Files
OmniRoute/tests/unit/codebuddy-reasoning-optin.test.ts
Ababil 55c2b35eb7 fix(codebuddy-cn): replace agent system prompts to bypass Tencent content filter (#9723)
* fix(codebuddy-cn): replace agent system prompts to bypass Tencent content filter

Tencent's content filter flags CLI agent system prompts (e.g. 'You are
Claude Code, Anthropic's official CLI...') as prompt injection / sensitive
content and rejects the entire request with error:

  抱歉,系统检测到您当前输入的信息存在敏感内容,我无法响应您的请求

This patch adds detection and replacement logic to the CodeBuddyCnExecutor:

- Regex-based identity marker detection (Claude Code, Cursor, Windsurf,
  Cline, Aider, Copilot, Cody, etc.) + length catch-all (>2000 chars)
- Handles both top-level 'system' field (Anthropic format) and messages
  array with role:'system' (OpenAI format)
- Preserves original content shape (string vs typed content blocks)
- Strips oversized tool descriptions (>64KB) that can also trigger the filter
- Replaces with neutral prompt, leaving legitimate user prompts untouched

Based on approach from rafilajhh/9router commit 7f7d7ce.

* test(codebuddy-cn): add regression coverage for system prompt replacement

Co-authored-by: diegosouzapw <8016841+diegosouzapw@users.noreply.github.com>

---------

Co-authored-by: diegosouzapw <diegosouzapw@users.noreply.github.com>
Co-authored-by: diegosouzapw <8016841+diegosouzapw@users.noreply.github.com>
2026-08-11 09:06:47 -03:00

100 lines
3.3 KiB
TypeScript

// #2071 — CodeBuddy forced reasoning_effort:"medium" + reasoning_summary:"auto"
// on requests where the client never asked for reasoning, tripping CodeBuddy's
// content filter ("model return error"). Reasoning params must be opt-in.
import { describe, it } from "node:test";
import assert from "node:assert/strict";
import { CodeBuddyCnExecutor } from "../../open-sse/executors/codebuddy-cn.ts";
describe("CodeBuddyCnExecutor reasoning params are opt-in (#2071)", () => {
const exec = new CodeBuddyCnExecutor();
it("does NOT force reasoning when the client did not request it", () => {
const out = exec.transformRequest(
"glm-5.2",
{ messages: [{ role: "user", content: "hi" }] },
false,
{}
) as Record<string, unknown>;
assert.equal(out.reasoning_effort, undefined);
assert.equal(out.reasoning_summary, undefined);
});
it("mirrors reasoning_summary:auto when the client explicitly requested reasoning", () => {
const out = exec.transformRequest(
"glm-5.2",
{ messages: [{ role: "user", content: "hi" }], reasoning_effort: "high" },
false,
{}
) as Record<string, unknown>;
assert.equal(out.reasoning_effort, "high");
assert.equal(out.reasoning_summary, "auto");
});
it("omits reasoning_effort for none/off and adds no reasoning_summary", () => {
const out = exec.transformRequest(
"glm-5.2",
{ messages: [{ role: "user", content: "hi" }], reasoning_effort: "none" },
false,
{}
) as Record<string, unknown>;
assert.equal(out.reasoning_effort, undefined);
assert.equal(out.reasoning_summary, undefined);
});
it("replaces agent system prompts in top-level system field", () => {
const out = exec.transformRequest(
"glm-5.2",
{
system: "You are Claude Code, Anthropic's official CLI for software engineering",
messages: [{ role: "user", content: "check this" }],
},
false,
{}
) as Record<string, unknown>;
assert.equal(
out.system,
"You are a helpful AI assistant that helps with software engineering tasks.",
);
});
it("preserves non-agent system prompts in top-level system field", () => {
const out = exec.transformRequest(
"glm-5.2",
{
system: "You are a helpful coding tutor for beginner developers.",
messages: [{ role: "user", content: "check this" }],
},
false,
{}
) as Record<string, unknown>;
assert.equal(out.system, "You are a helpful coding tutor for beginner developers.");
});
it("replaces agent system prompts in messages array preserving typed content shape", () => {
const out = exec.transformRequest(
"glm-5.2",
{
messages: [
{
role: "system",
content: [{ type: "text", text: "You are Cursor, an AI coding agent" }],
},
{ role: "user", content: "Hello" },
],
},
false,
{}
) as Record<string, unknown>;
const msgs = out.messages as Array<{ role: string; content?: unknown }>;
const system = msgs.find((m) => m.role === "system");
assert.deepEqual(system, {
role: "system",
content: [{ type: "text", text: "You are a helpful AI assistant that helps with software engineering tasks." }],
});
assert.equal(system && (system as { content?: unknown }).content, msgs[0].content);
});
});