Skip to content

revert: remove reasoning workarounds from PR #101, keep Claude reasoning enabled #248

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Draft
marius-kilocode wants to merge 1 commit into
base: dev
Choose a base branch
from
Draft

revert: remove reasoning workarounds from PR #101, keep Claude reasoning enabled #248

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
140 changes: 11 additions & 129 deletions packages/opencode/src/provider/transform.ts
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -69,14 +69,7 @@ export namespace ProviderTransform {
.filter((msg): msg is ModelMessage => msg !== undefined && msg.content !== "")
}

// kilocode_change - skip toolCallId normalization for OpenRouter/Kilo Gateway
// OpenRouter handles tool call IDs differently and modifying messages can break
// thinking/redacted_thinking blocks which must remain unchanged per Anthropic API
if (
model.api.id.includes("claude") &&
model.api.npm !== "@openrouter/ai-sdk-provider" &&
model.api.npm !== "@kilocode/kilo-gateway"
) {
if (model.api.id.includes("claude")) {
return msgs.map((msg) => {
if ((msg.role === "assistant" || msg.role === "tool") && Array.isArray(msg.content)) {
msg.content = msg.content.map((part) => {
Expand Down Expand Up @@ -257,89 +250,13 @@ export namespace ProviderTransform {
export function message(msgs: ModelMessage[], model: Provider.Model, options: Record<string, unknown>) {
msgs = unsupportedParts(msgs, model)
msgs = normalizeMessages(msgs, model, options)

// kilocode_change - skip caching for OpenRouter/Kilo Gateway to avoid modifying thinking blocks
// Anthropic API requires thinking/redacted_thinking blocks to remain exactly unchanged
const isOpenRouterOrKilo =
model.api.npm === "@openrouter/ai-sdk-provider" || model.api.npm === "@kilocode/kilo-gateway"

// kilocode_change - strip thinking/reasoning blocks for OpenRouter/Kilo Gateway
// Anthropic's API requires thinking blocks to be EXACTLY unchanged, but our storage
// reconstructs them which counts as modification. Stripping them is safe because
// the reasoning was already shown to the user and doesn't need to be sent back.
if (isOpenRouterOrKilo) {
// Helper to strip reasoning-related data from provider options/metadata
const stripReasoningData = (opts: Record<string, any> | undefined) => {
if (!opts) return undefined // Return undefined instead of empty object to clean up
const result = { ...opts }
// Strip from openrouter namespace
if (result.openrouter) {
result.openrouter = { ...result.openrouter }
delete result.openrouter.reasoning_details
delete result.openrouter.reasoning
delete result.openrouter.thinking
}
// Strip from kilo namespace
if (result.kilo) {
result.kilo = { ...result.kilo }
delete result.kilo.reasoning_details
delete result.kilo.reasoning
delete result.kilo.thinking
}
// Strip from anthropic namespace
if (result.anthropic) {
result.anthropic = { ...result.anthropic }
delete result.anthropic.thinking
delete result.anthropic.reasoning
}
return result
}

msgs = msgs.flatMap((msg): ModelMessage[] => {
// Handle string content (just strip metadata)
if (!Array.isArray(msg.content)) {
const result = { ...msg, providerOptions: stripReasoningData(msg.providerOptions) }
if ("experimental_providerMetadata" in msg) {
;(result as any).experimental_providerMetadata = stripReasoningData(
(msg as any).experimental_providerMetadata,
)
}
return [result]
}
// Filter out reasoning parts from content and strip metadata from remaining parts
const filtered = msg.content
.filter(
(part: any) => part.type !== "thinking" && part.type !== "redacted_thinking" && part.type !== "reasoning",
)
.map((part: any) => ({
...part,
providerOptions: stripReasoningData(part.providerOptions),
providerMetadata: stripReasoningData(part.providerMetadata),
experimental_providerMetadata: stripReasoningData(part.experimental_providerMetadata),
}))

// Providers may reject empty array content; drop empty messages.
if (filtered.length === 0) return []

// Also strip from message-level options/metadata
const result = { ...msg, content: filtered, providerOptions: stripReasoningData(msg.providerOptions) }
if ("experimental_providerMetadata" in msg) {
;(result as any).experimental_providerMetadata = stripReasoningData(
(msg as any).experimental_providerMetadata,
)
}
return [result as ModelMessage]
})
}

if (
!isOpenRouterOrKilo &&
(model.providerID === "anthropic" ||
model.api.id.includes("anthropic") ||
model.api.id.includes("claude") ||
model.id.includes("anthropic") ||
model.id.includes("claude") ||
model.api.npm === "@ai-sdk/anthropic")
model.providerID === "anthropic" ||
model.api.id.includes("anthropic") ||
model.api.id.includes("claude") ||
model.id.includes("anthropic") ||
model.id.includes("claude") ||
model.api.npm === "@ai-sdk/anthropic"
) {
msgs = applyCaching(msgs, model.providerID)
}
Expand Down Expand Up @@ -447,51 +364,17 @@ export namespace ProviderTransform {

// kilocode_change start
case "@kilocode/kilo-gateway":
// kilocode_change - adaptive thinking with effort levels
// TODO: Enable when @ai-sdk/anthropic supports thinking.type: "adaptive"
const ADAPTIVE_THINKING_ENABLED = false
if (ADAPTIVE_THINKING_ENABLED && id.includes("claude-opus-4-6")) {
return {
low: {
thinking: { type: "adaptive" },
output_config: { effort: "low" },
},
medium: {
thinking: { type: "adaptive" },
output_config: { effort: "medium" },
},
high: {
thinking: { type: "adaptive" },
output_config: { effort: "high" },
},
max: {
thinking: { type: "adaptive" },
output_config: { effort: "max" },
},
}
}
// Claude/Anthropic models support reasoning via effort levels through OpenRouter API
// OpenRouter uses OpenAI-style effort names: xhigh=95%, high=80%, medium=50%, low=20%, minimal=10%
// kilocode_change - expose "max" (Anthropic naming) to users, mapped to "xhigh" (OpenRouter naming) on the wire
// Claude/Anthropic models use OpenRouter reasoning format
if (
model.id.includes("claude") ||
model.id.includes("anthropic") ||
model.api.id.includes("claude") ||
model.api.id.includes("anthropic")
) {
const ANTHROPIC_EFFORTS = ["none", "minimal", ...WIDELY_SUPPORTED_EFFORTS, "max"]
return Object.fromEntries(
ANTHROPIC_EFFORTS.map((effort) => [effort, { reasoning: { effort: effort === "max" ? "xhigh" : effort } }]),
)
return Object.fromEntries(OPENAI_EFFORTS.map((effort) => [effort, { reasoning: { effort } }]))
}
// GPT models via Kilo need encrypted reasoning content to avoid org_id mismatch
if (!model.id.includes("gpt") && !model.id.includes("gemini-3")) return {}
// kilocode_change - Codex models use object-based reasoning format for OpenRouter
// OpenRouter expects { reasoning: { effort: "high" } } format
// See: https://openrouter.ai/docs/api/api-reference/chat/send-chat-completion-request#request.body.reasoning
if (model.id.includes("codex")) {
return Object.fromEntries(OPENAI_EFFORTS.map((effort) => [effort, { reasoning: { effort } }]))
}
return Object.fromEntries(
OPENAI_EFFORTS.map((effort) => [
effort,
Expand Down Expand Up @@ -875,11 +758,11 @@ export namespace ProviderTransform {
return { thinkingConfig: { thinkingBudget: 0 } }
}
if (model.providerID === "openrouter" || model.api.npm === "@kilocode/kilo-gateway") {
// kilocode_change - add Kilo Gateway support with model-specific handling
// kilocode_change
if (model.api.id.includes("google")) {
return { reasoning: { enabled: false } }
}
// Claude models need reasoning.effort format (OpenRouter API)
// kilocode_change - Claude models use reasoning.effort format via OpenRouter API
if (
model.id.includes("claude") ||
model.id.includes("anthropic") ||
Expand All @@ -888,7 +771,6 @@ export namespace ProviderTransform {
) {
return { reasoning: { effort: "minimal" } }
}
// Other models use reasoningEffort (AI SDK format)
return { reasoningEffort: "minimal" }
}
return {}
Expand Down
Loading
Loading