feat: add prompt_cache_key

[CellenLee]

Related GitHub Issue

None

Roo Code Task Context (Optional)

Description

Regardless of whether the service provider provides an explicit prompt cache option, adding prompt_cache_key helps the LLM provider dispatch to a more cache-friendly server, then the prefix caching provided by vllm and sglang will work.

So I created this PR to add prompt_cache_key to every LLM provider. In the long run, this will help LLM service providers optimize the scheduling of RooCode requests, thereby reducing TTFT and financial consumption.

Test Procedure

Pre-Submission Checklist

• Issue Linked: This PR is linked to an approved GitHub Issue (see "Related GitHub Issue" above).
• Scope: My changes are focused on the linked issue (one major feature/fix per PR).
• Self-Review: I have performed a thorough self-review of my code.
• Testing: New and/or updated tests have been added to cover my changes (if applicable).
• Documentation Impact: I have considered if my changes require documentation updates (see "Documentation Updates" section below).
• Contribution Guidelines: I have read and agree to the Contributor Guidelines.

Screenshots / Videos

POST /v1/chat/completions

{
	"model": "gpt-5",
	"temperature": 0,
	"messages": [{
		"role": "system",
	}, {
		"role": "user",
	}, {
		"role": "assistant",
	}, {
		"role": "user",
	}, {
		"role": "assistant",
	}, {
		"role": "user",
	}],
	"stream": true,
	"stream_options": {
		"include_usage": true
	},
	"prompt_cache_key": "6ab1234d-c123-1234-12ab-aaaaaaaaaaaa"
}

Documentation Updates

Does this PR necessitate updates to user-facing documentation?

• No documentation updates are required.
• Yes, documentation updates are required. (Please describe what needs to be updated or link to a PR in the docs repository).

Additional Notes

Get in Touch

---

Important

Adds prompt_cache_key to enhance caching across multiple LLM providers, optimizing request scheduling and reducing costs.

• Behavior:
  • Adds prompt_cache_key to completePrompt and createMessage functions across multiple LLM providers to enhance caching.
  • Affects providers like OpenAiHandler, OpenAiNativeHandler, OllamaHandler, QwenCodeHandler, and others.
  • Aims to optimize request scheduling and reduce costs by improving cache utilization.
• Functions:
  • Updates completePrompt in mistral.ts, native-ollama.ts, ollama.ts, openai-native.ts, openai.ts, openrouter.ts, qwen-code.ts, requesty.ts, unbound.ts, vercel-ai-gateway.ts, vscode-lm.ts, and xai.ts to include prompt_cache_key.
  • Updates createMessage in the same files to include prompt_cache_key.
• Misc:
  • Minor code style adjustments in codebaseSearchTool.ts and c-sharp.ts.

^{This description was created by for 54820c9. You can customize this summary. It will automatically update as commits are pushed.}

[roomote]

I found some issues that need attention. See inline comments for details.

[roomote]

P0: Official OpenAI endpoints may reject unknown request args (e.g., 'Unrecognized request argument: prompt_cache_key'). Please gate 'prompt_cache_key' and 'safety_identifier' so they’re only sent to endpoints that accept them (OpenRouter, vLLM/sglang gateways, etc.). Otherwise this can cause 400s for users on api.openai.com.

[roomote]

P1: The OpenAI SDK param types may not permit extra fields. If you keep these fields, ensure types allow them or route via a supported 'extra body' mechanism. Otherwise TS type-check or runtime validation could fail depending on the SDK/version.

[roomote]

P0: Same gating concern as above — adding 'prompt_cache_key' and 'safety_identifier' to official OpenAI Chat Completions can trigger 'unrecognized argument' errors. Please conditionally include based on baseURL/provider capability (or behind a feature flag).

[roomote]

P0/P1: Non-streaming request includes 'prompt_cache_key' and 'safety_identifier'. Official OpenAI endpoints may reject unknown fields and the SDK types may not allow extra props. Suggest gating by endpoint or providing a type-safe extension path.

[hannesrudolph]

I am very sorry for the late response to this! This PR does not meet contribution standards.

Specific issues:

• No linked issue or defined problem statement.
• Scope creep: includes unrelated formatting changes.
• Implementation causes instability and breaks existing functionality.

Please open a tracked issue that clearly defines the problem before resubmitting. Each PR must focus on a single, well-scoped change with tests and documentation updates where applicable.