@nanobpm/nano-workforce 0.188.1 → 0.189.0

package/operations/enrolAgenticWorker.ts

@@ -13,6 +13,12 @@
 import type { Capability } from "@nanobpm/agentic/protocol";
 import { resolveEnrolment } from "../app/agentic/vocab/enrol.ts";
 import { DurableResumeRegistry } from "../app/durableResume.ts";
+import {
+  HarnessProtocolRegistry,
+  isStaleProtocol,
+  minHarnessProtocol,
+  staleHarnessPolicy,
+} from "../app/harnessProtocol.ts";
 import { envVar } from "../app/version.ts";
 import type { EnrolResult } from "../nano-generated/api-io.d.ts";
 import { defineOperation } from "../nano-generated/operations.ts";
@@ -73,6 +79,22 @@ export default defineOperation("enrolAgenticWorker", async ({ req, body }, app)
     app.log.warn("enrolAgenticWorker rejected: non-boolean durableResume");
     return { status: 400, body: { error: "`durableResume` must be a boolean when provided" } };
   }
+  // The harness-protocol enrolment attribute (issue #802) — a non-negative integer the harness
+  // advertises declaring which machine-readable artifacts it emits (AgentInstance, transcript flush,
+  // result envelope). A directly-invoked delegate bypasses the OpenAPI runtime validation, so guard the
+  // type here: a non-integer / negative value would corrupt the persisted staleness gate.
+  if (
+    body.harnessProtocol !== undefined &&
+    (typeof body.harnessProtocol !== "number" ||
+      !Number.isInteger(body.harnessProtocol) ||
+      body.harnessProtocol < 0)
+  ) {
+    app.log.warn("enrolAgenticWorker rejected: non-integer harnessProtocol");
+    return {
+      status: 400,
+      body: { error: "`harnessProtocol` must be a non-negative integer when provided" },
+    };
+  }
 
   // Fold a top-level `host` into the capability when the capability didn't carry its own — a worker
   // may declare its host either on the capability or beside it (ADR 0059 `{ capability, host }`).
@@ -102,19 +124,57 @@ export default defineOperation("enrolAgenticWorker", async ({ req, body }, app)
     }
   }
 
+  // Harness-protocol enrolment gate (issue #802): record the advertised protocol per instance so the
+  // app can expose it in getAgenticSupply / the registry and gate on it. Recorded even on a downgrade
+  // (a re-enrol WITHOUT a version clears a stale-healthy value to absent → stale). Like durable-resume
+  // it needs a non-blank `instance` and is best-effort — a registry write hiccup must not fail enrol.
+  if (app.data && instanceKey) {
+    try {
+      await new HarnessProtocolRegistry(app.data).recordEnrolment(instanceKey, body.harnessProtocol);
+    } catch (err) {
+      app.log.warn("enrolAgenticWorker: harness-protocol record failed", { instance: instanceKey, err: String(err) });
+    }
+  }
+
+  // Derive staleness from the just-advertised protocol against the configured minimum (absent version =
+  // stale, the #802 signature). Under the `refuse` policy a stale harness is handed an EMPTY SERVE set
+  // so it wins no job leases — a REGISTER→SERVE gate, never an engine/job-protocol change. Under the
+  // default `flag` policy the SERVE set is unchanged (no routing regression); the worker is only
+  // flagged so the cockpit can surface it for drain.
+  const harnessStale = isStaleProtocol(body.harnessProtocol, minHarnessProtocol());
+  const refuseRouting = harnessStale && staleHarnessPolicy() === "refuse";
+  if (refuseRouting) {
+    app.log.warn("enrolAgenticWorker: refusing routing for a stale harness", {
+      instance: body.instance,
+      harnessProtocol: body.harnessProtocol,
+      minHarnessProtocol: minHarnessProtocol(),
+    });
+  }
+
   const result: EnrolResult = {
-    serve: [...resolved.serve],
-    roles: resolved.roles.map((role) => {
-      const out: EnrolResult["roles"][number] = { token: role.token, seatsDistinctFamily: role.seatsDistinctFamily };
-      if (role.weight !== undefined) out.weight = role.weight;
-      return out;
-    }),
+    serve: refuseRouting ? [] : [...resolved.serve],
+    roles: refuseRouting
+      ? []
+      : resolved.roles.map((role) => {
+          const out: EnrolResult["roles"][number] = { token: role.token, seatsDistinctFamily: role.seatsDistinctFamily };
+          if (role.weight !== undefined) out.weight = role.weight;
+          return out;
+        }),
     demandVersion: resolved.demandVersion,
     leaseTtl: resolved.leaseTtl,
+    // Always surface the staleness verdict so the caller (and the cockpit) can see a stale harness even
+    // when it advertised no version at all.
+    harnessStale,
   };
   if (body.instance !== undefined) result.instance = body.instance;
   if (body.durableResume !== undefined) result.durableResume = body.durableResume;
+  if (body.harnessProtocol !== undefined) result.harnessProtocol = body.harnessProtocol;
 
-  app.log.info("agentic enrol resolved", { instance: body.instance, serve: result.serve, family: capability.family });
+  app.log.info("agentic enrol resolved", {
+    instance: body.instance,
+    serve: result.serve,
+    family: capability.family,
+    harnessStale,
+  });
   return { status: 200, body: result };
 });

package/operations/getAgenticRegistry.ts

@@ -20,6 +20,6 @@ export default defineOperation("getAgenticRegistry", async ({ req }, app) => {
     app.log.warn("getAgenticRegistry rejected: missing/invalid shared secret");
     return { status: 401, body: { error: "unauthorized" } };
   }
-  const report = await computeRegistryReport(app.log);
+  const report = await computeRegistryReport(app.log, app.data);
   return { status: 200, body: toWireReport(report) };
 });

package/operations/getAgenticSupply.test.ts

@@ -5,6 +5,7 @@
 // by instance, family/host/jobKeys/liveness) — driven through a REAL AgenticHub + in-memory transport
 // exactly as the presence family is exercised, so the singleton the operation reads is the live one.
 import { DatabaseSync } from "node:sqlite";
+import { readFileSync } from "node:fs";
 import { test } from "node:test";
 import { AgenticHub } from "@nanobpm/agentic/channel";
 import type { Authenticator, ChannelConnection, ChannelTransport } from "@nanobpm/agentic/channel";
@@ -39,6 +40,19 @@ function memData(db: SqliteDb): DataLayer {
   return { source: () => ({ db }) } as unknown as DataLayer;
 }
 
+/** Wrap an existing DatabaseSync as a SqliteDb (so the presence store and a raw insert share one db). */
+function memSqliteOver(db: DatabaseSync): SqliteDb {
+  return {
+    exec: (sql) => db.exec(sql),
+    run: (sql, params = []) => {
+      const r = db.prepare(sql).run(...(params as never[]));
+      return { changes: Number(r.changes), lastInsertRowid: Number(r.lastInsertRowid) };
+    },
+    all: <T = Record<string, unknown>>(sql: string, params: unknown[] = []) =>
+      db.prepare(sql).all(...(params as never[])) as T[],
+  };
+}
+
 function memTransport(): { transport: ChannelTransport; connect(conn: ChannelConnection): void } {
   let onConnection: ((conn: ChannelConnection) => void) | undefined;
   const transport: ChannelTransport = {
@@ -259,3 +273,69 @@ test("#738 drift: the supply advertises the producer's instance-scoped stream id
     await hub.close();
   }
 });
+
+// Harness staleness (issue #802): the supply report flags a worker whose harness protocol is below the
+// minimum / absent, joined by instance from the app's harness-protocol registry.
+test("#802: flags harnessStale per worker, joining the harness-protocol registry by instance", async () => {
+  const raw = new DatabaseSync(":memory:");
+  raw.exec("PRAGMA foreign_keys = ON;");
+  raw.exec(readFileSync(new URL("../db/migrations/107_worker_harness_protocol.sql", import.meta.url), "utf8"));
+  const now = new Date().toISOString();
+  // wk-a advertised a healthy protocol (>= min 1); wk-a's presence row is minted below.
+  raw.prepare("INSERT INTO worker_harness_protocol (instance, harness_protocol, updated_at) VALUES (?, ?, ?)").run("wk-a", 2, now);
+
+  // A data layer providing BOTH the presence `source().db` handle and the RAD `table()` surface over
+  // the SAME db, so the presence family and the harness registry read one store.
+  const sqlite = memSqliteOver(raw);
+  const quote = (id: string) => `"${id.replace(/"/g, '""')}"`;
+  const table = (name: string, pk = "id") => ({
+    // biome-ignore lint/suspicious/noExplicitAny: test-only gateway.
+    async find(where: any = {}): Promise<any[]> {
+      const keys = Object.keys(where);
+      const clause = keys.length ? `WHERE ${keys.map((k) => `${quote(k)} = ?`).join(" AND ")}` : "";
+      return raw.prepare(`SELECT * FROM ${quote(name)} ${clause}`).all(...keys.map((k) => where[k])) as any[];
+    },
+    // biome-ignore lint/suspicious/noExplicitAny: test-only gateway.
+    async findOne(where: any = {}): Promise<any> {
+      return (await this.find(where))[0];
+    },
+    _pk: pk,
+  });
+  const data = {
+    source: () => ({ db: sqlite }),
+    table,
+    // The raw-SQL surface the harness registry's bounded `WHERE instance IN (…)` read binds to, over
+    // the SAME db as `table`/presence.
+    // biome-ignore lint/suspicious/noExplicitAny: test-only gateway.
+    open: () => ({ query: async (sql: string, params: any[] = []) => raw.prepare(sql).all(...params) as any[] }),
+  } as unknown as DataLayer;
+
+  const transport = memTransport();
+  const hub = new AgenticHub({ transport: transport.transport, authenticator, sweepIntervalMs: 0 });
+  await family.mount({ hub, registry: hub.registry, transport: transport.transport as never, data, log: noopLog() });
+  const healthyConn = fakeConn("c1", "leafA");
+  transport.connect(healthyConn.conn);
+  await flush();
+  healthyConn.feed({ lane: "control", family: "register", seq: 1, payload: { instance: "wk-a", capability: {} } });
+  const staleConn = fakeConn("c2", "leafA");
+  transport.connect(staleConn.conn);
+  await flush();
+  // wk-b registers but has NO harness-protocol row → absent = stale.
+  staleConn.feed({ lane: "control", family: "register", seq: 1, payload: { instance: "wk-b", capability: {} } });
+  await flush();
+
+  const withData = { log: noopLog(), data } as unknown as AppApi;
+  try {
+    const res = (await handler(input(), withData)) as { status: number; body: { workers: Array<Record<string, unknown>> } };
+    assertEquals(res.status, 200);
+    const byInstance = new Map(res.body.workers.map((w) => [w.instance, w]));
+    assertEquals(byInstance.get("wk-a")?.harnessStale, false, "healthy protocol is not stale");
+    assertEquals(byInstance.get("wk-a")?.harnessProtocol, 2);
+    assertEquals(byInstance.get("wk-b")?.harnessStale, true, "no recorded protocol = stale");
+    assertEquals("harnessProtocol" in (byInstance.get("wk-b") ?? {}), false, "no protocol echoed for a stale worker");
+  } finally {
+    family.teardown?.();
+    await hub.close();
+    raw.close();
+  }
+});

package/operations/getAgenticSupply.ts

@@ -23,6 +23,7 @@
 import { type ClaimRegistry, currentClaimRegistry } from "../app/agentic/claim-registry.ts";
 import { currentCorrelation, type JobCorrelation } from "../app/agentic/correlation.ts";
 import { currentPresenceRegistry, type SupplyWorker } from "../app/agentic/families/presence.family.ts";
+import { assessWorkers, type HarnessAssessment } from "../app/harnessProtocol.ts";
 import { envVar } from "../app/version.ts";
 import type { AgenticJobCorrelation, AgenticSupplyReport, AgenticSupplyWorker } from "../nano-generated/api-io.d.ts";
 import { defineOperation } from "../nano-generated/operations.ts";
@@ -36,7 +37,7 @@ const SECRET = envVar("NANO_PR_WEBHOOK_SECRET") ?? "";
 // (`composeStreamId(instance, jobKey)`, issue #738) when the claim registry knows a current claim for
 // it (#713) — keyed by the CLAIM, not by the connection — so drilling in opens the LIVE job's terminal
 // (the exact stream the producer writes) even before any transcript lands.
-function toWorker(w: SupplyWorker, claims: ClaimRegistry | undefined): AgenticSupplyWorker {
+function toWorker(w: SupplyWorker, claims: ClaimRegistry | undefined, harness: HarnessAssessment | undefined): AgenticSupplyWorker {
   const out: AgenticSupplyWorker = {
     instance: w.instance,
     identity: w.identity,
@@ -44,9 +45,13 @@ function toWorker(w: SupplyWorker, claims: ClaimRegistry | undefined): AgenticSu
     jobKeys: [...w.jobKeys],
     live: w.live,
     staleMs: w.staleMs,
+    // Harness staleness (issue #802) — an absent registry entry / unmounted data layer reads as stale
+    // (fail loud). Distinct from the liveness `staleMs` heartbeat grade above.
+    harnessStale: harness?.stale ?? true,
   };
   if (w.family !== undefined) out.family = w.family;
   if (w.host !== undefined) out.host = w.host;
+  if (harness?.harnessProtocol !== undefined) out.harnessProtocol = harness.harnessProtocol;
   return out;
 }
 
@@ -82,11 +87,18 @@ export default defineOperation("getAgenticSupply", async ({ req }, app) => {
   // process-instance / plan context surfaced in `correlations`, but no longer feeds visibility.
   const correlation = currentCorrelation();
   const snapshot = registry.snapshot(claims ? { jobKeysFor: (instance) => claims.jobKeysFor(instance) } : {});
+  // Harness-staleness (issue #802): assess every visible worker's advertised protocol against the
+  // configured minimum — the ONE canonical staleness derivation (no second heuristic). Best-effort:
+  // an unmounted data layer / read failure reads as stale (fail loud).
+  const harness = await assessWorkers(app.data, snapshot.workers.map((w) => w.instance));
   const report: AgenticSupplyReport = {
     count: snapshot.count,
     generatedAt: new Date().toISOString(),
-    workers: snapshot.workers.map((w) => toWorker(w, claims)),
-    leaves: snapshot.leaves.map((leaf) => ({ token: leaf.token, workers: leaf.workers.map((w) => toWorker(w, claims)) })),
+    workers: snapshot.workers.map((w) => toWorker(w, claims, harness.get(w.instance))),
+    leaves: snapshot.leaves.map((leaf) => ({
+      token: leaf.token,
+      workers: leaf.workers.map((w) => toWorker(w, claims, harness.get(w.instance))),
+    })),
     correlations: correlation ? correlation.snapshot().correlations.map(toCorrelation) : [],
   };
   return { status: 200, body: report };

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@nanobpm/nano-workforce",
-  "version": "0.188.1",
+  "version": "0.189.0",
   "description": "Nano Workforce — an Agent Graph Orchestration application for Agentic SDLC: durable BPMN processes that coordinate a graph of AI agents across the software delivery lifecycle.",
   "type": "module",
   "main": "main.ts",

package/pages/cockpit/mount.js

@@ -84,6 +84,10 @@ function workerView(worker, staleAfterMs, byJobKey) {
     correlations,
     liveness: liveness(worker, staleAfterMs),
     staleMs: worker.staleMs,
+    // Fail loud: default a missing harness verdict to STALE (mirrors the typed cockpit view and the
+    // server's fail-loud assessment) so an older/cached response can't hide a stale/unknown worker.
+    harnessStale: worker.harnessStale ?? true,
+    ...(worker.harnessProtocol !== undefined ? { harnessProtocol: worker.harnessProtocol } : {}),
   };
 }
 
@@ -141,6 +145,7 @@ function workerRow(doc, worker, onDrill, onOpenWorker) {
   row.setAttribute("data-worker", worker.instance);
   row.setAttribute("data-liveness", worker.liveness);
   row.setAttribute("data-stream", worker.stream);
+  row.setAttribute("data-harness-stale", String(worker.harnessStale));
 
   const nameCell = el(doc, "td", "cockpit-td cockpit-supply-name");
   nameCell.appendChild(dot(doc, worker.liveness));
@@ -162,6 +167,19 @@ function workerRow(doc, worker, onDrill, onOpenWorker) {
     if (onDrill) drill.addEventListener("click", () => onDrill(worker.stream));
     nameCell.appendChild(drill);
   }
+  // A stale harness silently swallows machine-readable artifacts (issue #802) — surface it as a
+  // distinct badge (mirrors app/agentic/cockpit/supply-render.ts).
+  if (worker.harnessStale) {
+    const badge = el(
+      doc,
+      "span",
+      "cockpit-supply-harness-stale",
+      worker.harnessProtocol === undefined ? "stale harness" : `stale harness (v${worker.harnessProtocol})`,
+    );
+    badge.setAttribute("data-harness-stale", "true");
+    badge.setAttribute("title", "Harness protocol below the configured minimum (or none advertised); jobs may dead-end.");
+    nameCell.appendChild(badge);
+  }
   row.appendChild(nameCell);
 
   row.appendChild(el(doc, "td", "cockpit-td cockpit-supply-family", worker.family));

package/test/worldDb.ts

@@ -69,12 +69,18 @@ function gateway(db: DatabaseSync, name: string, pk: string) {
  * that decorates `table` on the returned source sees its decoration inside the transaction too. */
 type MemDataSource = {
   table: (name: string, pk?: string) => ReturnType<typeof gateway>;
+  query<T = unknown>(sql: string, params?: unknown[]): Promise<T[]>;
   tx<T>(fn: (t: MemDataSource) => Promise<T>): Promise<T>;
 };
 
 function openDataSource(db: DatabaseSync): MemDataSource {
   const ds: MemDataSource = {
     table: (name, pk = "id") => gateway(db, name, pk),
+    // The row-returning raw-SQL surface (mirrors the runtime `DataSource.query`) so tests exercise a
+    // handler's bounded `WHERE … IN (…)` reads against genuine SQLite rather than a mock.
+    async query<T = unknown>(sql: string, params: unknown[] = []): Promise<T[]> {
+      return db.prepare(sql).all(...params.map(coerce)) as T[];
+    },
     async tx(fn) {
       db.exec("BEGIN");
       try {