Prevent assistant image hash placeholders from leaking into follow-up LLM prompts

[Copilot]

Generated images were being persisted as internal markdown placeholders containing file-hash:..., and those placeholders could be replayed back into later assistant context. After a few image generations, the model could start echoing hash-like strings instead of treating prior outputs as images.

• Prompt serialization

  • Strip internal assistant image placeholders from outbound OpenRouter text before sending history back to the model.
  • Keep image inputs intact by continuing to hydrate the real attachments from file_hashes.

• Supported placeholder formats

  • Handles the current persisted form:
    • ![file-hash:HASH](...)
  • Handles legacy forms:
    • ![alt](file-hash:HASH)
    • ![alt](blob:file-hash:HASH)

• Regression coverage

  • Added focused tests around buildOpenRouterMessages() to verify:
    • placeholder text is removed from assistant messages
    • whitespace is normalized after removal
    • images are still attached from file_hashes

const orMessages = await buildOpenRouterMessages([
  {
    role: 'assistant',
    content: 'Here is the image.\n\n![file-hash:abc123](...)',
    file_hashes: JSON.stringify(['abc123']),
  },
]);

// outbound assistant text no longer contains the internal placeholder
// but the hydrated image is still present as an image_url part

• Files

  • app/core/auth/openrouter-build.ts
  • app/core/auth/__tests__/openrouter-build.test.ts

• • Manual verification screenshot of the chat test harness:
    • https://github.com/user-attachments/assets/2e5a9fba-4f6f-47fc-abce-1800a982cfd9

Warning

Firewall rules blocked me from connecting to one or more addresses (expand for details)

I tried to connect to the following addresses, but was blocked by firewall rules:

• 0.0.0.0.in-addr.arpa
  • Triggering command: /home/REDACTED/work/_temp/ghcca-node/node/bin/node node /home/REDACTED/work/or3-chat/or3-chat/node_modules/.bin/nuxt dev --host 0.0.0.0 --port 3000 ash it} --global /usr/local/bin/git cat /pro�� (dns block)
• api.fontshare.com
  • Triggering command: /home/REDACTED/work/_temp/ghcca-node/node/bin/node node /home/REDACTED/work/or3-chat/or3-chat/node_modules/.bin/nuxt build (dns block)
  • Triggering command: /home/REDACTED/work/_temp/ghcca-node/node/bin/node node /home/REDACTED/work/or3-chat/or3-chat/node_modules/.bin/nuxt build hat/node_modules/vite/node_modules/@esbuild/linux-x64/bin/esbuild -c o | grep "physical id" | sort |uniq | wc -l (dns block)
  • Triggering command: /home/REDACTED/work/_temp/ghcca-node/node/bin/node node /home/REDACTED/work/or3-chat/or3-chat/node_modules/.bin/nuxt dev --host 0.0.0.0 --port 3000 ash it} --global /usr/local/bin/git cat /pro�� (dns block)
• api.fontsource.org
  • Triggering command: /home/REDACTED/work/_temp/ghcca-node/node/bin/node node /home/REDACTED/work/or3-chat/or3-chat/node_modules/.bin/nuxt build (dns block)
  • Triggering command: /home/REDACTED/work/_temp/ghcca-node/node/bin/node node /home/REDACTED/work/or3-chat/or3-chat/node_modules/.bin/nuxt build hat/node_modules/vite/node_modules/@esbuild/linux-x64/bin/esbuild -c o | grep "physical id" | sort |uniq | wc -l (dns block)
  • Triggering command: /home/REDACTED/work/_temp/ghcca-node/node/bin/node node /home/REDACTED/work/or3-chat/or3-chat/node_modules/.bin/nuxt dev --host 0.0.0.0 --port 3000 ash it} --global /usr/local/bin/git cat /pro�� (dns block)
• fonts.bunny.net
  • Triggering command: /home/REDACTED/work/_temp/ghcca-node/node/bin/node node /home/REDACTED/work/or3-chat/or3-chat/node_modules/.bin/nuxt build (dns block)
  • Triggering command: /home/REDACTED/work/_temp/ghcca-node/node/bin/node node /home/REDACTED/work/or3-chat/or3-chat/node_modules/.bin/nuxt build hat/node_modules/vite/node_modules/@esbuild/linux-x64/bin/esbuild -c o | grep "physical id" | sort |uniq | wc -l (dns block)
  • Triggering command: /home/REDACTED/work/_temp/ghcca-node/node/bin/node node /home/REDACTED/work/or3-chat/or3-chat/node_modules/.bin/nuxt dev --host 0.0.0.0 --port 3000 ash it} --global /usr/local/bin/git cat /pro�� (dns block)
• fonts.google.com
  • Triggering command: /home/REDACTED/work/_temp/ghcca-node/node/bin/node node /home/REDACTED/work/or3-chat/or3-chat/node_modules/.bin/nuxt build (dns block)
  • Triggering command: /home/REDACTED/work/_temp/ghcca-node/node/bin/node node /home/REDACTED/work/or3-chat/or3-chat/node_modules/.bin/nuxt build hat/node_modules/vite/node_modules/@esbuild/linux-x64/bin/esbuild -c o | grep "physical id" | sort |uniq | wc -l (dns block)
  • Triggering command: /home/REDACTED/work/_temp/ghcca-node/node/bin/node node /home/REDACTED/work/or3-chat/or3-chat/node_modules/.bin/nuxt dev --host 0.0.0.0 --port 3000 ash it} --global /usr/local/bin/git cat /pro�� (dns block)

If you need me to access, download, or install something from one of these locations, you can either:

• Configure Actions setup steps to set up my environment, which run before the firewall is enabled
• Add the appropriate URLs or hosts to the custom allowlist in this repository's Copilot coding agent settings (admins only)

Original prompt

When im talking in a chat and i have an image generation llm model it works ok, but after i one or two generations the llm starts sending back the image hash.

What i think is happening is that the image hash is being sent to the llm instead of the image, and it's confused/recognizing the hash pattern, so instead of generating another image it generates a hash.

Look into it and see if you can solve this issue

---

📍 Connect Copilot coding agent with Jira, Azure Boards or Linear to delegate work to Copilot in one click without leaving your project management tool.

[cloudflare-workers-and-pages]

Deploying or3-chat with    Cloudflare Pages

Latest commit:	166a6ed
Status:	🚫  Build failed.

View logs