polycli-runtime

@bbingz/polycli-runtime

@bbingz/polycli-runtime 是 polycli v2 的 provider runtime 集成层。

它提供八家 adapter：

• claude
• copilot
• opencode
• pi
• gemini
• kimi
• qwen
• minimax

Scope

这个包负责：

• provider registry
• availability / auth probes
• provider-specific args builder
• foreground prompt execution
• streaming execution
• stream / log parsing

这个包不负责：

• 持久化 job state
• background job orchestration
• command markdown / prompt 模板
• 把多家 provider 硬压成同一种协议

Public Surface

统一入口：

• getProviderRuntime()
• listProviderRuntimes()
• runProviderPrompt()
• runProviderPromptStreaming()

Provider registry 常量：

• PROVIDER_IDS
• PROVIDER_OPERATION_NAMES

PROVIDER_OPERATION_NAMES 当前明确只有：

• prompt

也就是说，这一层只承诺 runtime 级 prompt 执行能力，不承诺已经内置 command-level 的 task / review / adversarial-review 语义编排。

Provider-specific helpers：

• buildClaudeInvocation()
• buildCopilotInvocation()
• buildGeminiInvocation()
• buildKimiInvocation()
• buildQwenInvocation()
• buildQwenEnv()
• buildMiniMaxInvocation()
• buildOpenCodeInvocation()
• buildPiInvocation()
• parseClaudeJsonResult()
• parseClaudeStreamText()
• parseCopilotStreamText()
• parseGeminiStreamText()
• parseKimiStreamText()
• parseQwenStreamText()
• parseMiniMaxResponseBlocks()
• parseOpenCodeJsonResult()
• parseOpenCodeStreamText()
• parsePiStreamText()
• extractMiniMaxResponseFromLogText()
• extractMiniMaxLogPath()
• extractClaudeText()
• extractCopilotText()
• extractGeminiText()
• extractKimiText()
• extractOpenCodeText()
• extractPiText()
• stripAnsiSgr()

Provider-specific runtime methods：

• getClaudeAvailability() / getClaudeAuthStatus() / runClaudePrompt() / runClaudePromptStreaming()
• getCopilotAvailability() / getCopilotAuthStatus() / runCopilotPrompt() / runCopilotPromptStreaming()
• getGeminiAvailability() / getGeminiAuthStatus() / runGeminiPrompt() / runGeminiPromptStreaming()
• getKimiAvailability() / getKimiAuthStatus() / runKimiPrompt() / runKimiPromptStreaming()
• getQwenAvailability() / getQwenAuthStatus() / runQwenPrompt() / runQwenPromptStreaming()
• getMiniMaxAvailability() / getMiniMaxAuthStatus() / runMiniMaxPrompt() / runMiniMaxPromptStreaming()
• getOpenCodeAvailability() / getOpenCodeAuthStatus() / runOpenCodePrompt() / runOpenCodePromptStreaming()
• getPiAvailability() / getPiAuthStatus() / runPiPrompt() / runPiPromptStreaming()

Intentional omission:

• listModels is not in the current public surface.
• Reason: upstream CLIs do not expose a stable enough cross-provider listing contract yet.
• Decision: keep model enumeration out of the runtime API until it can be implemented as a real capability instead of provider-specific best-effort wrappers.

Example

import {
  getProviderRuntime,
  runProviderPrompt
} from "./src/index.js";

const runtime = getProviderRuntime("qwen");
const availability = runtime.getAvailability(process.cwd());

const result = await runProviderPrompt({
  provider: "gemini",
  prompt: "ping",
  cwd: process.cwd()
});