,CAAC;oBACF,MAAM,CAAC,eAAe,GAAG,CAAC,CAAC;gBAC7B,CAAC;gBAED,0BAA0B;gBAC1B,MAAM,SAAS,GAAG,MAAM,KAAK,CAAC,GAAG,OAAO,CAAC,OAAO,CAAC,KAAK,EAAE,EAAE,CAAC,iBAAiB,EAAE;oBAC5E,MAAM,EAAE,MAAM;oBACd,OAAO,EAAE,EAAE,cAAc,EAAE,kBAAkB,EAAE,cAAc,EAAE,MAAM,EAAE;oBACvE,IAAI,EAAE,IAAI,CAAC,SAAS,CAAC,EAAE,EAAE,EAAE,kBAAkB,EAAE,IAAI,EAAE,EAAE,KAAK,EAAE,GAAG,EAAE,EAAE,CAAC;iBACvE,CAAC,CAAC;gBACH,MAAM,UAAU,GAAQ,MAAM,SAAS,CAAC,IAAI,EAAE,CAAC;gBAC/C,IAAI,UAAU,CAAC,OAAO,IAAI,KAAK,CAAC,OAAO,CAAC,UAAU,CAAC,IAAI,CAAC,EAAE,CAAC;oBACzD,MAAM,MAAM,GAAG,EAAE,CAAC,OAAO,CAAC;;;;;WAKzB,CAAC,CAAC;oBACH,MAAM,EAAE,GAAG,EAAE,CAAC,WAAW,CAAC,CAAC,KAAY,EAAE,EAAE;wBACzC,KAAK,MAAM,CAAC,IAAI,KAAK,EAAE,CAAC;4BACtB,MAAM,CAAC,GAAG,CACR,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC,QAAQ,IAAI,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC,IAAI,IAAI,EAAE,EAChD,CAAC,CAAC,MAAM,EAAE,CAAC,CAAC,aAAa,IAAI,IAAI,EACjC,IAAI,CAAC,SAAS,CAAC,CAAC,CAAC,UAAU,IAAI,EAAE,CAAC,EAAE,IAAI,CAAC,SAAS,CAAC,CAAC,CAAC,SAAS,IAAI,EAAE,CAAC,EACrE,CAAC,CAAC,YAAY,EAAE,CAAC,CAAC,YAAY,IAAI,IAAI,EAAE,CAAC,CAAC,aAAa,IAAI,IAAI,EAC/D,CAAC,CAAC,UAAU,IAAI,CAAC,EAAE,CAAC,CAAC,cAAc,IAAI,CAAC,EACxC,IAAI,CAAC,SAAS,CAAC,CAAC,CAAC,OAAO,IAAI,IAAI,CAAC,CAClC,CAAC;wBACJ,CAAC;oBACH,CAAC,CAAC,CAAC;oBACH,EAAE,CAAC,UAAU,CAAC,IAAI,CAAC,CAAC;oBACpB,MAAM,CAAC,iBAAiB,GAAG,UAAU,CAAC,IAAI,CAAC,MAAM,CAAC;gBACpD,CAAC;gBAED,MAAM,UAAU,GAAG,IAAI,CAAC,GAAG,EAAE,GAAG,OAAO,CAAC;gBACxC,EAAE,CAAC,OAAO,CAAC;;;;SAIV,CAAC,CAAC,GAAG,CAAC,IAAI,CAAC,SAAS,CAAC,MAAM,CAAC,EAAE,UAAU,EAAE,MAAM,CAAC,CAAC;gBAEnD,OAAO;oBACL,OAAO,EAAE,IAAI;oBACb,MAAM;oBACN,UAAU;oBACV,MAAM;oBACN,YAAY,EAAE,oBAAoB,EAAE;iBACrC,CAAC;YACJ,CAAC;YAAC,OAAO,GAAQ,EAAE,CAAC;gBAClB,MAAM,UAAU,GAAG,IAAI,CAAC,GAAG,EAAE,GAAG,OAAO,CAAC;gBACxC,EAAE,CAAC,OAAO,CAAC;;;;SAIV,CAAC,CAAC,GAAG,CAAC,GAAG,CAAC,OAAO,EAAE,UAAU,EAAE,MAAM,CAAC,CAAC;gBACxC,OAAO,EAAE,KAAK,EAAE,IAAI,EAAE,OAAO,EAAE,GAAG,CAAC,OAAO,EAAE,MAAM,EAAE,UAAU,EAAE,CAAC;YACnE,CAAC;QACH,CAAC;KACF;IAED,uEAAuE;IACvE;QACE,IAAI,EAAE,yBAAyB;QAC/B,WAAW,EACT,uJAAuJ;QACzJ,WAAW,EAAE;YACX,IAAI,EAAE,QAAQ;YACd,UAAU,EAAE;gBACV,IAAI,EAAE;oBACJ,IAAI,EAAE,QAAQ;oBACd,WAAW,EAAE,4CAA4C;iBAC1D;aACF;SACF;QACD,OAAO,EAAE,KAAK,EAAE,IAAI,EAAE,EAAE;YACtB,MAAM,EAAE,GAAG,KAAK,EAAE,CAAC;YACnB,MAAM,GAAG,GAAG,IAAI,CAAC,IAAI;gBACnB,CAAC,CAAC,EAAE,CAAC,OAAO,CAAC,mFAAmF,CAAC,CAAC,GAAG,CAAC,IAAI,CAAC,IAAI,CAAQ;gBACvH,CAAC,CAAC,EAAE,CAAC,OAAO,CAAC,kEAAkE,CAAC,CAAC,GAAG,EAAS,CAAC;YAEhG,IAAI,CAAC,GAAG,EAAE,CAAC;gBACT,OAAO;oBACL,KAAK,EAAE,IAAI;oBACX,OAAO,EAAE,IAAI,CAAC,IAAI;wBAChB,CAAC,CAAC,gBAAgB,IAAI,CAAC,IAAI,uCAAuC;wBAClE,CAAC,CAAC,2DAA2D;oBAC/D,GAAG,EAAE,sDAAsD;iBAC5D,CAAC;YACJ,CAAC;YAED,OAAO;gBACL,UAAU,EAAE,GAAG,CAAC,WAAW;gBAC3B,WAAW,EAAE,GAAG,CAAC,YAAY;gBAC7B,OAAO,EAAE,GAAG,CAAC,OAAO;gBACpB,gBAAgB,EAAE,aAAa,CAAC,GAAG,CAAC,iBAAiB,CAAC;gBACtD,aAAa,EAAE,aAAa,CAAC,GAAG,CAAC,cAAc,CAAC;gBAChD,QAAQ,EAAE,GAAG,CAAC,SAAS;gBACvB,YAAY,EAAE,oBAAoB,EAAE;aACrC,CAAC;QACJ,CAAC;KACF;IAED,uEAAuE;IACvE;QACE,IAAI,EAAE,sBAAsB;QAC5B,WAAW,EACT,2KAA2K;QAC7K,WAAW,EAAE;YACX,IAAI,EAAE,QAAQ;YACd,UAAU,EAAE;gBACV,KAAK,EAAE;oBACL,IAAI,EAAE,QAAQ;oBACd,IAAI,EAAE,CAAC,UAAU,EAAE,YAAY,EAAE,QAAQ,EAAE,YAAY,EAAE,SAAS,CAAC;oBACnE,WAAW,EAAE,4BAA4B;iBAC1C;gBACD,KAAK,EAAE;oBACL,IAAI,EAAE,QAAQ;oBACd,WAAW,EAAE,qCAAqC;iBACnD;aACF;SACF;QACD,OAAO,EAAE,KAAK,EAAE,IAAI,EAAE,EAAE;YACtB,MAAM,EAAE,GAAG,KAAK,EAAE,CAAC;YACnB,MAAM,KAAK,GAAG,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC,KAAK,IAAI,EAAE,EAAE,GAAG,CAAC,CAAC;YAE9C,MAAM,KAAK,GAAG,IAAI,CAAC,KAAK;gBACtB,CAAC,CAAC,qGAAqG;gBACvG,CAAC,CAAC,6EAA6E,CAAC;YAElF,MAAM,IAAI,GAAG,IAAI,CAAC,KAAK;gBACrB,CAAC,CAAC,EAAE,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,GAAG,CAAC,IAAI,CAAC,KAAK,EAAE,KAAK,CAAU;gBACnD,CAAC,CAAC,EAAE,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,GAAG,CAAC,KAAK,CAAU,CAAC;YAE1C,IAAI,IAAI,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;gBACtB,OAAO;oBACL,KAAK,EAAE,IAAI;oBACX,OAAO,EAAE,kEAAkE;iBAC5E,CAAC;YACJ,CAAC;YAED,iBAAiB;YACjB,MAAM,OAAO,GAA0B,EAAE,CAAC;YAC1C,KAAK,MAAM,CAAC,IAAI,IAAI,EAAE,CAAC;gBACrB,MAAM,KAAK,GAAG,CAAC,CAAC,aAAa,IAAI,SAAS,CAAC;gBAC3C,IAAI,CAAC,OAAO,CAAC,KAAK,CAAC;oBAAE,OAAO,CAAC,KAAK,CAAC,GAAG,EAAE,CAAC;gBACzC,OAAO,CAAC,KAAK,CAAC,CAAC,IAAI,CAAC;oBAClB,EAAE,EAAE,CAAC,CAAC,EAAE;oBACR,IAAI,EAAE,CAAC,CAAC,IAAI;oBACZ,MAAM,EAAE,CAAC,CAAC,MAAM;oBAChB,UAAU,EAAE,CAAC,CAAC,WAAW;oBACzB,cAAc,EAAE,CAAC,CAAC,gBAAgB;oBAClC,aAAa,EAAE,CAAC,CAAC,eAAe;oBAChC,UAAU,EAAE,aAAa,CAAC,CAAC,CAAC,WAAW,CAAC;oBACxC,SAAS,EAAE,aAAa,CAAC,CAAC,CAAC,UAAU,CAAC;oBACtC,OAAO,EAAE,aAAa,CAAC,CAAC,CAAC,OAAO,CAAC;iBAClC,CAAC,CAAC;YACL,CAAC;YAED,qBAAqB;YACrB,MAAM,YAAY,GAA2B,EAAE,CAAC;YAChD,IAAI,CAAC;gBACH,MAAM,IAAI,GAAG,EAAE,CAAC,OAAO,CACrB,6FAA6F,CAC9F,CAAC,GAAG,EAAW,CAAC;gBACjB,KAAK,MAAM,CAAC,IAAI,IAAI;oBAAE,YAAY,CAAC,CAAC,CAAC,aAAa,CAAC,GAAG,CAAC,CAAC,KAAK,CAAC;YAChE,CAAC;YAAC,MAAM,CAAC,CAAC,yBAAyB,CAAC,CAAC;YAErC,OAAO;gBACL,YAAY,EAAE,IAAI,CAAC,MAAM;gBACzB,iBAAiB,EAAE,YAAY;gBAC/B,OAAO,EAAE,OAAO;gBAChB,YAAY,EAAE,oBAAoB,EAAE;aACrC,CAAC;QACJ,CAAC;KACF;IAED,uEAAuE;IACvE;QACE,IAAI,EAAE,mBAAmB;QACzB,WAAW,EACT,kKAAkK;QACpK,WAAW,EAAE;YACX,IAAI,EAAE,QAAQ;YACd,UAAU,EAAE,EAAE;SACf;QACD,OAAO,EAAE,KAAK,IAAI,EAAE;YAClB,MAAM,EAAE,GAAG,KAAK,EAAE,CAAC;YAEnB,YAAY;YACZ,IAAI,QAAQ,GAAQ,IAAI,CAAC;YACzB,IAAI,CAAC;gBACH,MAAM,GAAG,GAAG,EAAE,CAAC,OAAO,CAAC,0DAA0D,CAAC,CAAC,GAAG,EAAS,CAAC;gBAChG,IAAI,GAAG,EAAE,CAAC;oBACR,QAAQ,GAAG;wBACT,GAAG,GAAG;wBACN,aAAa,EAAE,aAAa,CAAC,GAAG,CAAC,aAAa,CAAC;qBAChD,CAAC;gBACJ,CAAC;YACH,CAAC;YAAC,MAAM,CAAC,CAAC,yBAAyB,CAAC,CAAC;YAErC,iCAAiC;YACjC,IAAI,SAAS,GAAU,EAAE,CAAC;YAC1B,IAAI,CAAC;gBACH,SAAS,GAAG,EAAE,CAAC,OAAO,CAAC;;;;;;SAMtB,CAAC,CAAC,GAAG,EAAW,CAAC;YACpB,CAAC;YAAC,MAAM,CAAC,CAAC,yBAAyB,CAAC,CAAC;YAErC,6BAA6B;YAC7B,IAAI,YAAY,GAAU,EAAE,CAAC;YAC7B,IAAI,CAAC;gBACH,YAAY,GAAG,EAAE,CAAC,OAAO,CAAC;;;;SAIzB,CAAC,CAAC,GAAG,EAAW,CAAC;YACpB,CAAC;YAAC,MAAM,CAAC,CAAC,yBAAyB,CAAC,CAAC;YAErC,cAAc;YACd,IAAI,UAAU,GAAG,CAAC,EAAE,WAAW,GAAG,CAAC,EAAE,UAAU,GAAG,CAAC,CAAC;YACpD,IAAI,CAAC;gBACH,UAAU,GAAI,EAAE,CAAC,OAAO,CAAC,2CAA2C,CAAC,CAAC,GAAG,EAAU,EAAE,CAAC,IAAI,CAAC,CAAC;gBAC5F,WAAW,GAAI,EAAE,CAAC,OAAO,CAAC,mDAAmD,CAAC,CAAC,GAAG,EAAU,EAAE,CAAC,IAAI,CAAC,CAAC;gBACrG,UAAU,GAAI,EAAE,CAAC,OAAO,CAAC,kDAAkD,CAAC,CAAC,GAAG,EAAU,EAAE,CAAC,IAAI,CAAC,CAAC;YACrG,CAAC;YAAC,MAAM,CAAC,CAAC,0BAA0B,CAAC,CAAC;YAEtC,yBAAyB;YACzB,IAAI,WAAW,GAAQ,IAAI,CAAC;YAC5B,IAAI,CAAC;gBACH,MAAM,WAAW,GAAG,EAAE,CAAC,OAAO,CAAC;;;;SAI9B,CAAC,CAAC,GAAG,EAAS,CAAC;gBAChB,IAAI,WAAW,IAAI,WAAW,CAAC,KAAK,GAAG,CAAC,EAAE,CAAC;oBACzC,WAAW,GAAG;wBACZ,cAAc,EAAE,WAAW,CAAC,eAAe,IAAI,CAAC;wBAChD,WAAW,EAAE,WAAW,CAAC,KAAK;qBAC/B,CAAC;gBACJ,CAAC;YACH,CAAC;YAAC,MAAM,CAAC,CAAC,yBAAyB,CAAC,CAAC;YAErC,OAAO;gBACL,QAAQ;gBACR,iBAAiB,EAAE,SAAS;gBAC5B,YAAY;gBACZ,UAAU,EAAE,EAAE,MAAM,EAAE,UAAU,EAAE,OAAO,EAAE,WAAW,EAAE,MAAM,EAAE,UAAU,EAAE;gBAC5E,WAAW;gBACX,YAAY,EAAE,oBAAoB,EAAE;aACrC,CAAC;QACJ,CAAC;KACF;IAED,uEAAuE;IACvE;QACE,IAAI,EAAE,sBAAsB;QAC5B,WAAW,EACT,gLAAgL;QAClL,WAAW,EAAE;YACX,IAAI,EAAE,QAAQ;YACd,UAAU,EAAE,EAAE;SACf;QACD,OAAO,EAAE,KAAK,IAAI,EAAE;YAClB,IAAI,GAAG,GAAG,oBAAoB,EAAE,CAAC;YACjC,IAAI,CAAC,GAAG,EAAE,CAAC;gBACT,IAAI,CAAC;oBACH,MAAM,IAAI,GAAG,MAAM,yBAAyB,CAAC,KAAK,EAAE,EAAE,IAAI,CAAC,CAAC;oBAC5D,GAAG,GAAG,oBAAoB,IAAI,EAAE,CAAC;gBACnC,CAAC;gBAAC,OAAO,GAAQ,EAAE,CAAC;oBAClB,OAAO,EAAE,KAAK,EAAE,IAAI,EAAE,OAAO,EAAE,8BAA8B,GAAG,CAAC,OAAO,EAAE,EAAE,CAAC;gBAC/E,CAAC;YACH,CAAC;YACD,OAAO;gBACL,GAAG;gBACH,KAAK,EAAE,CAAC,oCAAoC,EAAE,mCAAmC,EAAE,mCAAmC,CAAC;gBACvH,GAAG,EAAE,gFAAgF;aACtF,CAAC;QACJ,CAAC;KACF;CACF,CAAC"}

package/dist/tools/metaTools.js

@@ -211,6 +211,60 @@ const METHODOLOGY_CONTENT = {
             bundle_analysis: "npm run perf:bundle",
         },
     },
+    flywheel_ui_dogfood: {
+        title: "Flywheel Mode: UI Dogfood + Walkthrough Video QA",
+        description: "Continuous UI verification loop: capture a full walkthrough video + screenshots, run Gemini-based video/screenshot QA, fix root causes (5-whys), and re-capture evidence until the UI is stable, accessible, and performant. Do not stop at 'build passes' — ship dogfood evidence.",
+        steps: [
+            {
+                step: 1,
+                name: "Launch + Trajectory",
+                description: "Start the dev server and publish a visible trajectory link in the UI within 60 seconds. The trajectory is the audit trail of iterations and evidence.",
+                tools: ["record_learning"],
+                action: "Start dev server. Ensure `/dogfood` shows the latest manifest + walkthrough. Record a learning if trajectory/evidence is missing.",
+            },
+            {
+                step: 2,
+                name: "Capture Evidence (Scribe + Video)",
+                description: "Run a full route walkthrough capture (screens + interactions) so issues are visible and replayable.",
+                tools: ["capture_responsive_suite", "capture_ui_screenshot"],
+                action: "Run `npm run dogfood:full:local` to generate `public/dogfood/*` (gallery + walkthrough video + frames + scribe).",
+            },
+            {
+                step: 3,
+                name: "Gemini QA (Video + Screens)",
+                description: "Use Gemini video/screenshot understanding to flag UI regressions: flashes, layout shift, focus traps, broken empty states, and performance jank.",
+                tools: ["analyze_ui_screenshot"],
+                action: "Run `npm run dogfood:qa:gemini` and review results in `/dogfood`. Convert actionable issues into deterministic fixes (no band-aids).",
+            },
+            {
+                step: 4,
+                name: "Root Cause + Fix (5 Whys)",
+                description: "At each checkpoint, immediately seek issues and follow symptom → intermediate state → root cause. Fix the cause so the symptom is impossible.",
+                tools: ["run_closed_loop", "record_learning"],
+                action: "Do a 5-whys chain. Implement the smallest correct fix. Re-run `tsc` + build. Record learnings for any repeated failure mode.",
+            },
+            {
+                step: 5,
+                name: "Motion Safety (Seizure / Flash Policy)",
+                description: "Remove or soften high-contrast flashes (large-area pulses/fades). Ship stable backgrounds and subtle loading states. Always honor prefers-reduced-motion.",
+                tools: ["run_quality_gate"],
+                action: "FAIL if any route shows full-viewport flashing/pulsing/fade transitions. Prefer per-view fallbacks and non-animated skeletons for large surfaces.",
+            },
+            {
+                step: 6,
+                name: "Poll + Adapt",
+                description: "Poll every 60 seconds. After 3 consecutive failures on the same issue, change strategy (instrument, isolate, or rollback).",
+                tools: ["record_learning"],
+                action: "Keep iterating until the dogfood artifacts + QA runs show no new P0/P1 issues.",
+            },
+        ],
+        commands: {
+            capture: "npm run dogfood:full:local",
+            gemini_qa: "npm run dogfood:qa:gemini",
+            build: "npm run build",
+            typecheck: "npx tsc --noEmit",
+        },
+    },
     agentic_vision: {
         title: "Agentic Vision (AI-Powered Visual Verification)",
         description: "Use AI vision models to analyze UI screenshots programmatically. The Discover-Capture-Analyze-Manipulate-Iterate-Gate loop provides automated visual QA that goes beyond what rule-based checks can catch. Gemini with code execution provides the richest analysis (zoom, crop, compute within the model). Falls back to GPT-5-mini, Claude, or OpenRouter vision.",
@@ -1242,6 +1296,117 @@ const METHODOLOGY_CONTENT = {
             "For production: implement interrupt handling (user speaks while TTS is playing)",
         ],
     },
+    analyst_diagnostic: {
+        title: "Analyst Diagnostic — Root Cause Over Bandaids",
+        description: "Guide yourself like an analyst diagnosing the root cause, NOT a junior dev slapping on a bandaid. Mandatory for all bug work.",
+        steps: [
+            {
+                step: 1,
+                name: "Reproduce",
+                description: "Confirm the exact failure mode before touching any code.",
+                tools: ["search_all_knowledge"],
+                action: "Check if this root cause is already known. If yes, apply the known fix. If no, proceed to step 2.",
+            },
+            {
+                step: 2,
+                name: "Trace Upstream",
+                description: "Walk from symptom → intermediate state → root cause. Don't stop at the first error you see.",
+                tools: ["run_recon", "log_recon_finding"],
+                action: "Use recon tools to trace the chain. Log each intermediate finding.",
+            },
+            {
+                step: 3,
+                name: "Ask 'Why' 5 Times",
+                description: "Each answer should go one level deeper into the system. If you can't get to 5, you haven't found the root cause yet.",
+                action: "Document each 'why' level. The 5th answer is usually the real fix.",
+            },
+            {
+                step: 4,
+                name: "Fix the Cause",
+                description: "The right fix makes the symptom impossible, not just invisible.",
+                tools: ["resolve_gap"],
+                action: "Implement the fix. It should address the root cause from step 3, not the symptom from step 1.",
+            },
+            {
+                step: 5,
+                name: "Verify No Sideways Shift",
+                description: "Bandaids move bugs sideways. Check adjacent behavior to confirm the fix didn't introduce new issues.",
+                tools: ["log_test_result", "run_closed_loop"],
+                action: "Run tests. Check adjacent functionality. Confirm the bug can't recur.",
+            },
+            {
+                step: 6,
+                name: "Record the Root Cause",
+                description: "Document what you found so the next person doesn't re-discover it.",
+                tools: ["record_learning"],
+                action: "Call record_learning with the root cause, not just the fix.",
+            },
+        ],
+        redFlags: [
+            "Adding try/catch that swallows errors without understanding them",
+            "Adding ?. optional chaining to mask undefined instead of finding why it's undefined",
+            "Adding 'as any' to silence type errors instead of fixing the type mismatch",
+            "Adding timeouts/retries to paper over race conditions",
+            "Deleting a failing test instead of fixing the code it tests",
+            "'It works now' without understanding why it didn't before",
+        ],
+    },
+    scenario_testing: {
+        title: "Scenario-Based Testing — Real Human Behavior at Scale",
+        description: "Tests that don't model real human behavior are false confidence. Every test must be scenario-based: start from a real user persona and goal, simulate realistic behavior, and verify at scale. Covers all behavior angles and both short-running (burst) and long-running (sustained) scenarios.",
+        steps: [
+            {
+                step: 1,
+                name: "Define the User Persona",
+                description: "Who is the user in this scenario? First-timer, power user, distracted user, adversarial actor, mobile user on a slow network, or concurrent session? The persona determines which failure modes matter.",
+                action: "Before writing any assertion, write: 'User: <persona>. Goal: <what they want to achieve>. Prior state: <what's already in the system>.' If you can't answer these, you don't have a test scenario yet.",
+            },
+            {
+                step: 2,
+                name: "Specify Action Sequence + Timing",
+                description: "What does the user do, in what order, with what timing between actions? Rushed clicks differ from deliberate actions. Concurrent users produce race conditions that single-user tests miss entirely.",
+                action: "Document: 'Actions: [step-by-step sequence with timing]. Concurrency: [1 user / N concurrent].' For concurrent scenarios, specify what overlaps and at what rate.",
+            },
+            {
+                step: 3,
+                name: "Set the Scale Axis",
+                description: "A test that passes at 1 user is necessary but not sufficient. Define what happens at 10x and 100x. Single-user happy paths cannot catch connection pool exhaustion, rate limit throttling, lock contention, or thundering herd.",
+                action: "For every scenario, write: 'Scale: 1x (baseline) / 10x (concurrent) / 100x (sustained).' If you only have the 1x case, the test suite has a known gap — document it.",
+            },
+            {
+                step: 4,
+                name: "Set the Duration Axis",
+                description: "Short-running tests (burst/spike) and long-running tests (sustained load, state accumulation) catch different failure modes. A system that handles 10 requests in 1 second may fail after 10,000 requests over 30 minutes due to memory leaks, DB growth, stale caches, or token expiry.",
+                action: "For each scenario, specify: 'Duration: single request / session-length / multi-day accumulation.' If you have only short-running coverage, add a long-running counterpart or mark the gap explicitly.",
+            },
+            {
+                step: 5,
+                name: "Cover All Behavior Angles",
+                description: "Six required angles per feature: (1) happy path, (2) all sad paths, (3) concurrent access, (4) degraded conditions (slow network, auth expiry, partial API failures), (5) long-running accumulation (DB growth, stale cache), (6) adversarial inputs (injection, replay, unexpected payloads).",
+                action: "Use a checklist: happy ✓ / sad paths ✓ / concurrent ✓ / degraded ✓ / long-running ✓ / adversarial ✓. Missing boxes are documented gaps, not implicit coverage.",
+            },
+            {
+                step: 6,
+                name: "Verify Side Effects, Not Just Return Values",
+                description: "Assertions that only check return values miss the failure modes that matter in production: state changes, side effects, downstream consequences. Did the DB update? Did the cache invalidate? Did the audit log record? Did the notification fire?",
+                action: "After every action assertion, add assertions for: state in the DB, downstream side effects, observable UI change, and any async consequences. If you can't assert them, add instrumentation.",
+            },
+        ],
+        antiPatterns: [
+            "Simple unit tests with no scenario context",
+            "Tests that only cover the happy path",
+            "Tests that mock everything and test nothing real",
+            "Tests that pass at 1 user and are never run at 10+",
+            "Hard-coded user state assumptions that won't hold in production",
+            "'It passes in CI' declared without production-realistic data volume or concurrency",
+            "Declaring a feature 'tested' after a single integration test with synthetic clean data",
+            "Assertions only on return values — no state or side effect checks",
+        ],
+        scenarioAnatomy: {
+            template: "Scenario: <name> | User: <persona> | Goal: <user goal> | Prior state: <system state before> | Actions: <sequence with timing> | Scale: <1x/10x/100x> | Duration: <single/session/multi-day> | Expected: <state + side effects + UI> | Edge cases: <degraded/adversarial/partial>",
+            example: "Scenario: Power user submits form on slow 3G | User: power user, 50+ prior submissions, impatient | Goal: submit expense report | Prior state: DB has 10,000 existing reports, cache warm | Actions: fill form (5s), click submit, network drops to 3G mid-request, retry after 2s | Scale: 10 concurrent submitters | Duration: 30-minute session with 20 submissions | Expected: exactly-once submission, retry shows loading, no duplicate DB row, audit log entry | Edge cases: double-click, browser back mid-submit, session expiry at step 3",
+        },
+    },
     overview: {
         title: "NodeBench Development Methodology — Overview",
         description: "A dual-loop system for rigorous development. The inner loop (6-Phase Verification) ensures correctness. The outer loop (Eval-Driven Development) ensures improvement. Together they form the AI Flywheel.",
@@ -1274,6 +1439,8 @@ const METHODOLOGY_CONTENT = {
                     toon_format: "TOON Format — Token-Oriented Object Notation for ~40% token savings on LLM payloads",
                     seo_audit: "SEO Audit — technical SEO, content analysis, performance, WordPress security",
                     voice_bridge: "Voice Bridge — STT/TTS/LLM pipeline design, scaffold generation, latency benchmarking",
+                    analyst_diagnostic: "Analyst Diagnostic — root-cause diagnosis over bandaids. Mandatory for all bug work: reproduce, trace upstream, ask 'why' 5 times, fix the cause not the symptom",
+                    scenario_testing: "Scenario-Based Testing — no shallow tests, real user personas, scale axis (1x/10x/100x), duration axis (burst + sustained), all behavior angles (happy/sad/adversarial/concurrent/degraded)",
                 },
             },
             {
@@ -1554,7 +1721,7 @@ export function createMetaTools(allTools) {
         },
         {
             name: "getMethodology",
-            description: 'Get step-by-step guidance for a development methodology. Topics: verification, eval, flywheel, mandatory_flywheel, reconnaissance, quality_gates, ui_ux_qa, agentic_vision, closed_loop, learnings, project_ideation, tech_stack_2026, telemetry_setup, agents_md_maintenance, agent_bootstrap, autonomous_maintenance, parallel_agent_teams, self_reinforced_learning, academic_paper_writing, agent_evaluation, controlled_evaluation, overview. Call with topic "overview" to see all available methodologies.',
+            description: 'Get step-by-step guidance for a development methodology. Topics: verification, eval, flywheel, mandatory_flywheel, reconnaissance, quality_gates, ui_ux_qa, flywheel_ui_dogfood, agentic_vision, closed_loop, learnings, project_ideation, tech_stack_2026, telemetry_setup, agents_md_maintenance, agent_bootstrap, autonomous_maintenance, parallel_agent_teams, self_reinforced_learning, academic_paper_writing, agent_evaluation, controlled_evaluation, scenario_testing, overview. Call with topic "overview" to see all available methodologies.',
             inputSchema: {
                 type: "object",
                 properties: {
@@ -1568,6 +1735,7 @@ export function createMetaTools(allTools) {
                             "reconnaissance",
                             "quality_gates",
                             "ui_ux_qa",
+                            "flywheel_ui_dogfood",
                             "agentic_vision",
                             "closed_loop",
                             "learnings",
@@ -1582,6 +1750,7 @@ export function createMetaTools(allTools) {
                             "academic_paper_writing",
                             "agent_evaluation",
                             "controlled_evaluation",
+                            "scenario_testing",
                             "overview",
                         ],
                         description: "Which methodology to explain",