Imported from jeremylongshore/tons-of-skills-marketplace (skills/.curated/exa-advanced-troubleshooting/SKILL.md). Install upstream with npx skills add jeremylongshore/tons-of-skills-marketplace --skill exa-advanced-troubleshooting. Copyright stays with the author (MIT).

Exa Advanced Troubleshooting

Prerequisites

• A bounded non-sensitive reproduction, affected environment/version, expected behavior, and redaction policy.
• Authority to inspect aggregate diagnostics and a named owner for escalation.

Output

• A minimal redacted root-cause record with evidence, remediation, owner, and verification result.
• A safe vendor/internal escalation packet without raw sensitive queries, result content, or API keys.

Examples

Reproduce a retrieval-quality issue with a sanitized query and approved domain filter, compare response metadata and aggregate timing to the expected contract, then test one scoped configuration fix. If unresolved, submit only the redacted request category/options/correlation evidence through the approved support channel.

Overview

Deep debugging for complex Exa issues: latency spikes, intermittent failures, result quality degradation, and content retrieval failures. All Exa error responses include a requestId — always capture it.

Instructions

Step 1: Layer-by-Layer Diagnostics

import Exa from "exa-js";

interface DiagnosticResult {
  layer: string;
  success: boolean;
  latencyMs: number;
  details: string;
}

async function diagnoseExa(): Promise<DiagnosticResult[]> {
  const results: DiagnosticResult[] = [];
  const exa = new Exa(process.env.EXA_API_KEY);

  // Layer 1: DNS + Network
  let start = performance.now();
  try {
    const resp = await fetch("https://api.exa.ai", { method: "HEAD" });
    results.push({
      layer: "network",
      success: true,
      latencyMs: performance.now() - start,
      details: `HTTP ${resp.status}`,
    });
  } catch (err: any) {
    results.push({
      layer: "network",
      success: false,
      latencyMs: performance.now() - start,
      details: err.message,
    });
    return results; // No point continuing if network fails
  }

  // Layer 2: Authentication
  start = performance.now();
  try {
    await exa.search("auth test", { numResults: 1 });
    results.push({
      layer: "auth",
      success: true,
      latencyMs: performance.now() - start,
      details: "API key valid",
    });
  } catch (err: any) {
    results.push({
      layer: "auth",
      success: false,
      latencyMs: performance.now() - start,
      details: `${err.status}: ${err.message}`,
    });
    if (err.status === 401 || err.status === 402) return results;
  }

  // Layer 3: Neural search
  start = performance.now();
  try {
    const r = await exa.search("test neural search quality", {
      type: "neural",
      numResults: 3,
    });
    results.push({
      layer: "neural-search",
      success: true,
      latencyMs: performance.now() - start,
      details: `${r.results.length} results, top score: ${r.results[0]?.score.toFixed(3)}`,
    });
  } catch (err: any) {
    results.push({
      layer: "neural-search",
      success: false,
      latencyMs: performance.now() - start,
      details: `${err.status}: ${err.message}`,
    });
  }

  // Layer 4: Content retrieval
  start = performance.now();
  try {
    const r = await exa.searchAndContents("content retrieval test", {
      numResults: 1,
      text: { maxCharacters: 500 },
      highlights: { maxCharacters: 200 },
    });
    const hasText = !!r.results[0]?.text;
    const hasHighlights = !!r.results[0]?.highlights?.length;
    results.push({
      layer: "content-retrieval",
      success: hasText,
      latencyMs: performance.now() - start,
      details: `text: ${hasText}, highlights: ${hasHighlights}`,
    });
  } catch (err: any) {
    results.push({
      layer: "content-retrieval",
      success: false,
      latencyMs: performance.now() - start,
      details: `${err.status}: ${err.message}`,
    });
  }

  // Layer 5: findSimilar
  start = performance.now();
  try {
    const r = await exa.findSimilar("https://nodejs.org", { numResults: 2 });
    results.push({
      layer: "find-similar",
      success: r.results.length > 0,
      latencyMs: performance.now() - start,
      details: `${r.results.length} similar pages found`,
    });
  } catch (err: any) {
    results.push({
      layer: "find-similar",
      success: false,
      latencyMs: performance.now() - start,
      details: `${err.status}: ${err.message}`,
    });
  }

  return results;
}

// Print diagnostic report
const results = await diagnoseExa();
console.log("=== Exa Diagnostic Report ===");
for (const r of results) {
  const icon = r.success ? "PASS" : "FAIL";
  console.log(`[${icon}] ${r.layer}: ${r.latencyMs.toFixed(0)}ms — ${r.details}`);
}

Step 2: Latency Profiling

async function profileLatency(query: string, iterations = 5) {
  const exa = new Exa(process.env.EXA_API_KEY);
  const timings: { type: string; ms: number }[] = [];

  for (const type of ["instant", "fast", "auto", "neural"] as const) {
    for (let i = 0; i < iterations; i++) {
      const start = performance.now();
      try {
        await exa.search(query, { type, numResults: 3 });
        timings.push({ type, ms: performance.now() - start });
      } catch {
        timings.push({ type, ms: -1 }); // -1 indicates failure
      }
    }
  }

  // Summarize
  const grouped = new Map<string, number[]>();
  for (const t of timings) {
    if (!grouped.has(t.type)) grouped.set(t.type, []);
    if (t.ms > 0) grouped.get(t.type)!.push(t.ms);
  }

  console.log(`\nLatency profile for: "${query}"`);
  for (const [type, times] of grouped) {
    const sorted = times.sort((a, b) => a - b);
    const p50 = sorted[Math.floor(sorted.length * 0.5)];
    const p95 = sorted[Math.floor(sorted.length * 0.95)];
    console.log(`  ${type}: p50=${p50?.toFixed(0)}ms, p95=${p95?.toFixed(0)}ms`);
  }
}

Step 3: Content Retrieval Debugging

// When getContents or searchAndContents returns empty text
async function debugContentRetrieval(url: string) {
  const exa = new Exa(process.env.EXA_API_KEY);
  const configs = [
    { name: "default", opts: { text: true } },
    { name: "livecrawl-preferred", opts: { text: true, livecrawl: "preferred" as const, livecrawlTimeout: 15000 } },
    { name: "livecrawl-always", opts: { text: true, livecrawl: "always" as const, livecrawlTimeout: 15000 } },
    { name: "highlights-only", opts: { highlights: { maxCharacters: 500 } } },
    { name: "summary-only", opts: { summary: true } },
  ];

  console.log(`\nContent retrieval debug for: ${url}`);
  for (const { name, opts } of configs) {
    try {
      const result = await exa.getContents([url], opts as any);
      const r = result.results[0];
      console.log(`  ${name}: text=${r?.text?.length || 0} chars, highlights=${r?.highlights?.length || 0}`);
    } catch (err: any) {
      console.log(`  ${name}: ERROR ${err.status} — ${err.message}`);
    }
  }
}

Step 4: Support Escalation Template

## Exa Support Escalation

**Severity:** P[1-4]
**RequestId:** [from error response]
**Timestamp:** [ISO 8601 from error]
**SDK:** exa-js [version from npm list]

### Issue Summary
[One paragraph description]

### Steps to Reproduce
1. Initialize Exa client
2. Call [method] with [parameters]
3. Observe [error/unexpected behavior]

### Expected vs Actual
- Expected: [behavior]
- Actual: [behavior]

### Diagnostic Results
[Output from diagnoseExa() function]

### Evidence
- Latency profile attached
- Content retrieval debug output
- Error response with requestId

Error Handling

Resources

• Exa Error Codes
• Exa Support
• Exa Rate Limits

Next Steps

For load testing, see exa-load-scale.