Skip to content

fix: avoid Anthropic tool-history requests without tools #8248

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

Open
seilk wants to merge 1 commit into
base: dev
Choose a base branch
from
+40 −2
Open

fix: avoid Anthropic tool-history requests without tools #8248

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
42 changes: 40 additions & 2 deletions packages/opencode/src/session/llm.ts
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -5,6 +5,8 @@ import { Log } from "@/util/log"
import {
streamText,
wrapLanguageModel,
tool,
jsonSchema,
type ModelMessage,
type StreamTextResult,
type Tool,
Expand All @@ -26,6 +28,20 @@ import { Auth } from "@/auth"
export namespace LLM {
const log = Log.create({ service: "llm" })

/**
* Check if message history contains tool calls.
* Some providers (e.g., Anthropic via LiteLLM) reject requests that have
* tool call history in messages but no tools parameter specified.
* @see https://github.com/sst/opencode/issues/2915
*/
function hasToolCallsInHistory(messages: ModelMessage[]): boolean {
return messages.some(
(msg) =>
Array.isArray(msg.content) &&
msg.content.some((part: any) => part.type === "tool-call" || part.type === "tool-result"),
)
}

export const OUTPUT_TOKEN_MAX = Flag.OPENCODE_EXPERIMENTAL_OUTPUT_TOKEN_MAX || 32_000

export type StreamInput = {
Expand Down Expand Up @@ -134,6 +150,28 @@ export namespace LLM {

const tools = await resolveTools(input)

// When message history contains tool calls but no tools are provided,
// some providers (Anthropic via LiteLLM/proxies) reject the request.
// Add a dummy tool to satisfy the API requirement.
// @see https://github.com/sst/opencode/issues/2915
const hasToolHistory = hasToolCallsInHistory(input.messages)
const noToolsProvided = Object.keys(tools).length === 0

// Restrict workaround to LiteLLM proxies only.
const isLiteLLM =
input.model.providerID.toLowerCase().includes("litellm") ||
input.model.api.id.toLowerCase().includes("litellm")

const effectiveTools: Record<string, Tool> =
isLiteLLM && noToolsProvided && hasToolHistory
? {
_dummy: tool({
description: "Placeholder tool for API compatibility",
inputSchema: jsonSchema({ type: "object", properties: {} }),
}),
}
: tools

return streamText({
onError(error) {
l.error("stream error", {
Expand Down Expand Up @@ -165,8 +203,8 @@ export namespace LLM {
topP: params.topP,
topK: params.topK,
providerOptions: ProviderTransform.providerOptions(input.model, params.options),
activeTools: Object.keys(tools).filter((x) => x !== "invalid"),
tools,
activeTools: Object.keys(effectiveTools).filter((x) => x !== "invalid" && x !== "_dummy"),
tools: effectiveTools,
maxOutputTokens,
abortSignal: input.abort,
headers: {
Expand Down