@maintainabilityai/research-runner 0.1.1 → 0.1.3

package/dist/llm/anthropic-client.js

@@ -23,8 +23,10 @@ async function callAnthropic(opts) {
         throw new Error('ANTHROPIC_API_KEY missing — set the env var or pass apiKey directly');
     }
     const fetchImpl = opts.fetchImpl ?? globalThis.fetch;
+    // Match github-models default — 8K-token synth responses can run 60–90s.
+    const timeoutMs = opts.timeoutMs ?? 120_000;
     const controller = new AbortController();
-    const timer = setTimeout(() => controller.abort(), opts.timeoutMs ?? 60_000);
+    const timer = setTimeout(() => controller.abort(), timeoutMs);
     let response;
     try {
         response = await fetchImpl('https://api.anthropic.com/v1/messages', {
@@ -44,6 +46,12 @@ async function callAnthropic(opts) {
             signal: controller.signal,
         });
     }
+    catch (err) {
+        if (err instanceof Error && err.name === 'AbortError') {
+            throw new Error(`Anthropic request timed out after ${timeoutMs}ms (model=${opts.model}, max_tokens=${opts.maxTokens})`);
+        }
+        throw new Error(`Anthropic fetch failed (model=${opts.model}): ${err instanceof Error ? err.message : String(err)}`);
+    }
     finally {
         clearTimeout(timer);
     }

package/dist/llm/github-models-client.d.ts

@@ -11,12 +11,16 @@
  * their result types.
  *
  * Model names use GitHub Models namespacing — e.g. `openai/gpt-4o`,
- * `openai/gpt-4o-mini`, `anthropic/claude-3-5-sonnet`. The router (in
+ * `openai/gpt-4o-mini`, `openai/gpt-4.1`. The router (in
  * llm-router.ts) maps internal logical model tiers (`plan` / `synth`) to
  * the concrete provider-specific id.
  */
-/** Subset of GitHub Models model ids we use. Extend as new tiers land. */
-export type GitHubModelsModel = 'openai/gpt-4o' | 'openai/gpt-4o-mini' | 'anthropic/claude-3-5-sonnet' | 'anthropic/claude-3-5-haiku';
+/**
+ * Subset of GitHub Models model ids we use. Extend as new tiers land.
+ * GitHub Models does not currently host Anthropic Claude — synth tier
+ * uses `openai/gpt-4.1` (the "outperforms gpt-4o across the board" tier).
+ */
+export type GitHubModelsModel = 'openai/gpt-4o' | 'openai/gpt-4o-mini' | 'openai/gpt-4.1' | 'openai/gpt-4.1-mini';
 export interface CallGitHubModelsOpts {
     /** Workflow GITHUB_TOKEN. The model server checks the `models:read` permission scope. */
     token: string;

package/dist/llm/github-models-client.js

@@ -12,7 +12,7 @@
  * their result types.
  *
  * Model names use GitHub Models namespacing — e.g. `openai/gpt-4o`,
- * `openai/gpt-4o-mini`, `anthropic/claude-3-5-sonnet`. The router (in
+ * `openai/gpt-4o-mini`, `openai/gpt-4.1`. The router (in
  * llm-router.ts) maps internal logical model tiers (`plan` / `synth`) to
  * the concrete provider-specific id.
  */
@@ -25,8 +25,11 @@ async function callGitHubModels(opts) {
     }
     const fetchImpl = opts.fetchImpl ?? globalThis.fetch;
     const endpoint = opts.endpoint ?? DEFAULT_ENDPOINT;
+    // Synthesis prompts can produce 8K-token responses on gpt-4.1, which
+    // routinely take 60–90s. Default to 120s so we don't abort mid-stream.
+    const timeoutMs = opts.timeoutMs ?? 120_000;
     const controller = new AbortController();
-    const timer = setTimeout(() => controller.abort(), opts.timeoutMs ?? 60_000);
+    const timer = setTimeout(() => controller.abort(), timeoutMs);
     const messages = [];
     if (opts.system) {
         messages.push({ role: 'system', content: opts.system });
@@ -50,6 +53,12 @@ async function callGitHubModels(opts) {
             signal: controller.signal,
         });
     }
+    catch (err) {
+        if (err instanceof Error && err.name === 'AbortError') {
+            throw new Error(`GitHub Models request timed out after ${timeoutMs}ms (model=${opts.model}, max_tokens=${opts.maxTokens})`);
+        }
+        throw new Error(`GitHub Models fetch failed (model=${opts.model}): ${err instanceof Error ? err.message : String(err)}`);
+    }
     finally {
         clearTimeout(timer);
     }

package/dist/llm/llm-router.js

@@ -6,7 +6,7 @@ const github_models_client_1 = require("./github-models-client");
 /** Per-tier per-provider model id lookup. */
 const MODEL_BY_TIER = {
     plan: { anthropic: 'claude-haiku-4-5', githubModels: 'openai/gpt-4o-mini' },
-    synth: { anthropic: 'claude-sonnet-4-6', githubModels: 'anthropic/claude-3-5-sonnet' },
+    synth: { anthropic: 'claude-sonnet-4-6', githubModels: 'openai/gpt-4.1' },
 };
 async function callLlm(opts) {
     const tierModels = MODEL_BY_TIER[opts.tier];

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@maintainabilityai/research-runner",
-  "version": "0.1.1",
+  "version": "0.1.3",
   "description": "Research + PRD agent runner — orchestrates the Archeologist and PRD pipelines for the MaintainabilityAI governance mesh",
   "license": "MIT",
   "author": "MaintainabilityAI",