abhigyanpatwari/GitNexus · error · Error

content filter triggered mid-stream. The generated content w

Error message

content filter triggered mid-stream. The generated content was blocked by content policy. Adjust your prompt and retry.

What it means

During SSE streaming, the client saw a content_filter flag in the stream before/at completion and discards the partial content, telling you the generated text was blocked by content policy mid-generation. This differs from the non-streaming Azure 400 case: the request was accepted and streaming started, then the provider's safety system cut it off.

Source

Thrown at gitnexus/src/core/wiki/llm-client.ts:515

        // Detect content filter finish reason — skip delta from this chunk
        if (choice?.finish_reason === 'content_filter') {
          contentFilterTriggered = true;
          continue;
        }

        const delta = choice?.delta?.content;
        if (delta) {
          content += delta;
          onChunk(content.length);
        }
      } catch {
        // Skip malformed SSE chunks
      }
    }
  }

  if (contentFilterTriggered) {
    throw new Error(
      'content filter triggered mid-stream. The generated content was blocked by content policy. Adjust your prompt and retry.',
    );
  }

  if (!content) {
    throw new Error('LLM returned empty streaming response');
  }

  return { content };
}

View on GitHub (pinned to aac7515d2a)

Solutions

1. Adjust the prompt/source content: rename or exclude the module whose page keeps triggering the filter
2. Loosen/adjust the provider's content-filter configuration (Azure: filter settings per deployment)
3. Switch to a different provider or model with different filtering
4. Retry — mid-stream filters are often borderline-score dependent

try {
  await callLLM(prompt, config, undefined, { onChunk });
} catch (err) {
  if (err instanceof Error && err.message.includes('content filter triggered mid-stream')) {
    // partial content already streamed via onChunk — persist it as a stub and continue
    return { content: partialContent + '\n\n[Generation blocked by content filter]' };
  }
  throw err;
}

When it happens

Trigger: Streaming call (useStream) where the provider emits a content-filter finish reason or error frame partway through; model begins a doc page whose generated content (often security/exploit-adjacent code text) trips filters after several tokens.

Common situations: Wiki generation for repos with security-tooling, weapon, drug, or graphic naming; providers with strict streaming filters (Azure OpenAI, some gateways); intermittent — same repo can pass or fail depending on sampled content.

Related errors

• Azure content filter blocked this request. The prompt trigge
• LLM returned empty streaming response
• Streaming PDG manifest collides with a structural node CSV f
• Streaming PDG manifest collides with a structural relationsh
• PdgEmitSink: ${errors.length} streamed CSV writer(s) hit an

AI-assisted analysis of abhigyanpatwari/GitNexus@aac7515d2a (2026-08-20). Data as JSON: /api/errors/a7a1a5e9c59b00c2. Report an issue: GitHub.