Skip to content

armgit / Orkas

Public Imported from GitHub
0

Orkas is an open-source (MIT), local-first AI team for your desktop: a Commander turns your goal into to-dos and coordinates 16 built-in specialist agents plus Claude Code, Codex, OpenCode, OpenClaw and Hermes, with projects, scheduled automations and connectors. Free app for macOS & Windows; built-in models or your own keys.

Default metacognition prompts + reflection digest to English

aiservice8c6b555ba087
diff --git a/src/core-agent/src/evolution/metacognition.ts b/src/core-agent/src/evolution/metacognition.ts
index 782ef963..78c702cd 100644
--- a/src/core-agent/src/evolution/metacognition.ts
+++ b/src/core-agent/src/evolution/metacognition.ts
@@ -165,7 +165,9 @@ export function shouldReflect(
* listed in COMPETENCE.md. Simple keyword matching.
*/
function hitsKnownWeakness(metrics: RunMetrics, competence: string): boolean {
- // Look for a "已知弱点" or "weaknesses" section in competence
+ // Look for a "weaknesses" section (or its Chinese equivalent "已知弱点",
+ // kept so user-authored COMPETENCE.md from earlier zh-default builds still
+ // matches without a migration pass).
const weaknessSection = extractSection(competence, ['已知弱点', 'weaknesses', 'weak']);
if (!weaknessSection) return false;
@@ -218,83 +220,83 @@ export function buildAdaptiveReviewPrompt(
skillHealthReport?: string,
conversationDigest?: string,
): string {
- const base = '根据以下对话摘要,考虑改进你的技能和自我认知。\n\n'
- + '**重要:网络超时、连接中断、速率限制等瞬态错误是环境问题,不是技能或能力缺陷。'
- + '遇到这类错误时不要在 COMPETENCE.md 标记为弱点,不要修改或删除相关技能,'
- + '不要在 LEARNING_STRATEGIES.md 建议回避该工具。**\n\n';
+ const base = 'Based on the conversation digest below, consider improving your skills and self-awareness.\n\n'
+ + '**Important: transient errors (network timeouts, dropped connections, rate limits) are environmental issues, not skill or capability deficits. '
+ + 'Do not mark them as weaknesses in COMPETENCE.md, do not modify or delete the related skills, '
+ + 'and do not advise avoiding the tool in LEARNING_STRATEGIES.md.**\n\n';
let focus: string;
switch (primaryFocus) {
case 'error_recovery':
- focus = '你在这个任务中从错误中恢复了。'
- + '请把解决方法提取为可复用的技能,重点记录让恢复成功的非显而易见的步骤。';
+ focus = 'You recovered from an error during this task. '
+ + 'Extract the resolution into a reusable skill, focusing on the non-obvious steps that made recovery succeed.';
break;
case 'user_correction':
- focus = '用户纠正了你的做法。'
- + '请识别你做错了什么,创建/更新技能来记录正确的做法。'
- + '同时更新 COMPETENCE.md 记录这个弱点。';
+ focus = 'The user corrected your approach. '
+ + 'Identify what you got wrong, then create or update a skill that captures the correct way. '
+ + 'Also update COMPETENCE.md to record this weakness.';
break;
case 'skill_ineffective':
- focus = '你加载的技能没有帮到这个任务。'
- + '分析原因:技能过时了?太泛化?还是不匹配?'
- + '请修补或删除无效的技能。';
+ focus = 'A skill you loaded did not help on this task. '
+ + 'Diagnose why: outdated? too generic? mismatched? '
+ + 'Patch or delete the ineffective skill.';
break;
case 'known_weakness':
- focus = '这个任务命中了你的已知弱点。'
- + '重点加强这个领域。如果你成功了,更新 COMPETENCE.md 反映进步。';
+ focus = 'This task hit one of your known weaknesses. '
+ + 'Strengthen this area. If you succeeded, update COMPETENCE.md to reflect the progress.';
break;
case 'weakness_succeeded':
- focus = '你之前标记为弱点的能力在这次任务中表现正常。'
- + '请更新 COMPETENCE.md:移除或降级该弱点条目。'
- + '不要修改或删除相关技能——它们工作正常。';
+ focus = 'A capability you previously flagged as a weakness performed normally on this task. '
+ + 'Update COMPETENCE.md: remove or downgrade that weakness entry. '
+ + 'Do not modify or delete the related skills — they are working as expected.';
break;
case 'periodic_review':
- focus = '距离上次反思已经过了一段时间,做一次定期回顾。'
- + '查看 COMPETENCE.md 和 LEARNING_STRATEGIES.md:'
- + '合并相近条目、删除已经不再适用或重复的内容、'
- + '把近期反复出现的稳定模式提炼成新技能。'
- + '如果近期没有新的可保存内容,直接说"无需保存",不要为了反思而反思。';
+ focus = 'Some time has passed since the last reflection — do a periodic review. '
+ + 'Look at COMPETENCE.md and LEARNING_STRATEGIES.md: '
+ + 'merge similar entries, drop content that no longer applies or is duplicated, '
+ + 'and distill recently-recurring stable patterns into new skills. '
+ + 'If there is nothing new worth saving, just say "nothing to save" — do not reflect for the sake of reflecting.';
break;
case 'complexity':
default:
- focus = '这是一个复杂的任务。'
- + '考虑是否有值得保存为技能的方法或模式。';
+ focus = 'This is a complex task. '
+ + 'Consider whether any approach or pattern is worth saving as a skill.';
break;
}
const parts = [
base,
- `**重点**: ${focus}`,
+ `**Focus**: ${focus}`,
];
if (conversationDigest) {
parts.push(
'',
- '**对话摘要**:',
+ '**Conversation digest**:',
conversationDigest,
);
}
parts.push(
'',
- '**当前自我评估 (COMPETENCE.md)**:',
- competence || '(尚无自我评估 — 请考虑创建一份。)',
+ '**Current self-assessment (COMPETENCE.md)**:',
+ competence || '(No self-assessment yet — consider creating one.)',
'',
- '**可用学习策略 (LEARNING_STRATEGIES.md)**:',
- strategies || '(尚无策略记录 — 请自行判断。)',
+ '**Available learning strategies (LEARNING_STRATEGIES.md)**:',
+ strategies || '(No strategy log yet — use your own judgment.)',
);
if (skillHealthReport) {
- parts.push('', '**技能健康度报告**:', skillHealthReport);
+ parts.push('', '**Skill health report**:', skillHealthReport);
}
parts.push(
'',
- '反思完成后,你可以:',
- '1. 通过 skill_manage 工具创建/修补/删除技能',
- '2. 通过 metacognition 工具更新 COMPETENCE.md',
- '3. 通过 metacognition 工具更新 LEARNING_STRATEGIES.md',
- '4. 如果没有值得保存的内容,直接说"无需保存"',
+ 'After reflecting, you can:',
+ '1. Create / patch / delete skills via the skill_manage tool',
+ '2. Update COMPETENCE.md via the metacognition tool',
+ '3. Update LEARNING_STRATEGIES.md via the metacognition tool',
+ '4. If nothing is worth saving, just say "nothing to save"',
);
return parts.join('\n');
diff --git a/src/core-agent/test/metacognition.test.ts b/src/core-agent/test/metacognition.test.ts
index 75fda034..97d524c3 100644
--- a/src/core-agent/test/metacognition.test.ts
+++ b/src/core-agent/test/metacognition.test.ts
@@ -292,86 +292,86 @@ describe('shouldReflect', () => {
describe('buildAdaptiveReviewPrompt', () => {
it('builds error_recovery focused prompt', () => {
const prompt = buildAdaptiveReviewPrompt('error_recovery', '', '');
- expect(prompt).toContain('从错误中恢复');
+ expect(prompt).toContain('recovered from an error');
expect(prompt).toContain('skill_manage');
expect(prompt).toContain('metacognition');
});
it('builds user_correction focused prompt', () => {
const prompt = buildAdaptiveReviewPrompt('user_correction', '', '');
- expect(prompt).toContain('纠正');
+ expect(prompt).toContain('corrected your approach');
expect(prompt).toContain('COMPETENCE');
});
it('builds skill_ineffective focused prompt', () => {
const prompt = buildAdaptiveReviewPrompt('skill_ineffective', '', '');
- expect(prompt).toContain('没有帮到');
- expect(prompt).toContain('修补或删除');
+ expect(prompt).toContain('did not help');
+ expect(prompt).toContain('Patch or delete');
});
it('builds known_weakness focused prompt', () => {
const prompt = buildAdaptiveReviewPrompt('known_weakness', '', '');
- expect(prompt).toContain('已知弱点');
- expect(prompt).toContain('加强');
+ expect(prompt).toContain('known weaknesses');
+ expect(prompt).toContain('Strengthen');
});
it('builds complexity default prompt', () => {
const prompt = buildAdaptiveReviewPrompt('complexity', '', '');
- expect(prompt).toContain('复杂');
+ expect(prompt).toContain('complex task');
});
it('includes competence content', () => {
- const prompt = buildAdaptiveReviewPrompt('complexity', '我擅长 Python', '');
- expect(prompt).toContain('我擅长 Python');
- expect(prompt).not.toContain('尚无自我评估');
+ const prompt = buildAdaptiveReviewPrompt('complexity', 'I am strong at Python', '');
+ expect(prompt).toContain('I am strong at Python');
+ expect(prompt).not.toContain('No self-assessment yet');
});
it('shows placeholder when no competence', () => {
const prompt = buildAdaptiveReviewPrompt('complexity', '', '');
- expect(prompt).toContain('尚无自我评估');
+ expect(prompt).toContain('No self-assessment yet');
});
it('includes strategies content', () => {
- const prompt = buildAdaptiveReviewPrompt('complexity', '', '错误提取法');
- expect(prompt).toContain('错误提取法');
- expect(prompt).not.toContain('尚无策略记录');
+ const prompt = buildAdaptiveReviewPrompt('complexity', '', 'Error extraction pattern');
+ expect(prompt).toContain('Error extraction pattern');
+ expect(prompt).not.toContain('No strategy log yet');
});
it('includes skill health report when provided', () => {
const report = '- docker-debug: effectiveness 0.83, healthy';
const prompt = buildAdaptiveReviewPrompt('complexity', '', '', report);
- expect(prompt).toContain('技能健康度报告');
+ expect(prompt).toContain('Skill health report');
expect(prompt).toContain('docker-debug');
});
it('omits skill health section when not provided', () => {
const prompt = buildAdaptiveReviewPrompt('complexity', '', '');
- expect(prompt).not.toContain('技能健康度报告');
+ expect(prompt).not.toContain('Skill health report');
});
it('includes conversation digest when provided', () => {
- const digest = '使用的工具: bash, read_file\n加载的技能: docker-debug\n--- 对话最终回复 ---\n修复了 Docker 容器无法启动的问题';
+ const digest = 'Tools used: bash, read_file\nSkills loaded: docker-debug\n--- final reply ---\nFixed Docker container that failed to start';
const prompt = buildAdaptiveReviewPrompt('error_recovery', '', '', undefined, digest);
expect(prompt).toContain('docker-debug');
- expect(prompt).toContain('Docker 容器');
- expect(prompt).toContain('对话最终回复');
+ expect(prompt).toContain('Docker container');
+ expect(prompt).toContain('final reply');
});
it('omits conversation digest content when not provided', () => {
const prompt = buildAdaptiveReviewPrompt('complexity', '', '');
- expect(prompt).not.toContain('对话最终回复');
+ expect(prompt).not.toContain('final reply');
});
it('builds weakness_succeeded focused prompt', () => {
const prompt = buildAdaptiveReviewPrompt('weakness_succeeded', '', '');
- expect(prompt).toContain('表现正常');
- expect(prompt).toContain('移除或降级');
- expect(prompt).toContain('不要修改或删除相关技能');
+ expect(prompt).toContain('performed normally');
+ expect(prompt).toContain('remove or downgrade');
+ expect(prompt).toContain('Do not modify or delete the related skills');
});
it('includes transient error guidance in all prompts', () => {
const prompt = buildAdaptiveReviewPrompt('complexity', '', '');
- expect(prompt).toContain('瞬态错误是环境问题');
- expect(prompt).toContain('不要在 COMPETENCE.md 标记为弱点');
+ expect(prompt).toContain('transient errors');
+ expect(prompt).toContain('Do not mark them as weaknesses in COMPETENCE.md');
});
});
diff --git a/src/main/features/reflection-digest.ts b/src/main/features/reflection-digest.ts
index fab876fe..8bd674ef 100644
--- a/src/main/features/reflection-digest.ts
+++ b/src/main/features/reflection-digest.ts
@@ -92,25 +92,25 @@ export function aggregateSession(messages: any[], into: SessionMetrics): void {
export function formatDigest(metrics: SessionMetrics, sinceMs: number, nowMs: number): string {
if (metrics.sessionsAnalyzed === 0) {
- return `近期 (${isoDate(sinceMs)} 至 ${isoDate(nowMs)}) 无新增对话活动。`;
+ return `No new conversation activity in the range ${isoDate(sinceMs)} to ${isoDate(nowMs)}.`;
}
const parts: string[] = [];
- const range = `${isoDate(metrics.earliestTs ?? sinceMs)} 至 ${isoDate(metrics.latestTs ?? nowMs)}`;
- parts.push(`基于 ${metrics.sessionsAnalyzed} 个会话 (${range}):`);
+ const range = `${isoDate(metrics.earliestTs ?? sinceMs)} to ${isoDate(metrics.latestTs ?? nowMs)}`;
+ parts.push(`Based on ${metrics.sessionsAnalyzed} session(s) (${range}):`);
const sortedTools = Object.entries(metrics.toolCalls).sort((a, b) => b[1] - a[1]).slice(0, TOOL_TOP_N);
if (sortedTools.length) {
parts.push('');
- parts.push('**工具调用**:');
- for (const [name, n] of sortedTools) parts.push(`- ${name}: ${n} 次`);
+ parts.push('**Tool calls**:');
+ for (const [name, n] of sortedTools) parts.push(`- ${name}: ${n}`);
}
if (metrics.errorCount > 0) {
parts.push('');
- parts.push(`**错误总数**: ${metrics.errorCount}`);
+ parts.push(`**Total errors**: ${metrics.errorCount}`);
if (metrics.errorSamples.length) {
- parts.push('**错误样本**:');
+ parts.push('**Error samples**:');
for (const s of metrics.errorSamples) parts.push(`- ${s}`);
}
}
@@ -118,7 +118,7 @@ export function formatDigest(metrics: SessionMetrics, sinceMs: number, nowMs: nu
const sortedSkills = Object.entries(metrics.skillsLoaded).sort((a, b) => b[1] - a[1]).slice(0, SKILLS_TOP_N);
if (sortedSkills.length) {
parts.push('');
- parts.push('**加载的技能**:');
+ parts.push('**Skills loaded**:');
for (const [id, n] of sortedSkills) parts.push(`- ${id}${n > 1 ? ` (${n}x)` : ''}`);
}
diff --git a/test/main/features/reflection-digest.test.ts b/test/main/features/reflection-digest.test.ts
index 290b7c9f..2b6f0d52 100644
--- a/test/main/features/reflection-digest.test.ts
+++ b/test/main/features/reflection-digest.test.ts
@@ -128,7 +128,7 @@ describe('reflection-digest › formatDigest', () => {
it('returns "no activity" when nothing was analyzed', async () => {
const mod = await loadModule();
const out = mod.formatDigest(mod.emptyMetrics(), Date.parse('2026-04-21'), Date.parse('2026-04-23'));
- expect(out).toMatch(/无新增对话活动/);
+ expect(out).toMatch(/No new conversation activity/);
});
it('formats a populated digest with all sections', async () => {
@@ -143,10 +143,10 @@ describe('reflection-digest › formatDigest', () => {
m.latestTs = Date.parse('2026-04-23T10:00:00Z');
const out = mod.formatDigest(m, m.earliestTs, Date.now());
- expect(out).toContain('3 个会话');
- expect(out).toContain('bash: 5 次');
- expect(out).toContain('web_search: 2 次');
- expect(out).toContain('错误总数**: 2');
+ expect(out).toContain('3 session');
+ expect(out).toContain('bash: 5');
+ expect(out).toContain('web_search: 2');
+ expect(out).toContain('Total errors**: 2');
expect(out).toContain('rate limit');
expect(out).toContain('recover-foo (3x)');
});
@@ -157,8 +157,8 @@ describe('reflection-digest › formatDigest', () => {
m.sessionsAnalyzed = 1;
m.toolCalls = { bash: 1 };
const out = mod.formatDigest(m, Date.now() - 1000, Date.now());
- expect(out).not.toContain('错误总数');
- expect(out).not.toContain('错误样本');
+ expect(out).not.toContain('Total errors');
+ expect(out).not.toContain('Error samples');
});
it('caps tool list at top-N most-called', async () => {
@@ -208,7 +208,7 @@ describe('reflection-digest › buildAgentReflectionDigest', () => {
it('returns "no activity" when no matching conversations exist', async () => {
const mod = await loadModule();
const out = await mod.buildAgentReflectionDigest(TEST_UID, '_default', Date.now() - 86400000);
- expect(out).toMatch(/无新增对话活动/);
+ expect(out).toMatch(/No new conversation activity/);
});
it('aggregates only conversations matching the target agent_id', async () => {
@@ -243,7 +243,7 @@ describe('reflection-digest › buildAgentReflectionDigest', () => {
const mod = await loadModule();
const since = Date.now() - 86400000; // last 24h
const digest = await mod.buildAgentReflectionDigest(TEST_UID, '_default', since);
- expect(digest).toMatch(/无新增对话活动/);
+ expect(digest).toMatch(/No new conversation activity/);
});
it('tolerates malformed jsonl lines and missing files', async () => {