A team of long-lived bots for DeepSeek Harness, with main bots, bot-to-bot messages and group chats.
Wait for the memory summary itself in the memory panel tests
Fei241135d44d7fd2f0
diff --git a/ds-bot/test/memory.test.js b/ds-bot/test/memory.test.jsindex 1c9bfef7..099b9ed4 100644--- a/ds-bot/test/memory.test.js+++ b/ds-bot/test/memory.test.js@@ -177,6 +177,21 @@ function start({ home = mkdtempSync(join(tmpdir(), 'ds-bot-memory-')), replies =const settle = () => new Promise(resolve => setTimeout(resolve, 20))+// The summary is written to disk after the model answers; a fixed wait is too short on+// a slow disk and leaves the write running into the next step or the cleanup.+async function summarized(call, botId = 'b2') {+ for (let tries = 0; tries < 200; tries += 1) {+ if ((await call('memory-view', { botId })).value.summarizing === false) return+ await settle()+ }+ throw new Error('the memory summary did not finish')+}++async function until(check) {+ for (let tries = 0; tries < 200 && !check(); tries += 1) await settle()+ assert.ok(check(), 'the condition did not hold within 4 s')+}+test('remember saves one entry with its source, and forget removes it, both in the journal', async (t) => {const { tool, read, journal, stop } = start()t.after(() => stop())@@ -336,7 +351,7 @@ test('the memory panel shows every entry and a summary the Bot\'s model writes oconst asked = (await call('memory-view', { botId: 'b2', summarize: true })).valueassert.deepEqual(asked.entries.map(entry => [entry.scope, entry.topic, entry.text]), [['own', null, 'Likes tea']])assert.equal(asked.summarizing, true)- await settle()+ await summarized(call)const done = (await call('memory-view', { botId: 'b2', summarize: true })).valueassert.deepEqual([done.summary.text, done.fresh, done.summarizing, done.error], ['## Overview\nYou like tea.', true, false, null])assert.equal(requests.length, 1)@@ -347,13 +362,13 @@ test('the memory panel shows every entry and a summary the Bot\'s model writes oawait tool('remember', { text: 'Lives in Hangzhou' }, 'b2')replies.push(new Error('provider down'))assert.equal((await call('memory-view', { botId: 'b2', summarize: true })).value.fresh, false)- await settle()+ await summarized(call)const failed = (await call('memory-view', { botId: 'b2', summarize: true })).valueassert.deepEqual([failed.summary.text, failed.fresh, failed.error], ['## Overview\nYou like tea.', false, 'provider down'])assert.equal(requests.length, 2)replies.push('## Overview\nYou like tea and live in Hangzhou.')await call('memory-view', { botId: 'b2', regenerate: true })- await settle()+ await summarized(call)assert.equal((await call('memory-view', { botId: 'b2' })).value.summary.text, '## Overview\nYou like tea and live in Hangzhou.')assert.equal((await call('memory-view', { botId: 'nope' })).error.message, 'Unknown Bot')})@@ -367,7 +382,7 @@ test('a summary cut off at the output limit is written once more with a bigger bt.after(() => stop())await tool('remember', { text: 'Likes tea' }, 'b2')await call('memory-view', { botId: 'b2', summarize: true })- await settle()+ await summarized(call)const done = (await call('memory-view', { botId: 'b2' })).valueassert.deepEqual([done.summary?.text, done.fresh, done.error], ['## Overview\nYou like tea.', true, null])assert.deepEqual(requests.map(request => request.maxTokens), [2048, 8192])@@ -383,11 +398,11 @@ test('a summary cut off at both budgets keeps the old one and says the reasoningt.after(() => stop())await tool('remember', { text: 'Likes tea' }, 'b2')await call('memory-view', { botId: 'b2', summarize: true })- await settle()+ await summarized(call)// A new entry asks for a fresh summary; both its tries are cut off.await tool('remember', { text: 'Lives in Hangzhou' }, 'b2')await call('memory-view', { botId: 'b2', summarize: true })- await settle()+ await summarized(call)const failed = (await call('memory-view', { botId: 'b2' })).valueassert.equal(failed.summary.text, '## Overview\nYou like tea.')assert.match(failed.error, /reasoning.*output budget|output budget.*reasoning/i)@@ -399,14 +414,14 @@ test('the panel call asks for the model\'s lowest thinking level when it has not.after(() => first.stop())await first.tool('remember', { text: 'Likes tea' }, 'b2')await first.call('memory-view', { botId: 'b2', summarize: true })- await settle()+ await summarized(first.call)assert.equal(first.requests[0].reasoningEffort, 'low')const second = start({ replies: ['## ok'], reasoning: {} })t.after(() => second.stop())await second.tool('remember', { text: 'Likes tea' }, 'b2')await second.call('memory-view', { botId: 'b2', summarize: true })- await settle()+ await summarized(second.call)assert.equal(second.requests[0].reasoningEffort, undefined)})@@ -418,7 +433,7 @@ test('a summary the Bot\'s model refuses is written by the profile default modelt.after(() => stop())await tool('remember', { text: 'Likes tea' }, 'b2')await call('memory-view', { botId: 'b2', summarize: true })- await settle()+ await summarized(call)const done = (await call('memory-view', { botId: 'b2' })).valueassert.deepEqual([done.summary?.text, done.fresh, done.error], ['## Overview\nYou like tea.', true, null])assert.deepEqual(routes(requests), ['deepseek/v4@2048', 'deepseek/flash@2048'])@@ -431,7 +446,7 @@ test('a summary cut off at both budgets on the Bot\'s model is written by the det.after(() => stop())await tool('remember', { text: 'Likes tea' }, 'b2')await call('memory-view', { botId: 'b2', summarize: true })- await settle()+ await summarized(call)const done = (await call('memory-view', { botId: 'b2' })).valueassert.deepEqual([done.summary?.text, done.error], ['## Overview\nYou like tea.', null])assert.deepEqual(routes(requests), ['deepseek/v4@2048', 'deepseek/v4@8192', 'deepseek/flash@2048'])@@ -443,7 +458,7 @@ test('a summary that failed on both models is tried again once the Bot has anotht.after(() => stop())await tool('remember', { text: 'Likes tea' }, 'b2')await call('memory-view', { botId: 'b2', summarize: true })- await settle()+ await summarized(call)const failed = (await call('memory-view', { botId: 'b2', summarize: true })).valueassert.equal(failed.error, 'deepseek/v4: 401: Unauthorized; deepseek/flash: 503 Service Unavailable')assert.equal(failed.summarizing, false)@@ -453,7 +468,7 @@ test('a summary that failed on both models is tried again once the Bot has anothreplies.push('## Overview\nYou like tea.')const retried = (await call('memory-view', { botId: 'b2', summarize: true })).valueassert.deepEqual([retried.summarizing, retried.error], [true, null])- await settle()+ await summarized(call)const done = (await call('memory-view', { botId: 'b2' })).valueassert.deepEqual([done.summary?.text, done.error], ['## Overview\nYou like tea.', null])assert.deepEqual(routes(requests).slice(2), ['deepseek/flash@2048'])@@ -822,7 +837,7 @@ test('a queued review keeps the group lines that leave the log while it waits',await call('state', {})// A slow review of Writer's own chat holds the queue.ctx.emit('agent/status', { agent: { id: 'b2' }, status: 'idle' })- await settle()+ await until(() => requests.length === 1)assert.equal(requests.length, 1)ctx.emit('agent/status', { agent: { id: 'g3' }, status: 'idle' })await postToGroup(ctx, Array.from({ length: 50 }, (_, at) => `More ${at + 1}`))@@ -1021,7 +1036,7 @@ test('a memory summary names the Bot chat on both of its tries', async (t) => {t.after(() => stop())await tool('remember', { text: 'Likes tea' }, 'b2')await call('memory-view', { botId: 'b2', summarize: true })- await settle()+ await summarized(call)const view = (await call('memory-view', { botId: 'b2' })).valueassert.equal(view.error, null)assert.equal(view.summary?.text, 'You like tea.')