fix: Use --build-local in smoke-otel-tracing for latest api-proxy code (#3483)

* fix: use --build-local in smoke-otel-tracing to test latest api-proxy code

The workflow was using --skip-pull with pre-built GHCR images (v0.25.29)
which don't include the recently-merged otel.js module. Switch to
--build-local so the api-proxy container is built from source and
includes the OTEL tracing implementation.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* feat: inject stream_options for token tracking in spans

Streaming OpenAI/Copilot API responses only include usage data in the
final SSE chunk when the client sends stream_options: {include_usage: true}.
Without this, the token tracker never extracts usage and OTEL spans have
no gen_ai.usage.* attributes.

The api-proxy now injects this option automatically for OpenAI-compatible
providers (copilot, openai, opencode) when stream: true is set and
stream_options is not already present.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

---------

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Author: lpcox

.github/workflows/smoke-otel-tracing.lock.yml

@@ -22,7 +22,8 @@ COPY server.js logging.js metrics.js rate-limiter.js \
      proxy-request.js model-discovery.js management.js oidc-token-provider.js \
      oidc-token-provider-base.js \
      github-oidc.js aws-oidc-token-provider.js gcp-oidc-token-provider.js \
-     oidc-refresh-utils.js body-transform.js rate-limit.js websocket-proxy.js ./
+     oidc-refresh-utils.js body-transform.js rate-limit.js websocket-proxy.js \
+     otel.js ./
 COPY guards/ ./guards/
 COPY providers/ ./providers/
 

containers/api-proxy/Dockerfile

@@ -124,7 +124,40 @@ function injectSteeringMessage(body, provider, message) {
   return Buffer.from(JSON.stringify(parsed));
 }
 
+/**
+ * Inject `stream_options: { include_usage: true }` into OpenAI-compatible
+ * streaming requests so that the final SSE chunk includes token usage data.
+ * This is required for the token tracker (and OTEL spans) to capture usage.
+ *
+ * Only modifies the body when `stream: true` is present and `stream_options`
+ * is not already set.  Anthropic and Gemini use different mechanisms and are
+ * skipped.
+ *
+ * @param {Buffer} body
+ * @param {string} provider - 'openai' | 'copilot' | 'opencode' | 'anthropic' | 'gemini'
+ * @returns {{ body: Buffer, injected: boolean }|null}
+ */
+function injectStreamOptions(body, provider) {
+  // Only applies to OpenAI-compatible providers
+  if (provider === 'anthropic' || provider === 'gemini') return null;
+
+  let parsed;
+  try {
+    parsed = JSON.parse(body.toString('utf8'));
+  } catch {
+    return null;
+  }
+
+  if (!parsed || typeof parsed !== 'object' || Array.isArray(parsed)) return null;
+  if (!parsed.stream) return null;
+  if (parsed.stream_options) return null;
+
+  parsed.stream_options = { include_usage: true };
+  return { body: Buffer.from(JSON.stringify(parsed)), injected: true };
+}
+
 module.exports = {
   sanitizeNullToolCallTypes,
   injectSteeringMessage,
+  injectStreamOptions,
 };

containers/api-proxy/body-transform.js

@@ -13,7 +13,7 @@ const { generateRequestId, sanitizeForLog, logRequest } = require('./logging');
 const metrics = require('./metrics');
 const rateLimiter = require('./rate-limiter');
 const { buildUpstreamPath, shouldStripHeader } = require('./proxy-utils');
-const { sanitizeNullToolCallTypes, injectSteeringMessage } = require('./body-transform');
+const { sanitizeNullToolCallTypes, injectSteeringMessage, injectStreamOptions } = require('./body-transform');
 const { createRateLimitChecker } = require('./rate-limit');
 const { createProxyWebSocket } = require('./websocket-proxy');
 const {
@@ -346,6 +346,14 @@ function proxyRequest(req, res, targetHost, injectHeaders, provider, basePath =
       }
     }
 
+    // Inject stream_options.include_usage so streaming responses include token data
+    if (req.method === 'POST') {
+      const streamOpts = injectStreamOptions(body, provider);
+      if (streamOpts) {
+        body = streamOpts.body;
+      }
+    }
+
     const requestBytes = body.length;
     metrics.increment('request_bytes_total', { provider }, requestBytes);