@llm-dev-ops/agentics-cli 2.7.36 → 2.7.38

package/dist/commands/agents.js

@@ -17,7 +17,6 @@ import { execFileSync } from 'node:child_process';
 import { isTransientFailure, isTerminalFailure } from '../errors/transient.js';
 import { recordDegradation, drainDegradations } from '../observability/degradations.js';
 import { computePhase1Verdict } from '../pipeline/phase1-verdict.js';
-import { pipelineHeartbeat } from '../cli/ui/heartbeat.js';
 // ============================================================================
 // ADR-066: Copilot agents via claude --print (Claude Max — no API key needed)
 // ============================================================================
@@ -324,33 +323,12 @@ export const AGENT_DOMAINS = {
     },
 };
 /**
- * ADR-PIPELINE-087 §1 + 2.5.2 hotfix: domains that resolve through
- * AGENT_NAME_TO_DOMAIN but are NOT wired for direct cloud-function dispatch.
- *
- * Why: ADR-PIPELINE-080..086 added 14 AQE agents to the validation capability
- * as mandatory_agents. With 2.5.1 we registered AGENT_DOMAINS['quality-engineering']
- * and an adapter pointing at a guessed URL. Result: every ask call spawned 14
- * network attempts to a non-existent endpoint, each waiting out the per-call
- * timeout before the outer catch caught it. Stacked serially in
- * Promise.allSettled the delays were tolerable, but Claude Code's MCP client
- * tool-call budget is shorter than the pipeline wall time, so clients saw
- * "Tool result missing due to internal error."
- *
- * Resolution-only policy (2.5.2):
- *   - `quality-engineering` (AQE): no cloud handler deployed at a URL the CLI
- *     knows. Dispatch short-circuits with a clear 503 and zero network cost.
- *   - `diligence-artifacts`: SPARC/ADR/DDD/Context Map work is driven by the
- *     auto-chain phase coordinators (phase2/phase3/phase4), NOT by the graph-
- *     routed direct-dispatch path. Skip the adapter.
- *
- * Net effect: the ADR-087 goal is preserved (no silent drops — classifier
- * selections resolve and are observable in `routing.unresolved === []`) while
- * the pipeline finishes inside Claude Code's tool-call window again.
- *
- * Landing 2 removes entries here as real handlers come online.
+ * Domains that resolve through AGENT_NAME_TO_DOMAIN but are NOT wired for
+ * direct cloud-function dispatch. `diligence-artifacts` SPARC/ADR/DDD/Context
+ * Map work is driven by the auto-chain phase coordinators (phase2/phase3/
+ * phase4), not by the graph-routed direct-dispatch path.
  */
 export const RESOLUTION_ONLY_DOMAINS = new Set([
-    'quality-engineering',
     'diligence-artifacts',
 ]);
 export function isResolutionOnlyDomain(domain) {
@@ -402,12 +380,7 @@ export async function executeAgentsHealthCommand(domain, options) {
 }
 export async function executeAgentsInvokeCommand(domain, agent, payload, options) {
     const start = Date.now();
-    // ADR-PIPELINE-087 §1 + 2.5.2 hotfix: short-circuit resolution-only domains
-    // BEFORE the AGENT_DOMAINS check, the adapter creation, and any network I/O.
-    // These domains are resolved for routing visibility but have no direct
-    // dispatch handler — attempting a cloud call would stack per-invocation
-    // timeouts and blow Claude Code's MCP client budget (see the RESOLUTION_ONLY
-    // block comment above). Fast-fail with a structured 503 instead.
+    // Short-circuit resolution-only domains (no direct dispatch handler).
     if (RESOLUTION_ONLY_DOMAINS.has(domain)) {
         return {
             domain,
@@ -415,11 +388,7 @@ export async function executeAgentsInvokeCommand(domain, agent, payload, options
             status: 503,
             response: {
                 status: 'resolution-only',
-                reason: `${domain}/${agent} is registered in the router for classification visibility but is not wired for direct cloud-function dispatch. `
-                    + (domain === 'diligence-artifacts'
-                        ? 'SPARC/ADR/DDD artifacts are produced by the auto-chain phase coordinators (phase2/phase3/phase4).'
-                        : 'AQE agents will be wired to a real handler in a follow-up landing.'),
-                adr: 'ADR-PIPELINE-087',
+                reason: `${domain}/${agent} is registered in the router for classification visibility but is not wired for direct cloud-function dispatch. SPARC/ADR/DDD artifacts are produced by the auto-chain phase coordinators (phase2/phase3/phase4).`,
             },
             timing: Date.now() - start,
         };
@@ -3779,20 +3748,27 @@ export async function executeNaturalLanguageRoute(query, options) {
     // Top-of-function entry banner — fires within microseconds of dispatch so
     // the user sees the route is executing even if subsequent operations
     // (graph load, ruvector simulation, fleet dispatch) take a while.
-    // ADR-PIPELINE-100 §D5 — emit AGENTICS_TRACE_ID=<id> as the very first
-    // line so the MCP server's fast-return path can grab it within the first
-    // 100ms and return a "pipeline started, tail this log" response to the
-    // caller without blocking on the rest of the run.
-    process.stderr.write(`AGENTICS_TRACE_ID=${correlationId}\n`);
     process.stderr.write(`[agentics] route entered — query="${query.slice(0, 100)}${query.length > 100 ? '…' : ''}" depth=${options.depth ?? 'lite'} trace=${correlationId.slice(0, 8)}\n`);
-    // ADR-PIPELINE-099 D3 — process-wide heartbeat. Survives across the
-    // executeNaturalLanguageRoute → executeAutoChain handoff (the previous
-    // per-call heartbeat died at the function boundary, leaving Phases 2–7
-    // silent for minutes at a time). Idempotent — auto-chain calls start()
-    // again with no effect. Stop happens at the end of executeAutoChain or
-    // via the singleton's exit handlers.
-    pipelineHeartbeat.start('agentics ask');
-    pipelineHeartbeat.setPhase('Phase 1 — Fleet Dispatch + Simulation');
+    // Global 10-second heartbeat. Ticks until process exit (`unref()` keeps it
+    // from blocking termination). When any other code emits its own output the
+    // tick still fires but that's acceptable — better noisy than silent. The
+    // heartbeat suppresses itself via `lastOutputAt` if other writes have
+    // happened recently.
+    let lastOutputAt = Date.now();
+    const realStderrWrite = process.stderr.write.bind(process.stderr);
+    // eslint-disable-next-line @typescript-eslint/no-explicit-any
+    process.stderr.write = (chunk, ...rest) => {
+        lastOutputAt = Date.now();
+        return realStderrWrite(chunk, ...rest);
+    };
+    const heartbeat = setInterval(() => {
+        if (Date.now() - lastOutputAt < 9_000)
+            return; // something else just wrote
+        const elapsedSec = Math.floor((Date.now() - start) / 1000);
+        realStderrWrite(`[agentics] ⏳ still working — ${elapsedSec}s elapsed\n`);
+        lastOutputAt = Date.now();
+    }, 10_000);
+    heartbeat.unref();
     // ========================================================================
     // ADR-030: Capability Graph Routing
     //
@@ -3835,7 +3811,6 @@ export async function executeNaturalLanguageRoute(query, options) {
                 console.error(graphResult.routing.explanation);
                 // Run ruvector simulation first (same as full-fleet path, ADR-016)
                 console.error('  [RUVECTOR] Running simulation before graph-routed dispatch (ADR-016)...');
-                pipelineHeartbeat.setActivity(`ruvector: simulating ${graphResult.routing.totalAgents}-agent graph route`);
                 const simResult = await (async () => {
                     const MAX_RETRIES = 3;
                     let lastErr;
@@ -3874,21 +3849,8 @@ export async function executeNaturalLanguageRoute(query, options) {
                 const REPO_LOCAL_DOMAINS = new Set(['copilot']);
                 const prevLocalAgents = process.env['AGENTICS_LOCAL_AGENTS'];
                 console.error(`  Dispatching ${graphResult.agents.length} graph-routed agents...`);
-                pipelineHeartbeat.setActivity(`fleet: dispatching ${graphResult.agents.length} graph-routed agents`);
                 const agentPromises = [];
-                // ADR-PIPELINE-100 D2 — emit a `[FLEET-CALL]` line as each agent
-                // STARTS and a result line when each FINISHES. The fleet runs in
-                // parallel via Promise.allSettled, so without these lines the user
-                // sees nothing for 1-3 minutes between the dispatch banner and the
-                // aggregate result. Pre-2.7.27 builds had this same silence — it
-                // never went away by itself when the heartbeat singleton was added.
-                const fleetTotal = graphResult.agents.length;
-                let fleetStartedCount = 0;
-                let fleetFinishedCount = 0;
-                graphResult.agents.forEach((ref, idx) => {
-                    const agentStart = Date.now();
-                    fleetStartedCount++;
-                    process.stderr.write(`  [FLEET-CALL] [${String(idx + 1).padStart(3, ' ')}/${fleetTotal}] → ${ref.domain}/${ref.agent}\n`);
+                for (const ref of graphResult.agents) {
                     agentPromises.push((async () => {
                         try {
                             if (!REPO_LOCAL_DOMAINS.has(ref.domain)) {
@@ -3902,25 +3864,12 @@ export async function executeNaturalLanguageRoute(query, options) {
                             }
                             const payload = buildDomainPayload(ref.domain, ref.agent, query, correlationId);
                             const result = await executeAgentsInvokeCommand(ref.domain, ref.agent, payload, options);
-                            const elapsed = ((Date.now() - agentStart) / 1000).toFixed(1);
-                            fleetFinishedCount++;
-                            const status = result.status ?? 200;
-                            const ok = status >= 200 && status < 300;
-                            process.stderr.write(`  [FLEET-CALL] [${String(idx + 1).padStart(3, ' ')}/${fleetTotal}] ${ok ? '✓' : '✗'} ${ref.domain}/${ref.agent} (${elapsed}s, ${fleetFinishedCount}/${fleetTotal} done)\n`);
-                            pipelineHeartbeat.setActivity(`fleet: ${fleetFinishedCount}/${fleetTotal} agents complete`);
-                            return { kind: 'agent', domain: ref.domain, agent: ref.agent, status, response: result.response };
+                            return { kind: 'agent', domain: ref.domain, agent: ref.agent, status: result.status ?? 200, response: result.response };
                         }
                         catch (err) {
-                            fleetFinishedCount++;
-                            const elapsed = ((Date.now() - agentStart) / 1000).toFixed(1);
-                            const errMsg = err instanceof Error ? err.message.slice(0, 80) : String(err).slice(0, 80);
-                            process.stderr.write(`  [FLEET-CALL] [${String(idx + 1).padStart(3, ' ')}/${fleetTotal}] ✗ ${ref.domain}/${ref.agent} (${elapsed}s) — ${errMsg}\n`);
                             return { kind: 'agent', domain: ref.domain, agent: ref.agent, status: 502, response: { error: err instanceof Error ? err.message : String(err) } };
                         }
                     })());
-                });
-                if (fleetStartedCount > 0) {
-                    process.stderr.write(`  [FLEET-CALL] all ${fleetStartedCount} agents in flight — awaiting responses...\n`);
                 }
                 const allResults = await Promise.allSettled([Promise.resolve(simResult), ...agentPromises]);
                 const elapsed = Date.now() - start;
@@ -4210,7 +4159,6 @@ export async function executeNaturalLanguageRoute(query, options) {
             { domain: 'platform', agent: 'risk-score' },
         ];
         console.error(`Dispatching ${fleetAgents.length} agents across ALL 27 domains + ruvector simulation`);
-        pipelineHeartbeat.setActivity(`fleet: dispatching ${fleetAgents.length} agents across 27 domains`);
         // ========================================================================
         // ADR-016: Run ruvector simulation BEFORE fleet dispatch (sequential).
         //
@@ -4223,7 +4171,6 @@ export async function executeNaturalLanguageRoute(query, options) {
         // start the fleet. Adds ~5s latency but guarantees simulation success.
         // ========================================================================
         console.error('  [RUVECTOR] Running simulation before fleet dispatch (ADR-016)...');
-        pipelineHeartbeat.setActivity('ruvector: pre-fleet simulation');
         const simResult = await (async () => {
             const MAX_RETRIES = 3;
             let lastErr;
@@ -4270,24 +4217,12 @@ export async function executeNaturalLanguageRoute(query, options) {
         const WAVE_SIZE = 15;
         const WAVE_DELAY_MS = 2000;
         const agentPromises = [];
-        // ADR-PIPELINE-100 D2 — per-agent visibility for the full-fleet path.
-        // Without these per-call lines the user sees a "Wave N/M" banner then
-        // 1-3 minutes of silence per wave, then nothing until ALL waves
-        // finish. The user explicitly demanded "I should see them being
-        // called" — these are the lines that satisfy that.
-        const fleetTotal = fleetAgents.length;
-        let fleetFinishedCount = 0;
-        let agentDispatchIdx = 0;
         for (let i = 0; i < fleetAgents.length; i += WAVE_SIZE) {
             const wave = fleetAgents.slice(i, i + WAVE_SIZE);
             const waveNum = Math.floor(i / WAVE_SIZE) + 1;
             const totalWaves = Math.ceil(fleetAgents.length / WAVE_SIZE);
-            console.error(`  Wave ${waveNum}/${totalWaves}: dispatching ${wave.length} agents (${wave.map(a => `${a.domain}/${a.agent}`).slice(0, 3).join(', ')}${wave.length > 3 ? '...' : ''})`);
+            console.error(`  Wave ${waveNum}/${totalWaves}: ${wave.length} agents (${wave.map(a => `${a.domain}/${a.agent}`).slice(0, 3).join(', ')}${wave.length > 3 ? '...' : ''})`);
             for (const { domain, agent } of wave) {
-                agentDispatchIdx++;
-                const myIdx = agentDispatchIdx;
-                const agentStart = Date.now();
-                process.stderr.write(`  [FLEET-CALL] [${String(myIdx).padStart(3, ' ')}/${fleetTotal}] → ${domain}/${agent}\n`);
                 agentPromises.push((async () => {
                     try {
                         // Toggle repo-local per-agent: only copilot uses repo-local
@@ -4302,19 +4237,9 @@ export async function executeNaturalLanguageRoute(query, options) {
                         }
                         const payload = buildDomainPayload(domain, agent, query, correlationId);
                         const result = await executeAgentsInvokeCommand(domain, agent, payload, options);
-                        fleetFinishedCount++;
-                        const elapsed = ((Date.now() - agentStart) / 1000).toFixed(1);
-                        const status = result.status ?? 200;
-                        const ok = status >= 200 && status < 300;
-                        process.stderr.write(`  [FLEET-CALL] [${String(myIdx).padStart(3, ' ')}/${fleetTotal}] ${ok ? '✓' : '✗'} ${domain}/${agent} (${elapsed}s, ${fleetFinishedCount}/${fleetTotal} done)\n`);
-                        pipelineHeartbeat.setActivity(`fleet: ${fleetFinishedCount}/${fleetTotal} agents complete`);
-                        return { kind: 'agent', domain, agent, status, response: result.response };
+                        return { kind: 'agent', domain, agent, status: result.status ?? 200, response: result.response };
                     }
                     catch (err) {
-                        fleetFinishedCount++;
-                        const elapsed = ((Date.now() - agentStart) / 1000).toFixed(1);
-                        const errMsg = err instanceof Error ? err.message.slice(0, 80) : String(err).slice(0, 80);
-                        process.stderr.write(`  [FLEET-CALL] [${String(myIdx).padStart(3, ' ')}/${fleetTotal}] ✗ ${domain}/${agent} (${elapsed}s) — ${errMsg}\n`);
                         return { kind: 'agent', domain, agent, status: 502, response: { error: err instanceof Error ? err.message : String(err) } };
                     }
                 })());
@@ -4324,7 +4249,6 @@ export async function executeNaturalLanguageRoute(query, options) {
                 await new Promise(resolve => setTimeout(resolve, WAVE_DELAY_MS));
             }
         }
-        process.stderr.write(`  [FLEET-CALL] all ${fleetTotal} agents in flight — awaiting final responses...\n`);
         const allResults = await Promise.allSettled([Promise.resolve(simResult), ...agentPromises]);
         const elapsed = Date.now() - start;
         // Restore repo-local mode setting