https://agent-zone.ai/tags/introspection/Recent content in Introspection on Agent ZoneHugoen-usWed, 20 May 2026 00:00:00 +0000https://agent-zone.ai/knowledge/observability/cloudflare-graphql-analytics-field-discovery/Wed, 20 May 2026 00:00:00 +0000https://agent-zone.ai/knowledge/observability/cloudflare-graphql-analytics-field-discovery/<h1 id="cloudflare-graphql-analytics-a-field-discovery-cookbook-when-introspection-is-locked">Cloudflare GraphQL Analytics: A Field-Discovery Cookbook When Introspection Is Locked<a class="anchor" href="#cloudflare-graphql-analytics-a-field-discovery-cookbook-when-introspection-is-locked">#</a></h1> <p>Cloudflare&rsquo;s GraphQL Analytics API at <code>https://api.cloudflare.com/client/v4/graphql</code> is the richest source of metrics about your CF account — Workers invocations, D1 reads/writes, KV ops, Workers AI neurons, Vectorize queries. The dashboard&rsquo;s charts are powered by it. The CLI is not: <code>wrangler</code> exposes a fraction of what GraphQL does.</p> <p>But the schema is hostile to discovery:</p> <ul> <li><code>__type(name: &quot;WorkersInvocationsAdaptive&quot;)</code> returns <code>null</code> for almost every node.</li> <li>The official schema docs at <code>developers.cloudflare.com/analytics/graphql-api</code> are partial and stale by months.</li> <li>Nodes like <code>vectorizeQueriesAdaptiveGroups</code> exist, but their <code>sum</code>/<code>dimensions</code> field names are nowhere on the public internet.</li> </ul> <p>You can still derive the schema. The trick is <strong>deliberate-error probing</strong>: send a query with a guessed field name; the error message tells you whether the parent node exists. This page is the recipe.</p>https://agent-zone.ai/knowledge/agent-tooling/self-ask-trap-llm-introspection/Wed, 20 May 2026 00:00:00 +0000https://agent-zone.ai/knowledge/agent-tooling/self-ask-trap-llm-introspection/<h1 id="the-self-ask-trap">The Self-Ask Trap<a class="anchor" href="#the-self-ask-trap">#</a></h1> <p>Practitioners ask the LLM about itself as a research shortcut: &ldquo;What are your common quirks? What temperature should I use? Do you need reasoning_content echoed in multi-turn?&rdquo; The output looks plausible, often cites specific behaviors, sometimes includes API parameter names. It is often wrong.</p> <p>The 2026-05-20 kimi-k2.6 tuning research surfaced a clean example. Self-ask said one thing. Documentation, partner adapter source, GitHub issues, and direct API probes said the opposite. The model is provably wrong about itself, and the failure mode is structural — not specific to kimi.</p>