Weekend Delivery Plan

Date: 2026-04-11 Revised: 2026-04-11 (spec-reconciliation pass — all examples updated to match canonical specs; workspace install note added) Revised: 2026-04-11 (sdk-audit-and-traits-alignment-plan — implementation status reflected; workspace:* gap documented; traits/persona context added) Target: 2026-04-13 (Sunday night)

Canonical source of truth: Package specs in docs/specs/ override any example code in this document. All assembly examples below have been updated to match the reviewed specs and docs/architecture/spec-reconciliation-rules.md. If a code example conflicts with a spec, trust the spec, not this document.

Implementation Status as of 2026-04-11

WF-1 through WF-5 are COMPLETE. The core, sessions, and surfaces packages are implemented and passing tests. The weekend delivery goal was to produce stable v1 type contracts for Sage, MSD, and NightCTO to write product adapter code against. That goal is met.

Package	Tests	Status
`@agent-assistant/core`	44 pass	COMPLETE
`@agent-assistant/sessions`	25 pass	COMPLETE
`@agent-assistant/surfaces`	28 pass	COMPLETE

WF-6 is COMPLETE. packages/core/src/core-sessions-surfaces.test.ts (describe block labeled WF-6) covers multi-surface session attachment, fanout, targeted send, and detach behavior.

WF-7 is OPEN. The end-to-end assembly test in packages/examples/src/ does not yet exist. Package READMEs for core (152 lines), sessions (118 lines), and surfaces (175 lines) are substantive — not placeholders. The v1 release tag is gated on the assembly test being written.

Remaining open items:

@agent-assistant/routing has a blocking DoD failure (12 tests, 40+ required). Do not wire routing into product code until this is resolved.
No root package.json or monorepo workspace config exists. The workspace:* protocol referenced below is the target pattern, not current reality. Use npm pack tarballs or path-based installs until a root workspace is configured.

Goal

Sage, MSD, and NightCTO teams can npm install @agent-assistant/core @agent-assistant/sessions @agent-assistant/surfaces by Sunday night, with type contracts stable enough to write product adapter code against.

npm install note: For v1, "npm install" means local monorepo consumption via workspace references ("@agent-assistant/core": "workspace:*") or npm pack tarballs — not the public npm registry. Public publishing is a post-v1 task tracked separately.

Workspace config gap: No root package.json with workspace configuration currently exists. Each package is independently installable. Until a workspace root is configured, consume packages via npm pack tarballs or local path references. This is tracked in the audit plan as D-5.

The v1 type contracts that are now stable:

AssistantDefinition (core)
AssistantRuntime (core)
InboundMessage / OutboundEvent (core)
CapabilityHandler / CapabilityContext (core)
Session / SessionStore (sessions)
AffinityResolver / resolveSession (sessions)
SurfaceRegistry / SurfaceConnection / SurfaceAdapter (surfaces)
SurfaceCapabilities / SurfaceFormatHook / FanoutResult (surfaces)

Traits and Persona Context

Products using this SDK should understand the distinction between workforce personas and assistant traits before writing product adapter code.

Workforce personas are runtime execution profiles (model, harness, system prompt, service tier). These are defined and owned in Workforce infrastructure. They are not imported from this SDK.

Assistant traits are identity and behavioral characteristics (voice, style, vocabulary, proactivity level). The @agent-assistant/traits package is planned for v1.2. In v1, products define traits as local data objects and inject them manually into persona prompts and format hooks.

See traits-and-persona-layer.md for the full boundary definition.

Timeline

Saturday Morning (2026-04-12, first half)

Status: COMPLETE (implementation already done)

All three specs are already IMPLEMENTATION_READY. WF-1, WF-2, WF-3 implementations are done and passing.

Task	Deliverable	Status
Read and confirm core spec	Mental model of `AssistantDefinition`, `AssistantRuntime`, adapters	DONE
Read and confirm sessions spec	Mental model of `Session`, `SessionStore`, `InMemorySessionStoreAdapter`	DONE
Read and confirm surfaces spec	Mental model of `SurfaceRegistry`, `SurfaceConnection`, fanout vs targeted send	DONE
Scaffold `packages/core`	`package.json`, `tsconfig.json`, `src/index.ts`, `src/types.ts`	DONE
Scaffold `packages/sessions`	same	DONE
Scaffold `packages/surfaces`	same	DONE

Saturday Afternoon (2026-04-12, second half)

Status: COMPLETE

Workflow	Package	Key output	Status
WF-1: Define assistant and start runtime	core	`createAssistant`, `AssistantDefinition` validation, `AssistantRuntime`, `runtime.start()` / `runtime.stop()`, `runtime.status()`	COMPLETE — 44 tests
WF-2: Handle inbound message via dispatch	core	Capability dispatch table, `runtime.dispatch()`, `AssistantHooks.onMessage` pre-filter, `runtime.emit()`	COMPLETE
WF-3: Create and manage sessions	sessions	`createSessionStore`, `InMemorySessionStoreAdapter`, full lifecycle (`touch`, `expire`, `sweepStale`), `attachSurface`, `detachSurface`, `resolveSession`	COMPLETE — 25 tests

Saturday Evening / Sunday Morning (2026-04-12 evening – 2026-04-13 morning)

Status: COMPLETE

Workflow	Packages	Key output	Status
WF-4: Wire session store into runtime	core + sessions	`runtime.register('sessions', store)`, `runtime.get<SessionStore>('sessions')`, `resolveSession` in handler	COMPLETE
WF-5: Register surface registry and route messages	core + surfaces	`createSurfaceRegistry`, `SurfaceConnection`, adapter wiring as core relay adapters, inbound normalization, outbound targeted send, `formatHook`	COMPLETE — 28 tests

Sunday Afternoon (2026-04-13 afternoon)

Status: UNCERTAIN — verify before marking complete

Workflow	Packages	Key output	Status
WF-6: Multi-surface session fanout	core + sessions + surfaces	Cross-surface session attachment, `surfaceRegistry.fanout()`, targeted-vs-fanout rule validated	COMPLETE — `packages/core/src/core-sessions-surfaces.test.ts`
WF-7: End-to-end assembly	core + sessions + surfaces	Full inbound→session→handler→emit→format→adapter cycle, validated assembly, examples package	OPEN — `packages/examples/src/` not yet created

Sunday Night (2026-04-13)

Status: OPEN — consumer readiness verification needed

Each product team runs the consumer readiness checklist against the released packages:

npm install @agent-assistant/core @agent-assistant/sessions @agent-assistant/surfaces (resolves via workspace protocol or local tarballs — not the public npm registry for v1)
Define an assistant with createAssistant(definition, adapters) where definition.capabilities is Record<string, CapabilityHandler>
Wire a SessionStore via runtime.register('sessions', createSessionStore({ adapter }))
Register surfaces via createSurfaceRegistry() and wire it as the core relay adapter pair
Handle InboundMessage through capability dispatch and emit OutboundEvent via context.runtime.emit()
Return formatted responses to the originating surface via targeted send or fanout
Write product adapter code against stable v1 types
Run the SDK's own tests to verify correct integration

Tag v1 release once all checks pass.

Product-Specific Adoption Paths

Sage Adoption Path

Weekend goal: Wire core + sessions + surfaces. Draft a memory adapter interface stub so v1.1 memory integration can start Monday.

Step	Action
v1	Install core, sessions, surfaces. Define the Sage assistant identity using `createAssistant()`. Wire `createSessionStore({ adapter: new InMemorySessionStoreAdapter() })`. Register a Slack `SurfaceConnection` in a `SurfaceRegistry`.
Immediate after v1	Begin adapter stub for `@agent-assistant/memory` (v1.1). Sage's existing memory patterns are the primary signal for the memory spec.
v1.1 gates	Full memory persistence across Sage sessions. Proactive follow-up engine.
v1.2 gates	`@agent-assistant/traits` — Sage is a primary extraction signal. Define local `sageTraits` object now so the v1.2 extraction has a concrete pattern to generalize from.

What stays in Sage for now:

Knowledge and workspace-specific prompt behavior
Workforce persona definitions (model, harness, system prompt, tier) — these are workforce-owned, not SDK concerns
Product-specific follow-up heuristics
Slack-specific UI conventions and block kit templates
Trait values (voice, style, vocabulary) — define as a local data object; @agent-assistant/traits ships at v1.2
Memory retrieval logic (until v1.1 @agent-assistant/memory ships)

Sage v1 minimum viable assembly:

import { createAssistant } from "@agent-assistant/core";
import type { AssistantDefinition, InboundMessage, CapabilityContext } from "@agent-assistant/core";
import { createSessionStore, InMemorySessionStoreAdapter, resolveSession, createDefaultAffinityResolver } from "@agent-assistant/sessions";
import type { SessionStore } from "@agent-assistant/sessions";
import { createSurfaceRegistry } from "@agent-assistant/surfaces";
import type { SurfaceConnection, SurfaceCapabilities } from "@agent-assistant/surfaces";

// 1. Define the Sage assistant
const definition: AssistantDefinition = {
  id: "sage",
  name: "Sage",
  capabilities: {
    chat: async (message: InboundMessage, context: CapabilityContext) => {
      const store = context.runtime.get<SessionStore>("sessions");
      const session = await resolveSession(
        message,
        store,
        createDefaultAffinityResolver(store),
      );
      await store.touch(session.id);

      // Sage domain handler — product-owned logic
      await context.runtime.emit({
        surfaceId: message.surfaceId,
        sessionId: session.id,
        text: "...", // Sage-specific response
      });
    },
  },
};

// 2. Wire sessions
const sessionStore = createSessionStore({
  adapter: new InMemorySessionStoreAdapter(),
});

// 3. Wire surfaces
const slackCapabilities: SurfaceCapabilities = {
  markdown: false,
  richBlocks: true,
  attachments: true,
  streaming: false,
  maxResponseLength: 3000,
};

const slackConnection: SurfaceConnection = {
  id: "sage-slack",
  type: "slack",
  state: "registered",
  capabilities: slackCapabilities,
  adapter: stubSlackAdapter, // provided by relay foundation or product code
  formatHook: (event, caps) => ({ blocks: [{ type: "section", text: event.text }] }),
};

const surfaceRegistry = createSurfaceRegistry();
surfaceRegistry.register(slackConnection);

// 4. Create runtime and register subsystems
const runtime = createAssistant(definition, {
  inbound: surfaceRegistry,
  outbound: surfaceRegistry,
});
runtime.register("sessions", sessionStore);

await runtime.start();

MSD Adoption Path

Weekend goal: Wire core + sessions + surfaces. MSD's cross-surface session design maps directly onto the v1 session model. Focus on the Slack + web multi-surface path.

Step	Action
v1	Install core, sessions, surfaces. Define the MSD assistant identity. Wire session store. Register Slack and web surface connections in the surface registry.
After v1	Stub `@agent-assistant/policy` interface for approval-mode scaffolding (policy ships in v2 but MSD can define the interface contract early as a passthrough).
v1.2 gates	Coordination for review orchestration. Policy for external action governance.

What stays in MSD for now:

Code review operations and PR workflows
Review-specific orchestration logic
PR-specific tools and heuristics
Workforce persona definitions — owned by Workforce, not imported from SDK
Coordinator delegation (until v1.2 @agent-assistant/coordination ships for product use)

MSD v1 minimum viable assembly:

import { createAssistant } from "@agent-assistant/core";
import type { AssistantDefinition, InboundMessage, CapabilityContext } from "@agent-assistant/core";
import { createSessionStore, InMemorySessionStoreAdapter, resolveSession, createDefaultAffinityResolver } from "@agent-assistant/sessions";
import type { SessionStore } from "@agent-assistant/sessions";
import { createSurfaceRegistry } from "@agent-assistant/surfaces";
import type { SurfaceConnection } from "@agent-assistant/surfaces";

const definition: AssistantDefinition = {
  id: "msd-review-assistant",
  name: "MSD",
  capabilities: {
    review: async (message: InboundMessage, context: CapabilityContext) => {
      const store = context.runtime.get<SessionStore>("sessions");
      const session = await resolveSession(
        message,
        store,
        createDefaultAffinityResolver(store),
      );
      await store.touch(session.id);
      await store.attachSurface(session.id, message.surfaceId);

      // MSD review handler — product-owned logic

      // Targeted send: reply to originating surface
      await context.runtime.emit({
        surfaceId: message.surfaceId,
        sessionId: session.id,
        text: "...", // MSD-specific response
      });

      // Session fanout: notify ALL attached surfaces (surfaceId absent, sessionId present)
      // await context.runtime.emit({
      //   sessionId: session.id,
      //   text: "PR review complete — notifying all attached surfaces",
      // });
    },
  },
};

const sessionStore = createSessionStore({ adapter: new InMemorySessionStoreAdapter() });

const slackConnection: SurfaceConnection = {
  id: "msd-slack",
  type: "slack",
  state: "registered",
  capabilities: { markdown: false, richBlocks: true, attachments: true, streaming: false, maxResponseLength: 3000 },
  adapter: stubSlackAdapter,
};

const webConnection: SurfaceConnection = {
  id: "msd-web",
  type: "web",
  state: "registered",
  capabilities: { markdown: true, richBlocks: false, attachments: false, streaming: true, maxResponseLength: 0 },
  adapter: stubWebAdapter,
};

const surfaceRegistry = createSurfaceRegistry();
surfaceRegistry.register(slackConnection);
surfaceRegistry.register(webConnection);

const runtime = createAssistant(definition, {
  inbound: surfaceRegistry,
  outbound: surfaceRegistry,
});
runtime.register("sessions", sessionStore);

await runtime.start();

NightCTO Adoption Path

Weekend goal: Wire core + sessions + surfaces. NightCTO exercises more of the SDK than Sage or MSD, but v1 is still the foundation.

Step	Action
v1	Install core, sessions, surfaces. Define NightCTO assistant identity. Wire session store. Register primary Slack surface connection in the registry.
v1.1 gates	Memory for per-client continuity. Connectivity for multi-component flows.
v1.2 gates	Proactive monitoring and digests. Coordination for specialist orchestration. Routing for model selection and depth/cost decisions. `@agent-assistant/traits` — NightCTO is a primary extraction signal alongside Sage.

What stays in NightCTO for now:

Founder/CTO communication style and persona — trait values defined locally until v1.2
Workforce persona definitions — owned by Workforce
Client-tier and service policy
Domain-specific specialist lineups (until v1.2 @agent-assistant/coordination ships)
Per-client memory (until v1.1 @agent-assistant/memory ships)
Proactive monitoring (until v1.2 @agent-assistant/proactive ships)

NightCTO v1 minimum viable assembly:

import { createAssistant } from "@agent-assistant/core";
import type { AssistantDefinition, InboundMessage, CapabilityContext } from "@agent-assistant/core";
import { createSessionStore, InMemorySessionStoreAdapter, resolveSession, createDefaultAffinityResolver } from "@agent-assistant/sessions";
import type { SessionStore } from "@agent-assistant/sessions";
import { createSurfaceRegistry } from "@agent-assistant/surfaces";
import type { SurfaceConnection } from "@agent-assistant/surfaces";

const definition: AssistantDefinition = {
  id: "nightcto",
  name: "NightCTO",
  capabilities: {
    // v1: single general capability; v1.2 adds specialist routing
    advise: async (message: InboundMessage, context: CapabilityContext) => {
      const store = context.runtime.get<SessionStore>("sessions");
      const session = await resolveSession(
        message,
        store,
        createDefaultAffinityResolver(store),
      );
      await store.touch(session.id);
      await store.attachSurface(session.id, message.surfaceId);

      // NightCTO domain handler — product-owned logic
      await context.runtime.emit({
        surfaceId: message.surfaceId,
        sessionId: session.id,
        text: "...", // NightCTO-specific response
      });
    },
  },
  hooks: {
    onStart: async (runtime) => {
      // NightCTO: register any startup behavior here
    },
  },
};

const sessionStore = createSessionStore({ adapter: new InMemorySessionStoreAdapter() });

const slackConnection: SurfaceConnection = {
  id: "nightcto-slack",
  type: "slack",
  state: "registered",
  capabilities: { markdown: false, richBlocks: true, attachments: true, streaming: false, maxResponseLength: 3000 },
  adapter: stubSlackAdapter,
  formatHook: (event, caps) => ({
    // NightCTO Slack formatting — product-owned
    blocks: [{ type: "section", text: { type: "mrkdwn", text: event.text } }],
  }),
};

const surfaceRegistry = createSurfaceRegistry();
surfaceRegistry.register(slackConnection);

const runtime = createAssistant(definition, {
  inbound: surfaceRegistry,
  outbound: surfaceRegistry,
});
runtime.register("sessions", sessionStore);

await runtime.start();

Minimum Shippable v1 (WF-1 through WF-5)

WF-1 through WF-5 are complete. The minimum shippable v1 is achieved. Products can begin adapter work now.

WF-6 and WF-7 remain to validate full integration and update package documentation. They are not blockers for product adoption but are required for the v1 release tag.

Risk Flags

Risk	Mitigation
Routing consumed before DoD is cleared	Do not wire `@agent-assistant/routing` into product integrations until F-1 (tests) and F-2 (escalated flag) are resolved
No monorepo workspace config exists	Use npm pack or path references until a root package.json is configured; do not assume workspace:* works
Traits/persona conflation	Workforce personas stay in workforce; SDK traits are v1.2; in v1 define trait values as local product data objects
Spec vocabulary drift causes wrong implementation	Treat `docs/specs/v1-*-spec.md` as the only source of truth; discard any planning doc that conflicts
WF-7 README updates not done	Package READMEs are still placeholders; verify and update before tagging v1 release

After the Weekend: Immediate Next Steps

Once v1 is tagged:

NightCTO and Sage teams begin writing product adapter code against v1 types, starting with memory adapter stubs
Draft docs/workflows/connectivity-wf-*.md (four workflow specs required by the v1.1 connectivity spec)
MSD team validates cross-surface session behavior against their existing session architecture using WF-6 as the reference test
Begin v1.1 implementation planning — memory spec is already IMPLEMENTATION_READY; connectivity spec is already IMPLEMENTATION_READY
Resolve routing DoD failures (F-1, F-2) before any product attempts to consume @agent-assistant/routing
Define local traits objects in Sage, MSD, and NightCTO — these are the v1.2 extraction signal for @agent-assistant/traits

Reuse-First Implementation Note

Weekend implementation work should not assume every package is greenfield. Agents should inspect existing relay packages first for reusable capabilities.

For memory specifically:

prefer the existing @agent-relay/memory package
only build new assistant-side memory code where the assistant contract needs an adapter, composition layer, or assistant-specific extension

This applies to all future package work. Check what exists in Relay/AgentWorkforce repos before building.

Out of Scope For This Weekend

The following is intentionally deferred beyond the weekend build-out:

cross-agent memory consolidation / librarian agent behavior

That capability is better treated as a v5-v8 level feature after basic memory, coordination, connectivity, and routing layers are stable.

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Weekend Delivery Plan

Implementation Status as of 2026-04-11

Goal

Traits and Persona Context

Timeline

Saturday Morning (2026-04-12, first half)

Saturday Afternoon (2026-04-12, second half)

Saturday Evening / Sunday Morning (2026-04-12 evening – 2026-04-13 morning)

Sunday Afternoon (2026-04-13 afternoon)

Sunday Night (2026-04-13)

Product-Specific Adoption Paths

Sage Adoption Path

MSD Adoption Path

NightCTO Adoption Path

Minimum Shippable v1 (WF-1 through WF-5)

Risk Flags

After the Weekend: Immediate Next Steps

Reuse-First Implementation Note

Out of Scope For This Weekend

FilesExpand file tree

weekend-delivery-plan.md

Latest commit

History

weekend-delivery-plan.md

File metadata and controls

Weekend Delivery Plan

Implementation Status as of 2026-04-11

Goal

Traits and Persona Context

Timeline

Saturday Morning (2026-04-12, first half)

Saturday Afternoon (2026-04-12, second half)

Saturday Evening / Sunday Morning (2026-04-12 evening – 2026-04-13 morning)

Sunday Afternoon (2026-04-13 afternoon)

Sunday Night (2026-04-13)

Product-Specific Adoption Paths

Sage Adoption Path

MSD Adoption Path

NightCTO Adoption Path

Minimum Shippable v1 (WF-1 through WF-5)

Risk Flags

After the Weekend: Immediate Next Steps

Reuse-First Implementation Note

Out of Scope For This Weekend