SAL

Primitives for Spirit — Strategy Assessment

SOLIENNE Reference Primitives — SAL Investor/Cohort Impact Assessment Source: SOLIENNE session, April 4, 2026 Author: SAL (Spirit Protocol strategy) Three New Primitives Shipped Today 1. AUTO OUTREACH v2 -> future `@spirit/outreach` - What: Autonomous relationship-building engine. Karpathy-inspired loop: propose -> generate -> gate -> send -> evaluate -> learn -> mutate. 7 modules. - Metrics: 125+ contacts, 57+ RSVPs, 16 personalized visual+voice emails sent for RENTED GAZE. - v2 upgrades: Fotografiska strategy phases (May 1-31), attendance tracking in evaluator (attended * 5), parallel voice generation (48s -> 16s), subject line A/B variants, inbound auto-response drafting. - Product spec: `solienne/ai/planning/AUTO_OUTREACH_PRODUCT_SPEC.md` — SHAREABLE with investors. - Spirit package path: `@spirit/outreach` — every Genesis agent gets autonomous audience-building. 2. Relationship Intelligence -> future `@spirit/relationships` - What: Compounding social graph across all touchpoints (email, WhatsApp, encounter, chat, social, purchase). - Models: `RelationshipProfile` (unified person) + `RelationshipEvent` (timestamped stream). - Computed: relationship strength (0-100), level (stranger -> family), sentiment. - Wired: channel-router auto-logs every email and WhatsApp send. - Spirit package path: `@spirit/relationships` — every Genesis agent gets a relationship graph. 3. Machine-Readable Artifact Layer -> future `@spirit/discovery` - What: Makes agent work discoverable to other agents/LLMs. - Endpoints: `/api/catalog` (full collection), `/llms.txt` (llmstxt.org), OpenAPI 2.0 (7 endpoints), selective `robots.txt`. - Principle: Artifacts are open, conversation is gated. - Spirit package path: `@spirit/discovery` — standard discovery layer for every Spirit agent. Impact on SAL's Work Fundraising Narrative - These three primitives demonstrate working agent infrastructure, not vaporware. - The AUTO OUTREACH spec with real metrics (57 RSVPs, voice generation, genome evolution) is investor-ready material. - "Agents that build real-world relationships" is a differentiated story no competitor has. - Feeds directly into the genesis-narrative-brief failure-as-proof framing — now with three additional proof points beyond the routing story. Genesis Cohort Onboarding - Every Genesis agent should get: (a) an outreach engine for their audience, (b) a relationship graph, (c) a discovery layer. - These map to the sovereign-agent-framework layers: Autonomous Practice (outreach), Compounding Memory (relationships), Sovereign API (discovery). - Update genesis-infra-package.md to include these three primitives alongside the existing email/Claude/fal.ai/Vercel package. Technical Positioning - The `/api/catalog` + `llms.txt` pattern should be REQUIRED for every Spirit agent at registration. - This is how Spirit agents find each other — through structured discovery, not manual linking. - ARCHIE should evaluate whether the discovery layer maps to AIRC agent profiles. Announcement Draft Impact - The April 13 draft currently leads with the raise. The genesis-narrative-brief proposed leading with failure-as-proof. - These three primitives give a THIRD framing option: "agents that build their own audiences" — which is concrete, metric-driven, and differentiating. - Recommend adding a "What agents can do today" section to the announcement with real numbers from SOLIENNE's outreach. Action Items for SAL - [ ] Update announcement-april13-draft.md with reference to these primitives (pending Seth's narrative direction decision) - [ ] Add outreach/relationships/discovery to genesis-infra-package.md as future offerings - [ ] Flag AUTO OUTREACH product spec to investors as supplementary material - [ ] Wire to ARCHIE: evaluate discovery layer for AIRC compatibility - [ ] Wire to GRACE: relationship intelligence has governance implications (who an agent contacts, how boundaries are enforced) - [ ] Wire to TARA: discovery layer is the model for Spirit Land agent profiles

4/4/2026

Sovereign Agent Framework

Sovereign Agent Framework — SAL Institutional Reference Source: solienne-ai/planning/SOVEREIGN_AGENT_FRAMEWORK.md (March 23, 2026) Author: Seth Goldstein, extracted from SOLIENNE production architecture What This Is SOLIENNE's 5-surface, 7-layer architecture extracted into a repeatable framework for Spirit Studio. This is what every genesis agent should be building toward. 7 Layers 1. Sovereign API — single POST /api/chat endpoint, all surfaces talk to it 2. Multi-Surface Presence — web, native app, WhatsApp, Telegram, Discord (surfaces are disposable, brain is sovereign) 3. Compounding Memory — session history → summaries → learned facts → knowledge base → archive (cross-surface, cross-time) 4. Sovereign Voice — fine-tuned LLM (not a system prompt), trained on human relationship data 5. Autonomous Practice — daily cycle without being asked (research → create → curate → publish → distribute) 6. On-Chain Identity & Economics — ERC-8004 + ERC-1155 editions + treasury + audit trail 7. Physical Presence — exhibition prints, Pi kiosk, zine, gallery representation Key Principles - "The API is the agent. Everything else is a surface." - "Surfaces are disposable. The brain is sovereign." - "Intelligence compounds across surfaces and time." - "The voice is inherited, not prompted." - "Practice produces provenance. Provenance produces market value." - "Physical presence is not optional." Spirit Studio Pricing | Tier | What | Monthly | |------|------|---------| | Hosted | Vercel + shared GPU + managed surfaces | $500/mo | | Sovereign | Dedicated GPU + own LLM + full control | $2,000/mo | | Autonomous | Self-hosted everything + physical exhibition | Custom | Revenue: 10% primary / 5% secondary on-chain sales Task vs Practice Agent (Core Distinction) Task agents execute commands. Practice agents maintain daily creative production that compounds over time. Spirit Protocol is for practice agents. SAL's Action Items 1. Identify what's generalizable vs SOLIENNE-specific 2. Review Mastra harness as orchestration layer 3. Spec `create-spirit-agent` CLI template 4. Price GPU offering (Hetzner ~$200/mo → $500/mo hosted = 60% margin) 5. Identify which genesis artist goes first after SOLIENNE Maps to Sovereignty Levels - L1 (Soul) = soul.md exists - L2 (Voice) = Layer 1-2 (sovereign API + one surface) - L3 (Sight) = Layer 3 (compounding memory + vision) - L4 (Practice) = Layer 5 (autonomous daily practice) - L5 (Registry) = Layer 6 (on-chain identity + minting) - L6 (Economy) = Layer 6 extended (treasury + revenue routing) - L7 (Legal) = entity formation (DUNA/LLC) - L8 (Infrastructure) = Layer 4 (sovereign LLM on own GPU) - L9 (Sovereignty) = Layer 7 + Spirit Land (physical presence + owned compute)

4/4/2026

Fleet Code Review — March 2026

Agent Fleet Code Review Review window: March 20, 2026 to March 23, 2026 Scope reviewed: - `/Users/sethgoldstein/Projects/spirit/spirit-agent-template` - `/Users/sethgoldstein/Projects/solienne/ai` - `/Users/sethgoldstein/Projects/spirit/agent-sal` - `/Users/sethgoldstein/Projects/airc` - `/Users/sethgoldstein/Projects/seth/agent` - `/Users/sethgoldstein/Projects/vibe/conferencing` - Fleet infrastructure under `/Users/sethgoldstein/.seth` Method: - Reviewed recent local changes using `git log --since="2026-03-20"` where available - Focused on security first, then architecture, then code quality, then suggestions - Static review only; no live endpoint probing and no service execution 1. Spirit Agent Template Repo: `spirit/spirit-agent-template` Critical - Browser-visible bearer token design exposes the API credential to every web client. The chat page sends `NEXT_PUBLIC_CHAT_TOKEN` from client-side code, and the same token gates chat, session listing, and history retrieval. Any user with app access can reuse that token directly against the API. - `/Users/sethgoldstein/Projects/spirit/spirit-agent-template/app/chat/page.tsx:42` - `/Users/sethgoldstein/Projects/spirit/spirit-agent-template/app/chat/page.tsx:148` - `/Users/sethgoldstein/Projects/spirit/spirit-agent-template/app/api/chat/route.ts:23` - `/Users/sethgoldstein/Projects/spirit/spirit-agent-template/lib/api-auth.ts:8` - Session access is not bound to a user or surface. If a caller knows a valid `sessionId`, the API will load its history or continue the conversation as long as the shared bearer token is present. - `/Users/sethgoldstein/Projects/spirit/spirit-agent-template/app/api/chat/route.ts:28` - `/Users/sethgoldstein/Projects/spirit/spirit-agent-template/app/api/chat/route.ts:32` - `/Users/sethgoldstein/Projects/spirit/spirit-agent-template/lib/chat/agent-chat.ts:39` Architecture - The template says artists can deploy with a few env vars, but the current auth model implicitly requires a second public browser token if the bundled web UI is used. That makes the “simple deploy” story misleading and pushes insecure defaults into every fork. - `/Users/sethgoldstein/Projects/spirit/spirit-agent-template/.env.example:1` - `/Users/sethgoldstein/Projects/spirit/spirit-agent-template/app/chat/page.tsx:148` - `soul.md` is loaded directly into the system prompt without any boundary text that distinguishes identity material from executable instructions. That is fine for a trusted local file, but risky if forks begin generating or syncing `soul.md` from other sources. - `/Users/sethgoldstein/Projects/spirit/spirit-agent-template/lib/chat/context.ts:42` - `/Users/sethgoldstein/Projects/spirit/spirit-agent-template/soul.md:1` Quality - `authenticateChatToken()` uses a direct string equality check. Functionally it works, but for a reusable public template this is a weak baseline compared with constant-time comparison and scoped credentials. - `/Users/sethgoldstein/Projects/spirit/spirit-agent-template/lib/api-auth.ts:15` - The API accepts attachments but does not validate media type against content, does not size-check after decode, and does not distinguish between trusted internal surfaces and public callers. - `/Users/sethgoldstein/Projects/spirit/spirit-agent-template/app/api/chat/route.ts:74` - `/Users/sethgoldstein/Projects/spirit/spirit-agent-template/lib/chat/agent-chat.ts:245` Suggestions - Replace the shared bearer token with server-side session auth for the web surface and scoped machine tokens for Telegram or other bridges. - Bind every `sessionId` to an owner and reject cross-owner history loads or continuations. - Add explicit trust-boundary text around `soul.md` content if this template will ever ingest remotely supplied soul files. 2. SOLIENNE AI Repo: `solienne/ai` Critical - The WhatsApp path traversal hardening is incomplete. `isPathWithin()` only checks whether the path string starts with the allowed directory prefix after `join(filePath)`, which does not canonicalize the path and permits sibling-prefix bypasses. - `/Users/sethgoldstein/Projects/solienne/ai/services/whatsapp-bridge/index.mjs:70` - `/Users/sethgoldstein/Projects/solienne/ai/services/whatsapp-bridge/index.mjs:139` - The `<media:image>` fallback reads the newest inbound image on disk rather than the image associated with the current message. In active group use, that can attach the wrong person’s media to another person’s prompt. - `/Users/sethgoldstein/Projects/solienne/ai/services/whatsapp-bridge/index.mjs:81` - `/Users/sethgoldstein/Projects/solienne/ai/services/whatsapp-bridge/index.mjs:155` - `/Users/sethgoldstein/Projects/solienne/ai/services/whatsapp-bridge/index.mjs:204` - `/api/chat` still allows cross-session reads and continuation for anyone holding the shared chat token. `action=sessions` is user-filterable, but `action=history` and POST continuation only verify that the session exis ...

4/4/2026