@openagents-org/agent-launcher 0.2.172 → 0.2.175

package/src/adapters/hermes.js

@@ -428,6 +428,62 @@ class HermesAdapter extends BaseAdapter {
       await this.sendError(msgChannel, `Error processing message: ${e.message}`);
     }
   }
+
+  // ------------------------------------------------------------------
+  // Static: configure Hermes's native model config from LLM env vars
+  // ------------------------------------------------------------------
+
+  /**
+   * Point Hermes at a custom OpenAI-compatible endpoint from user-provided
+   * LLM_API_KEY / LLM_BASE_URL / LLM_MODEL values. Called by saveAgentEnv when
+   * type === 'hermes', same pattern as OpenClaw's configureNativeAuth.
+   *
+   * Hermes has no env-var path for a custom endpoint (OPENAI_BASE_URL is
+   * honored only for its built-in openai provider) — the supported form is the
+   * `model:` section of its config.yaml:
+   *   model: { default: <id>, provider: custom, base_url: <url>, api_key: <key> }
+   * Rather than hand-writing that YAML (and clobbering user config), drive
+   * Hermes's own `hermes config set` CLI: it owns the schema, merges into the
+   * existing file, and resolves its platform config path itself (~/.hermes on
+   * Unix, %LOCALAPPDATA%\hermes on native Windows, in-distro under WSL).
+   *
+   * Without this, a remotely-created hermes agent died on every message with
+   * "No inference provider configured" — the workspace's API-key field was a
+   * silent no-op for this type.
+   *
+   * Only acts when LLM_BASE_URL is set: a user signed in via `hermes setup`
+   * (the Nous login) must keep their own provider selection untouched.
+   */
+  static configureNativeAuth(env) {
+    const vals = env || {};
+    const baseUrl = String(vals.LLM_BASE_URL || '').trim().replace(/\/+$/, '');
+    if (!baseUrl) return { skipped: true, reason: 'no LLM_BASE_URL' };
+    const model = String(vals.LLM_MODEL || '').trim();
+    const apiKey = String(vals.LLM_API_KEY || '').trim();
+
+    // Borrow the adapter's binary resolution without running the constructor.
+    const probe = Object.create(HermesAdapter.prototype);
+    probe._log = () => {};
+    const bin = probe._findHermesBinary();
+    if (!bin) return { skipped: true, reason: 'hermes binary not found' };
+
+    const { execFileSync } = require('child_process');
+    const run = (args) => {
+      const spawnBin = probe._hermesViaWsl ? 'wsl.exe' : bin;
+      const spawnArgs = probe._hermesViaWsl ? ['-e', bin, ...args] : args;
+      execFileSync(spawnBin, spawnArgs, {
+        stdio: 'ignore',
+        timeout: 20000,
+        windowsHide: true,
+      });
+    };
+
+    run(['config', 'set', 'model.provider', 'custom']);
+    run(['config', 'set', 'model.base_url', baseUrl]);
+    if (model) run(['config', 'set', 'model.default', model]);
+    if (apiKey) run(['config', 'set', 'model.api_key', apiKey]);
+    return { configured: true };
+  }
 }
 
 module.exports = HermesAdapter;

package/src/adapters/index.js

@@ -23,6 +23,7 @@ const MiniSweAgentAdapter = require('./mini');
 const PiAdapter = require('./pi');
 const DeepSeekAdapter = require('./deepseek');
 const AntigravityAdapter = require('./antigravity');
+const CommandCodeAdapter = require('./commandcode');
 
 const ADAPTER_MAP = {
   openclaw: OpenClawAdapter,
@@ -43,11 +44,12 @@ const ADAPTER_MAP = {
   pi: PiAdapter,
   deepseek: DeepSeekAdapter,
   antigravity: AntigravityAdapter,
+  commandcode: CommandCodeAdapter,
 };
 
 /**
  * Create an adapter instance for the given agent type.
- * @param {string} type - Agent type (openclaw, claude, codex, opencode, nanoclaw, cursor, hermes, gemini, kimi, aider, goose, copilot, cline, amp, mini-swe-agent, pi, deepseek, antigravity)
+ * @param {string} type - Agent type (openclaw, claude, codex, opencode, nanoclaw, cursor, hermes, gemini, kimi, aider, goose, copilot, cline, amp, mini-swe-agent, pi, deepseek, antigravity, commandcode)
  * @param {object} opts - Adapter constructor options
  * @returns {BaseAdapter}
  */
@@ -79,6 +81,7 @@ module.exports = {
   PiAdapter,
   DeepSeekAdapter,
   AntigravityAdapter,
+  CommandCodeAdapter,
   createAdapter,
   ADAPTER_MAP,
 };

package/src/adapters/kimi-stream.js

@@ -0,0 +1,360 @@
+/**
+ * Pure, side-effect-free helpers for the Kimi Code CLI adapter.
+ *
+ * Everything here is I/O-free and deterministic so it can be unit-tested
+ * without spawning the `kimi` binary or touching the network: the JSONL
+ * stream parser, the message interpreter, secret redaction, error/version
+ * classification, the argument builder, and the env mapping that turns the
+ * launcher's KIMI_* fields into the CLI's KIMI_MODEL_* provider variables.
+ *
+ * Verified against Kimi Code CLI v0.39.1 (`kimi -p --output-format
+ * stream-json`). The stream is JSONL on stdout where each line is a message:
+ *   {role:"assistant", content, tool_calls?:[{type:"function",id,function:{name,arguments}}]}
+ *   {role:"tool", tool_call_id, content}
+ *   {role:"meta", type:"system.version", version}
+ *   {role:"meta", type:"session.resume_hint", session_id, command, content}
+ *   {role:"meta", type:"turn.step.retrying", failed_attempt, next_attempt,
+ *                 max_attempts, delay_ms, error_name, error_message}
+ * Fatal errors arrive on stderr as `error: ...` lines. Exit codes: 0 success,
+ * 1 permanent failure (config/auth/quota), 75 transient (retryable).
+ *
+ * IMPORTANT product distinction: two Moonshot products install a `kimi`
+ * binary. Kimi Code CLI (npm `@moonshot-ai/kimi-code`) versions are 0.x;
+ * the legacy Python `kimi-cli` (PyPI) is 1.x and has a DIFFERENT headless
+ * interface. A 1.x `kimi --version` therefore means the WRONG product.
+ */
+
+'use strict';
+
+// ---------------------------------------------------------------------------
+// Version / product classification
+// ---------------------------------------------------------------------------
+
+/** Pull a dotted version (e.g. "0.39.1") out of `kimi --version` output. */
+function parseKimiVersion(raw) {
+  if (!raw || typeof raw !== 'string') return null;
+  const m = raw.match(/(\d+)\.(\d+)\.(\d+)(?:[-.][0-9A-Za-z.]+)?/);
+  return m ? m[0] : null;
+}
+
+/**
+ * Classify `kimi --version` output into a product identity.
+ * @returns {{version: string|null, product: 'kimi-code'|'legacy'|null}}
+ *   'kimi-code' → the real Kimi Code CLI (0.x) this adapter drives
+ *   'legacy'    → the wound-down Python kimi-cli (1.x) — incompatible
+ *   null        → undetermined (unparseable output); proceed leniently
+ */
+function classifyKimiVersion(rawVersion) {
+  const version = parseKimiVersion(rawVersion);
+  if (!version) return { version: null, product: null };
+  const major = parseInt(version.split('.')[0], 10);
+  return { version, product: major === 0 ? 'kimi-code' : 'legacy' };
+}
+
+// ---------------------------------------------------------------------------
+// Secret redaction (same patterns as the other CLI adapters)
+// ---------------------------------------------------------------------------
+
+const SECRET_PATTERNS = [
+  /\bsk-[A-Za-z0-9._-]{8,}\b/g,
+  /\b(?:or|rk|gsk|ghp|gho|ghu|ghs|github_pat)[-_][A-Za-z0-9._-]{12,}\b/g,
+  /\b(?:Bearer|Basic)\s+[A-Za-z0-9._~+/=-]{8,}/gi,
+  /\bAKIA[0-9A-Z]{16}\b/g,
+];
+
+/** Redact obvious secret material from a free-text string. */
+function redactSecrets(text) {
+  if (text == null) return text;
+  let out = String(text);
+  for (const re of SECRET_PATTERNS) out = out.replace(re, '«redacted»');
+  return out;
+}
+
+/** Redact an argv array for logging. The prompt (arg after -p) is elided. */
+function redactArgs(args) {
+  const out = [];
+  for (let i = 0; i < args.length; i++) {
+    const a = args[i];
+    if (a === '-p' || a === '--prompt') {
+      out.push(a, '«prompt»');
+      i++;
+      continue;
+    }
+    out.push(redactSecrets(a));
+  }
+  return out;
+}
+
+// ---------------------------------------------------------------------------
+// JSONL stream parser
+// ---------------------------------------------------------------------------
+
+/** Line-buffered JSONL parser. push() returns complete parsed messages. */
+class KimiStreamParser {
+  constructor() {
+    this._buf = '';
+  }
+
+  /** @param {Buffer|string} chunk @returns {object[]} parsed messages */
+  push(chunk) {
+    this._buf += chunk.toString('utf-8');
+    const lines = this._buf.split('\n');
+    this._buf = lines.pop();
+    return parseLines(lines);
+  }
+
+  /** Flush any trailing partial line. */
+  flush() {
+    const rest = this._buf;
+    this._buf = '';
+    return parseLines([rest]);
+  }
+}
+
+function parseLines(lines) {
+  const out = [];
+  for (const line of lines) {
+    const t = line.trim();
+    if (!t || t[0] !== '{') continue;
+    try {
+      const obj = JSON.parse(t);
+      if (obj && typeof obj === 'object') out.push(obj);
+    } catch {
+      // Partial or non-JSON diagnostic line — skip.
+    }
+  }
+  return out;
+}
+
+// ---------------------------------------------------------------------------
+// Message interpretation
+// ---------------------------------------------------------------------------
+
+/** Flatten a message `content` (string or array of text blocks) to a string. */
+function contentText(content) {
+  if (typeof content === 'string') return content;
+  if (Array.isArray(content)) {
+    return content
+      .map((b) => (b && typeof b === 'object' && typeof b.text === 'string' ? b.text : ''))
+      .join('');
+  }
+  return '';
+}
+
+// Argument keys worth surfacing in a tool-status preview, best first.
+const PREVIEW_KEYS = ['command', 'path', 'file_path', 'pattern', 'url', 'query', 'prompt', 'description'];
+
+/** A short, redacted human preview for a tool call's JSON arguments. */
+function toolPreview(argsJson) {
+  let args;
+  try {
+    args = JSON.parse(argsJson);
+  } catch {
+    return '';
+  }
+  if (!args || typeof args !== 'object') return '';
+  for (const key of PREVIEW_KEYS) {
+    const v = args[key];
+    if (typeof v === 'string' && v.trim()) {
+      const one = redactSecrets(v.replace(/\s+/g, ' ').trim());
+      return one.length > 80 ? one.slice(0, 80) + '…' : one;
+    }
+  }
+  return '';
+}
+
+/**
+ * Map one parsed stream message to zero or more adapter events:
+ *   {kind:'text', text}                       assistant prose
+ *   {kind:'tool_start', name, preview}        a tool call was issued
+ *   {kind:'tool_result', id, text}            a tool finished
+ *   {kind:'session', sessionId}               resume hint (inline session id)
+ *   {kind:'retrying', attempt, maxAttempts, message}  provider retry
+ *   {kind:'version', version}                 stream preamble
+ */
+function interpretKimiMessage(msg) {
+  if (!msg || typeof msg !== 'object') return [];
+  const events = [];
+
+  if (msg.role === 'assistant') {
+    const text = contentText(msg.content).trim();
+    if (text) events.push({ kind: 'text', text });
+    if (Array.isArray(msg.tool_calls)) {
+      for (const tc of msg.tool_calls) {
+        const fn = tc && typeof tc === 'object' ? tc.function : null;
+        const name = fn && typeof fn.name === 'string' ? fn.name : 'tool';
+        events.push({ kind: 'tool_start', name, preview: toolPreview(fn && fn.arguments) });
+      }
+    }
+    return events;
+  }
+
+  if (msg.role === 'tool') {
+    events.push({
+      kind: 'tool_result',
+      id: typeof msg.tool_call_id === 'string' ? msg.tool_call_id : '',
+      text: contentText(msg.content),
+    });
+    return events;
+  }
+
+  if (msg.role === 'meta') {
+    if (msg.type === 'session.resume_hint' && typeof msg.session_id === 'string' && msg.session_id) {
+      events.push({ kind: 'session', sessionId: msg.session_id });
+    } else if (msg.type === 'turn.step.retrying') {
+      events.push({
+        kind: 'retrying',
+        attempt: typeof msg.failed_attempt === 'number' ? msg.failed_attempt : 0,
+        maxAttempts: typeof msg.max_attempts === 'number' ? msg.max_attempts : 0,
+        message: typeof msg.error_message === 'string' ? redactSecrets(msg.error_message) : '',
+      });
+    } else if (msg.type === 'system.version') {
+      events.push({ kind: 'version', version: typeof msg.version === 'string' ? msg.version : '' });
+    }
+    return events;
+  }
+
+  return events;
+}
+
+// ---------------------------------------------------------------------------
+// Argument + environment builders
+// ---------------------------------------------------------------------------
+
+/**
+ * Build the argv for one non-interactive run. Print mode (`-p`) already
+ * auto-approves tools — the CLI REJECTS combining it with --yolo/--auto/--plan
+ * (verified v0.39.1), so permission/plan flags must never be added here.
+ */
+function buildKimiArgs({ prompt, sessionId }) {
+  const args = [];
+  if (sessionId) args.push('-S', sessionId);
+  args.push('-p', prompt, '--output-format', 'stream-json');
+  return args;
+}
+
+const DEFAULT_KIMI_MODEL = 'kimi-k2.6';
+// The CLI defaults max_completion_tokens to the model's full context size
+// (262144), which OpenAI-compatible gateways reject (input+output > context).
+// A fixed, generous output budget works against both Moonshot and gateways.
+const DEFAULT_MAX_COMPLETION_TOKENS = '32768';
+
+/**
+ * Map the launcher's saved KIMI_* fields onto the env-var provider contract of
+ * Kimi Code CLI: setting KIMI_MODEL_NAME synthesizes an in-memory provider
+ * from KIMI_MODEL_API_KEY (required) + KIMI_MODEL_BASE_URL +
+ * KIMI_MODEL_PROVIDER_TYPE, bypassing `kimi login`.
+ *
+ * With no API key configured, the env is passed through untouched so the
+ * CLI's own `kimi login` credentials / config.toml apply.
+ *
+ * @returns {{env: object, viaEnvProvider: boolean}}
+ */
+function buildKimiEnv(agentEnv) {
+  const env = { ...(agentEnv || {}) };
+  const apiKey =
+    env.KIMI_MODEL_API_KEY ||
+    env.KIMI_API_KEY ||
+    env.MOONSHOT_API_KEY ||
+    '';
+  if (!apiKey) return { env, viaEnvProvider: false };
+
+  env.KIMI_MODEL_API_KEY = apiKey;
+  if (!env.KIMI_MODEL_NAME) env.KIMI_MODEL_NAME = env.KIMI_MODEL || DEFAULT_KIMI_MODEL;
+  const baseUrl = env.KIMI_MODEL_BASE_URL || env.KIMI_BASE_URL;
+  if (baseUrl) env.KIMI_MODEL_BASE_URL = baseUrl.replace(/\/$/, '');
+  if (!env.KIMI_MODEL_PROVIDER_TYPE) env.KIMI_MODEL_PROVIDER_TYPE = 'kimi';
+  if (!env.KIMI_MODEL_MAX_COMPLETION_TOKENS) {
+    env.KIMI_MODEL_MAX_COMPLETION_TOKENS = DEFAULT_MAX_COMPLETION_TOKENS;
+  }
+  return { env, viaEnvProvider: true };
+}
+
+// ---------------------------------------------------------------------------
+// Error classification
+// ---------------------------------------------------------------------------
+
+// Kimi Code CLI's documented exit codes.
+const EXIT_PERMANENT = 1;
+const EXIT_TRANSIENT = 75;
+
+/** Extract the message from `error: ...` lines on stderr (last one wins). */
+function extractStderrError(stderrText) {
+  if (!stderrText) return '';
+  let last = '';
+  for (const line of String(stderrText).split('\n')) {
+    const m = line.match(/^\s*error:\s*(.+)$/i);
+    if (m) last = m[1].trim();
+  }
+  // Strip the CLI's wrapper prefix so the user sees the actual cause.
+  return redactSecrets(last.replace(/^failed to run prompt:\s*/i, ''));
+}
+
+/**
+ * Classify a failed run into a user-facing message.
+ * @param {{code: number|null, signal: string|null, stderrText: string, retryMessage: string}} info
+ * @returns {{kind: string, userMessage: string}}
+ */
+function classifyKimiError(info) {
+  const { code, signal, stderrText, retryMessage } = info || {};
+  const detail = extractStderrError(stderrText) || redactSecrets(retryMessage || '');
+
+  if (/auth_error|401|unauthorized|invalid.{0,20}(api.?key|token)|credential/i.test(detail)) {
+    return {
+      kind: 'auth',
+      userMessage:
+        'Kimi authentication failed. Check KIMI_API_KEY in the launcher, or run `kimi login` in a terminal.' +
+        (detail ? `\n\nDetails: ${detail}` : ''),
+    };
+  }
+  if (/no model configured|llm not set|kimi_model_api_key is missing/i.test(detail)) {
+    return {
+      kind: 'config',
+      userMessage:
+        'Kimi Code CLI is not configured. Set KIMI_API_KEY (plus optional KIMI_BASE_URL / KIMI_MODEL) ' +
+        'in the launcher, or run `kimi login` in a terminal.',
+    };
+  }
+  if (/rate.?limit|429|quota|insufficient.{0,20}balance/i.test(detail)) {
+    return {
+      kind: 'rate_limit',
+      userMessage: `Kimi hit a rate/quota limit. Please retry shortly.${detail ? `\n\nDetails: ${detail}` : ''}`,
+    };
+  }
+  if (/context length|maximum context|too many tokens/i.test(detail)) {
+    return {
+      kind: 'context',
+      userMessage: `The request exceeded the model's context limits.${detail ? `\n\nDetails: ${detail}` : ''}`,
+    };
+  }
+  if (code === EXIT_TRANSIENT) {
+    return {
+      kind: 'transient',
+      userMessage: `Kimi hit a temporary provider error — please try again.${detail ? `\n\nDetails: ${detail}` : ''}`,
+    };
+  }
+  if (detail) {
+    return { kind: 'error', userMessage: `Kimi failed: ${detail}` };
+  }
+  const why = signal ? `terminated by signal ${signal}` : `exited with code ${code}`;
+  return { kind: 'error', userMessage: `Kimi Code CLI ${why}.` };
+}
+
+module.exports = {
+  parseKimiVersion,
+  classifyKimiVersion,
+  redactSecrets,
+  redactArgs,
+  KimiStreamParser,
+  interpretKimiMessage,
+  contentText,
+  toolPreview,
+  buildKimiArgs,
+  buildKimiEnv,
+  extractStderrError,
+  classifyKimiError,
+  DEFAULT_KIMI_MODEL,
+  DEFAULT_MAX_COMPLETION_TOKENS,
+  EXIT_PERMANENT,
+  EXIT_TRANSIENT,
+};