feat: add codex context budget controls

cli.js

@@ -93,6 +93,8 @@ const RECENT_CONFIGS_FILE = path.join(CONFIG_DIR, 'recent-configs.json');
 const WORKFLOW_DEFINITIONS_FILE = path.join(CONFIG_DIR, 'codexmate-workflows.json');
 const WORKFLOW_RUNS_FILE = path.join(CONFIG_DIR, 'codexmate-workflow-runs.jsonl');
 const DEFAULT_CLAUDE_MODEL = 'glm-4.7';
+const DEFAULT_MODEL_CONTEXT_WINDOW = 190000;
+const DEFAULT_MODEL_AUTO_COMPACT_TOKEN_LIMIT = 185000;
 const CODEX_BACKUP_NAME = 'codex-config';
 
 const DEFAULT_MODELS = ['gpt-5.3-codex', 'gpt-5.1-codex-max', 'gpt-4-turbo', 'gpt-4'];
@@ -226,6 +228,8 @@ function resolveWebHost(options = {}) {
 
 const EMPTY_CONFIG_FALLBACK_TEMPLATE = `model = "gpt-5.3-codex"
 model_reasoning_effort = "high"
+model_context_window = ${DEFAULT_MODEL_CONTEXT_WINDOW}
+model_auto_compact_token_limit = ${DEFAULT_MODEL_AUTO_COMPACT_TOKEN_LIMIT}
 disable_response_storage = true
 approval_policy = "never"
 sandbox_mode = "danger-full-access"
@@ -3168,6 +3172,8 @@ function buildDefaultConfigContent(initializedAt) {
 
 model_provider = "openai"
 model = "${defaultModel}"
+model_context_window = ${DEFAULT_MODEL_CONTEXT_WINDOW}
+model_auto_compact_token_limit = ${DEFAULT_MODEL_AUTO_COMPACT_TOKEN_LIMIT}
 
 [model_providers.openai]
 name = "openai"
@@ -3333,6 +3339,45 @@ function applyReasoningEffortToTemplate(template, reasoningEffort) {
  return content;
 }
 
+function normalizePositiveIntegerParam(value) {
+ if (value === undefined || value === null) {
+ return null;
+ }
+ const text = typeof value === 'number'
+ ? String(value)
+ : (typeof value === 'string' ? value.trim() : String(value).trim());
+ if (!text) {
+ return null;
+ }
+ if (!/^\d+$/.test(text)) {
+ return null;
+ }
+ const parsed = Number.parseInt(text, 10);
+ if (!Number.isSafeInteger(parsed) || parsed <= 0) {
+ return null;
+ }
+ return parsed;
+}
+
+function applyPositiveIntegerConfigToTemplate(template, key, value) {
+ let content = typeof template === 'string' ? template : '';
+ const normalized = normalizePositiveIntegerParam(value);
+ if (!key || normalized === null) {
+ return content;
+ }
+
+ const hasBom = content.charCodeAt(0) === 0xFEFF;
+ const lineEnding = content.includes('\r\n') ? '\r\n' : '\n';
+ if (hasBom) {
+ content = content.slice(1);
+ }
+ const escapedKey = key.replace(/[.*+?^${}()|[\]\\]/g, '\\$&');
+ const pattern = new RegExp(`^\\s*${escapedKey}\\s*=\\s*[^\\n]*\\n?`, 'gmi');
+ content = content.replace(pattern, '');
+ content = content.replace(new RegExp(`^(?:[\\t ]*${lineEnding})+`), '');
+ return `${hasBom ? '\uFEFF' : ''}${key} = ${normalized}${lineEnding}${content}`;
+}
+
 function getConfigTemplate(params = {}) {
  let content = EMPTY_CONFIG_FALLBACK_TEMPLATE;
  if (fs.existsSync(CONFIG_FILE)) {
@@ -3343,6 +3388,20 @@ function getConfigTemplate(params = {}) {
  }
  } catch (e) {}
  }
+ if (
+ params.modelAutoCompactTokenLimit !== undefined
+ && params.modelAutoCompactTokenLimit !== null
+ && normalizePositiveIntegerParam(params.modelAutoCompactTokenLimit) === null
+ ) {
+ return { error: 'modelAutoCompactTokenLimit must be a positive integer' };
+ }
+ if (
+ params.modelContextWindow !== undefined
+ && params.modelContextWindow !== null
+ && normalizePositiveIntegerParam(params.modelContextWindow) === null
+ ) {
+ return { error: 'modelContextWindow must be a positive integer' };
+ }
  const selectedProvider = typeof params.provider === 'string' ? params.provider.trim() : '';
  const selectedModel = typeof params.model === 'string' ? params.model.trim() : '';
  let template = normalizeTopLevelConfigWithTemplate(content, selectedProvider, selectedModel);
@@ -3352,11 +3411,54 @@ function getConfigTemplate(params = {}) {
  if (typeof params.reasoningEffort === 'string') {
  template = applyReasoningEffortToTemplate(template, params.reasoningEffort);
  }
+ if (!/^\s*model_auto_compact_token_limit\s*=.*$/m.test(template)) {
+ template = applyPositiveIntegerConfigToTemplate(
+ template,
+ 'model_auto_compact_token_limit',
+ DEFAULT_MODEL_AUTO_COMPACT_TOKEN_LIMIT
+ );
+ }
+ if (!/^\s*model_context_window\s*=.*$/m.test(template)) {
+ template = applyPositiveIntegerConfigToTemplate(
+ template,
+ 'model_context_window',
+ DEFAULT_MODEL_CONTEXT_WINDOW
+ );
+ }
+ if (params.modelAutoCompactTokenLimit !== undefined) {
+ template = applyPositiveIntegerConfigToTemplate(
+ template,
+ 'model_auto_compact_token_limit',
+ params.modelAutoCompactTokenLimit
+ );
+ }
+ if (params.modelContextWindow !== undefined) {
+ template = applyPositiveIntegerConfigToTemplate(
+ template,
+ 'model_context_window',
+ params.modelContextWindow
+ );
+ }
  return {
  template
  };
 }
 
+function readPositiveIntegerConfigValue(config, key) {
+ const options = arguments[2] && typeof arguments[2] === 'object' ? arguments[2] : {};
+ const useDefaultsWhenMissing = options.useDefaultsWhenMissing !== false;
+ if (!config || typeof config !== 'object' || !key) {
+ return '';
+ }
+ const raw = config[key];
+ if (raw === undefined && useDefaultsWhenMissing) {
+ if (key === 'model_context_window') return DEFAULT_MODEL_CONTEXT_WINDOW;
+ if (key === 'model_auto_compact_token_limit') return DEFAULT_MODEL_AUTO_COMPACT_TOKEN_LIMIT;
+ }
+ const normalized = normalizePositiveIntegerParam(raw);
+ return normalized === null ? '' : normalized;
+}
+
 function applyConfigTemplate(params = {}) {
  const template = typeof params.template === 'string' ? params.template : '';
  if (!template.trim()) {
@@ -3370,6 +3472,20 @@ function applyConfigTemplate(params = {}) {
  return { error: `模板 TOML 解析失败: ${e.message}` };
  }
 
+ if (
+ Object.prototype.hasOwnProperty.call(parsed, 'model_context_window')
+ && normalizePositiveIntegerParam(parsed.model_context_window) === null
+ ) {
+ return { error: '模板中的 model_context_window 必须是正整数' };
+ }
+
+ if (
+ Object.prototype.hasOwnProperty.call(parsed, 'model_auto_compact_token_limit')
+ && normalizePositiveIntegerParam(parsed.model_auto_compact_token_limit) === null
+ ) {
+ return { error: '模板中的 model_auto_compact_token_limit 必须是正整数' };
+ }
+
  if (!parsed.model_provider || typeof parsed.model_provider !== 'string') {
  return { error: '模板缺少 model_provider' };
  }
@@ -9976,22 +10092,38 @@ function createWebServer({ htmlPath, assetsDir, webDir, host, port, openBrowser
  let result;
 
  switch (action) {
- case 'status':
+ case 'status': {
  const statusConfigResult = readConfigOrVirtualDefault();
  const config = statusConfigResult.config;
  const serviceTier = typeof config.service_tier === 'string' ? config.service_tier.trim() : '';
  const modelReasoningEffort = typeof config.model_reasoning_effort === 'string' ? config.model_reasoning_effort.trim() : '';
+ const budgetReadOptions = {
+ useDefaultsWhenMissing: !hasConfigLoadError(statusConfigResult)
+ };
+ const modelContextWindow = readPositiveIntegerConfigValue(
+ config,
+ 'model_context_window',
+ budgetReadOptions
+ );
+ const modelAutoCompactTokenLimit = readPositiveIntegerConfigValue(
+ config,
+ 'model_auto_compact_token_limit',
+ budgetReadOptions
+ );
  result = {
  provider: config.model_provider || '未设置',
  model: config.model || '未设置',
  serviceTier,
  modelReasoningEffort,
+ modelContextWindow,
+ modelAutoCompactTokenLimit,
  configReady: !statusConfigResult.isVirtual,
  configErrorType: statusConfigResult.errorType || '',
  configNotice: statusConfigResult.reason || '',
  initNotice: consumeInitNotice()
  };
  break;
+ }
  case 'install-status':
  result = buildInstallStatusReport();
  break;
@@ -11464,11 +11596,26 @@ function buildMcpStatusPayload() {
  const config = statusConfigResult.config;
  const serviceTier = typeof config.service_tier === 'string' ? config.service_tier.trim() : '';
  const modelReasoningEffort = typeof config.model_reasoning_effort === 'string' ? config.model_reasoning_effort.trim() : '';
+ const budgetReadOptions = {
+ useDefaultsWhenMissing: !hasConfigLoadError(statusConfigResult)
+ };
+ const modelContextWindow = readPositiveIntegerConfigValue(
+ config,
+ 'model_context_window',
+ budgetReadOptions
+ );
+ const modelAutoCompactTokenLimit = readPositiveIntegerConfigValue(
+ config,
+ 'model_auto_compact_token_limit',
+ budgetReadOptions
+ );
  return {
  provider: config.model_provider || '未设置',
  model: config.model || '未设置',
  serviceTier,
  modelReasoningEffort,
+ modelContextWindow,
+ modelAutoCompactTokenLimit,
  configReady: !statusConfigResult.isVirtual,
  configErrorType: statusConfigResult.errorType || '',
  configNotice: statusConfigResult.reason || '',
@@ -11566,6 +11713,8 @@ const BUILTIN_WORKFLOW_DEFINITIONS = Object.freeze({
  model: { type: 'string' },
  serviceTier: { type: 'string' },
  reasoningEffort: { type: 'string' },
+ modelContextWindow: { type: ['string', 'number'] },
+ modelAutoCompactTokenLimit: { type: ['string', 'number'] },
  apply: { type: 'boolean' }
  },
  required: ['provider'],
@@ -11580,7 +11729,9 @@ const BUILTIN_WORKFLOW_DEFINITIONS = Object.freeze({
  provider: '{{input.provider}}',
  model: '{{input.model}}',
  serviceTier: '{{input.serviceTier}}',
- reasoningEffort: '{{input.reasoningEffort}}'
+ reasoningEffort: '{{input.reasoningEffort}}',
+ modelContextWindow: '{{input.modelContextWindow}}',
+ modelAutoCompactTokenLimit: '{{input.modelAutoCompactTokenLimit}}'
  }
  },
  {
@@ -12149,15 +12300,17 @@ function createMcpTools(options = {}) {
 
  pushTool({
  name: 'codexmate.config.template.get',
- description: 'Get Codex config template with optional provider/model/service tier/reasoning effort.',
+ description: 'Get Codex config template with optional provider/model/service tier/reasoning effort/context budget.',
  readOnly: true,
  inputSchema: {
  type: 'object',
  properties: {
  provider: { type: 'string' },
  model: { type: 'string' },
  serviceTier: { type: 'string' },
- reasoningEffort: { type: 'string' }
+ reasoningEffort: { type: 'string' },
+ modelContextWindow: { type: ['string', 'number'] },
+ modelAutoCompactTokenLimit: { type: ['string', 'number'] }
  },
  additionalProperties: false
  },

tests/e2e/test-config.js

@@ -22,6 +22,10 @@ module.exports = async function testConfig(ctx) {
  assert(typeof apiStatus.configReady === 'boolean', 'api status configReady missing');
  assert('modelReasoningEffort' in apiStatus, 'api status modelReasoningEffort missing');
  assert('serviceTier' in apiStatus, 'api status serviceTier missing');
+ assert('modelContextWindow' in apiStatus, 'api status modelContextWindow missing');
+ assert('modelAutoCompactTokenLimit' in apiStatus, 'api status modelAutoCompactTokenLimit missing');
+ assert(apiStatus.modelContextWindow === 190000, 'api status modelContextWindow mismatch');
+ assert(apiStatus.modelAutoCompactTokenLimit === 185000, 'api status modelAutoCompactTokenLimit mismatch');
 
  // ========== List API Tests ==========
  const apiList = await api('list');
@@ -83,26 +87,78 @@ module.exports = async function testConfig(ctx) {
  assert(typeof templateReasoningXhigh.template === 'string', 'get-config-template(reasoning xhigh) missing template');
  assert(/^\s*model_reasoning_effort\s*=\s*"xhigh"\s*$/m.test(templateReasoningXhigh.template), 'get-config-template(reasoning xhigh) missing model_reasoning_effort');
 
+ // ========== Get Config Template Tests - Context Budget ==========
+ const templateContextBudget = await api('get-config-template', {
+ provider: 'shadow',
+ model: 'shadow-model',
+ modelContextWindow: 200000,
+ modelAutoCompactTokenLimit: 195000
+ });
+ assert(typeof templateContextBudget.template === 'string', 'get-config-template(context budget) missing template');
+ assert(templateContextBudget.template.includes('model_provider = "shadow"'), 'get-config-template(context budget) missing provider override');
+ assert(templateContextBudget.template.includes('model = "shadow-model"'), 'get-config-template(context budget) missing model override');
+ assert(/^\s*model_context_window\s*=\s*200000\s*$/m.test(templateContextBudget.template), 'get-config-template(context budget) missing model_context_window');
+ assert(/^\s*model_auto_compact_token_limit\s*=\s*195000\s*$/m.test(templateContextBudget.template), 'get-config-template(context budget) missing model_auto_compact_token_limit');
+
  // ========== Get Config Template Tests - Combined ==========
  const templateCombined = await api('get-config-template', {
  provider: 'shadow',
  model: 'shadow-model',
  serviceTier: 'fast',
- reasoningEffort: 'high'
+ reasoningEffort: 'high',
+ modelContextWindow: 190000,
+ modelAutoCompactTokenLimit: 185000
  });
  assert(typeof templateCombined.template === 'string', 'get-config-template(combined) missing template');
  assert(/^\s*service_tier\s*=\s*"fast"\s*$/m.test(templateCombined.template), 'get-config-template(combined) missing service_tier');
  assert(/^\s*model_reasoning_effort\s*=\s*"high"\s*$/m.test(templateCombined.template), 'get-config-template(combined) missing model_reasoning_effort');
+ assert(/^\s*model_context_window\s*=\s*190000\s*$/m.test(templateCombined.template), 'get-config-template(combined) missing model_context_window');
+ assert(/^\s*model_auto_compact_token_limit\s*=\s*185000\s*$/m.test(templateCombined.template), 'get-config-template(combined) missing model_auto_compact_token_limit');
 
  const templateCombinedXhigh = await api('get-config-template', {
  provider: 'shadow',
  model: 'shadow-model',
  serviceTier: 'fast',
- reasoningEffort: 'xhigh'
+ reasoningEffort: 'xhigh',
+ modelContextWindow: 210000,
+ modelAutoCompactTokenLimit: 200000
  });
  assert(typeof templateCombinedXhigh.template === 'string', 'get-config-template(combined xhigh) missing template');
  assert(/^\s*service_tier\s*=\s*"fast"\s*$/m.test(templateCombinedXhigh.template), 'get-config-template(combined xhigh) missing service_tier');
  assert(/^\s*model_reasoning_effort\s*=\s*"xhigh"\s*$/m.test(templateCombinedXhigh.template), 'get-config-template(combined xhigh) missing model_reasoning_effort');
+ assert(/^\s*model_context_window\s*=\s*210000\s*$/m.test(templateCombinedXhigh.template), 'get-config-template(combined xhigh) missing model_context_window');
+ assert(/^\s*model_auto_compact_token_limit\s*=\s*200000\s*$/m.test(templateCombinedXhigh.template), 'get-config-template(combined xhigh) missing model_auto_compact_token_limit');
+
+ // ========== Apply Config Template Validation Tests ==========
+ const invalidContextBudgetApply = await api('apply-config-template', {
+ template: `model_provider = "shadow"
+model = "shadow-model"
+model_context_window = 0
+
+[model_providers.shadow]
+base_url = "https://example.test/v1"
+preferred_auth_method = "shadow-key"
+`
+ });
+ assert(
+ invalidContextBudgetApply.error === '模板中的 model_context_window 必须是正整数',
+ 'apply-config-template should reject invalid model_context_window'
+ );
+
+ const invalidAutoCompactApply = await api('apply-config-template', {
+ template: `model_provider = "shadow"
+model = "shadow-model"
+model_auto_compact_token_limit = "abc"
+
+[model_providers.shadow]
+base_url = "https://example.test/v1"
+preferred_auth_method = "shadow-key"
+`
+ });
+ assert(
+ invalidAutoCompactApply.error === '模板中的 model_auto_compact_token_limit 必须是正整数',
+ 'apply-config-template should reject invalid model_auto_compact_token_limit'
+ );
 
  // ========== Export Config Tests ==========
  const exportResult = await api('export-config', { includeKeys: true });
@@ -337,6 +393,21 @@ module.exports = async function testConfig(ctx) {
  await waitForServer(legacyPort);
 
  const legacyApi = (action, params) => postJson(legacyPort, { action, params }, 2000);
+ const legacyStatus = await legacyApi('status');
+ assert(legacyStatus.modelContextWindow === 190000, 'legacy status should default modelContextWindow');
+ assert(
+ legacyStatus.modelAutoCompactTokenLimit === 185000,
+ 'legacy status should default modelAutoCompactTokenLimit'
+ );
+ const legacyTemplateDefaults = await legacyApi('get-config-template', {});
+ assert(
+ /^\s*model_context_window\s*=\s*190000\s*$/m.test(legacyTemplateDefaults.template),
+ 'legacy get-config-template should restore default model_context_window'
+ );
+ assert(
+ /^\s*model_auto_compact_token_limit\s*=\s*185000\s*$/m.test(legacyTemplateDefaults.template),
+ 'legacy get-config-template should restore default model_auto_compact_token_limit'
+ );
  const legacyAddDup = await legacyApi('add-provider', {
  name: 'foo.bar',
  url: 'https://dup.example.com/v1',