an open-source alternative to the dots, bots and muses of the world, run an office of claude code/codex like agents on your laptop, sandboxes or anywhere, uses your existing subscriptions
HTML agent-harnessagent-orchestrationagentsai-agentsautonomous-agentsclaude-codecodexcoding-agentsdesktop-appdeveloper-toolselectrongemini-cliharnessharness-engineeringlocal-firstmemorymulti-agentopencodeorchestrationtypescript
test(bench): make the benchmark workload env-configurable
savvaska6538e10c16a
diff --git a/test/usage-cache-benchmark.manual.cjs b/test/usage-cache-benchmark.manual.cjsindex e21496dc..d35ac1af 100644--- a/test/usage-cache-benchmark.manual.cjs+++ b/test/usage-cache-benchmark.manual.cjs@@ -7,9 +7,9 @@* gets a new record appended, then every agent's own ~30s usage beat fires* (readAgentUsage(cwd, {sessionId})).*- * Reports wall-clock time AND total JSON.parse() calls (a direct proxy for- * "how many times was a line reprocessed") for both the pre-fix and- * post-fix transcript.ts, run back-to-back against the identical workload.+ * Prints per-round progress live (for recorded before/after demos) plus a+ * final summary: wall-clock time AND total JSON.parse() calls (a direct+ * proxy for "how many times was a line reprocessed").** Usage: node usage-cache-benchmark.cjs <path-to-transcript.ts> <label>*/@@ -24,8 +24,22 @@ if (!transcriptTsPath || !label) {process.exit(1);}-const N_AGENTS = 10;-const N_ROUNDS = 30;+// Workload size. An unparseable value is an error, not a silent fallback: a+// benchmark that quietly ignores the size you asked for reports a number for+// the wrong workload.+function benchCount(name, fallback) {+ const raw = process.env[name];+ if (raw === undefined || raw === '') return fallback;+ const n = Number(raw);+ if (!Number.isInteger(n) || n <= 0) {+ console.error(`${name} must be a positive integer, got: ${raw}`);+ process.exit(1);+ }+ return n;+}++const N_AGENTS = benchCount('BENCH_AGENTS', 10);+const N_ROUNDS = benchCount('BENCH_ROUNDS', 30);// Sandbox HOME so projectDir() never touches the real ~/.claude/projects.const FAKE_HOME = fs.mkdtempSync(path.join(os.tmpdir(), 'bench-home-'));@@ -66,15 +80,18 @@ let parseCount = 0;const realParse = JSON.parse;JSON.parse = (...args) => { parseCount++; return realParse(...args); };+console.log(`=== ${label} === ${N_AGENTS} agents sharing a dummy project dir, ${N_ROUNDS} rounds of "everyone active"\n`);const t0 = process.hrtime.bigint();-for (let round = 0; round < N_ROUNDS; round++) {+for (let round = 1; round <= N_ROUNDS; round++) {// Everyone is active: each agent's own file grows by one record.for (const a of agents) fs.appendFileSync(a.file, record(a.sessionId, round));// Everyone's ~30s usage beat fires, filtered to their own session.for (const a of agents) readAgentUsage(cwd, { sessionId: a.sessionId });+ const elapsedMs = Number(process.hrtime.bigint() - t0) / 1e6;+ console.log(` round ${String(round).padStart(2)}/${N_ROUNDS} elapsed ${elapsedMs.toFixed(1).padStart(7)}ms JSON.parse() calls so far: ${parseCount}`);}const t1 = process.hrtime.bigint();JSON.parse = realParse;const ms = Number(t1 - t0) / 1e6;-console.log(`${label}: ${N_AGENTS} agents x ${N_ROUNDS} rounds -> ${ms.toFixed(1)}ms wall-clock, ${parseCount} JSON.parse() calls`);+console.log(`\n${label} TOTAL: ${ms.toFixed(1)}ms wall-clock, ${parseCount} JSON.parse() calls\n`);