Wealth OS: spawn Investment Intelligence vertical + trade-gate MCP + Wealth IS ACL collapse

.claude/commands/invest-retro.md

@@ -0,0 +1,20 @@
+# /invest-retro
+
+Outcome-grounded retrospective — the learning loop. Reads trajectory records (thesis→action→outcome→lesson), scores decisions against realized outcomes, and distills lessons into per-agent `calibration_notes` in `engine/agents/catalog.json`.
+
+## Usage
+
+```
+/invest-retro [--since YYYY-WW] [--thesis <slug>]
+```
+
+## Contract
+
+Load `verticals/investment-intelligence/SKILL.md` and execute `verticals/investment-intelligence/engine/commands/retrospective.md` under it:
+
+- R5 non-advisory clause opens the output.
+- Every acted-on recommendation since the last retro must have a trajectory record (`engine/schemas/trajectory.schema.json`) — missing records are flagged as pipeline drift (SOUL test 4).
+- Lessons update agent calibration notes; systemic lessons propose skill/command edits as diffs for human review (never silent self-modification).
+- Cross-check the trade-gate audit JSONL: every executed order must trace to an approved intent. Discrepancy = escalate, do not rationalize.
+
+**Built on SIP** — invest-retro wrapper · v0.1

.claude/commands/invest-snapshot.md

@@ -0,0 +1,20 @@
+# /invest-snapshot
+
+Compose the cross-asset portfolio snapshot — ground truth before any decision. Reads (as available): Ghostfolio API (T0), DPI ledger state (`/wealth-dpi`), operator instance data (private, aggregate-only off-T0), Crypto IS House of Allocation outputs.
+
+## Usage
+
+```
+/invest-snapshot [--aggregate]   # --aggregate is forced automatically off-T0
+```
+
+## Contract
+
+Load `verticals/investment-intelligence/SKILL.md` and execute `verticals/investment-intelligence/engine/commands/portfolio-snapshot.md` under it:
+
+- R5 non-advisory clause opens the output.
+- Output validates against `engine/schemas/portfolio-snapshot.schema.json`.
+- **Data classification enforced:** real balances/positions are T0-only per `ROUTING.md`. On T1/T2 models, emit percentage weights and band-labels, never raw amounts.
+- A snapshot is not a strategy session — no recommendations from this command.
+
+**Built on SIP** — invest-snapshot wrapper · v0.1

.claude/commands/invest-strategy.md

@@ -0,0 +1,21 @@
+# /invest-strategy
+
+Run the Investment Intelligence weekly strategy session — the full pipeline (blind-parallel analysis → debate → risk gate → synthesis) producing a draft session document for human verdict. Namespaced `invest-*` to avoid collision with operator-instance commands.
+
+## Usage
+
+```
+/invest-strategy [--week YYYY-WW] [--mode weekly|rebalance]
+```
+
+## Contract
+
+Load `verticals/investment-intelligence/SKILL.md` (vertical-tier rules fire) and execute `verticals/investment-intelligence/engine/commands/weekly-strategy.md` under them:
+
+- R5 non-advisory clause opens the output. Non-waivable.
+- Pipeline order structural: analysis agents run blind-parallel per `engine/agents/catalog.json`; risk vetoes size, never direction; only the portfolio-manager emits TradeIntents.
+- Any non-DCA TradeIntent terminates as a pending-approval object via the trade-gate MCP (`mcp/trade-gate/`) — never a fabricated fill.
+- Real balances/positions are T0 data per `ROUTING.md`: on non-local models, work from aggregates.
+- Session artifact validates against `engine/schemas/strategy-session.schema.json`; acted-on recommendations get trajectory records.
+
+**Built on SIP** — invest-strategy wrapper · v0.1

.claude/commands/invest-thesis-debate.md

@@ -0,0 +1,20 @@
+# /invest-thesis-debate
+
+Bull/bear adversarial debate on a single investment thesis (TradingAgents-pattern, provenance: `verticals/investment-intelligence/docs/absorption/tradingagents.md`). Pressure-test before any new position thesis enters the portfolio.
+
+## Usage
+
+```
+/invest-thesis-debate <thesis-slug or free-text thesis>
+```
+
+## Contract
+
+Load `verticals/investment-intelligence/SKILL.md` and execute `verticals/investment-intelligence/engine/commands/thesis-debate.md` under it:
+
+- R5 non-advisory clause opens the output.
+- Bull and bear cases argued from independently-formed analysis-layer stances (blind-parallel first, collide second).
+- Verdict is a thesis-quality assessment (validated against `engine/schemas/thesis.schema.json`), never a buy/sell instruction.
+- Optional persona overlays (value/contrarian/quality lenses per `AGENTS.md` § persona overlays) increase debate diversity; they never bypass the risk layer.
+
+**Built on SIP** — invest-thesis-debate wrapper · v0.1

.claude/commands/siswarm.md

@@ -0,0 +1,61 @@
+# /siswarm — Investment-intelligence council (Sonnet + Opus swarm)
+
+Run the 10-agent investment-intelligence council locally, with per-agent model
+routing threaded from `verticals/investment-intelligence/engine/agents/catalog.json`.
+This is the executable counterpart to the `/si`//`/so` CLI-lane routers: `/si`
+decides *which* CLI to use; `/siswarm` actually *runs* a Sonnet+Opus council.
+
+## Usage
+
+```
+starlight swarm run "<operator context>"                 # dry-run plan (default)
+starlight swarm run "<operator context>" --live          # execute via headless claude
+starlight swarm run "<context>" --live --only technical  # single-agent cheap demo
+starlight swarm run "<context>" --live --concurrency 3 --timeout 240000
+```
+
+Dry-run prints the compiled plan (agent · layer · model · prompt size) and passes
+**no model to any CLI**. `--live` executes.
+
+## The three phases (from the catalog handoff protocol)
+
+| Phase | Agents | Models | Prompt sees |
+|---|---|---|---|
+| 1 — analysis (blind-parallel) | macro-risk · crypto-dca · defi-yield · fundamentals · technical | Sonnet ×4, **Haiku** (technical) | operator context only — **no peer output** |
+| 2 — risk | risk-manager · tax-optimizer · regulatory-risk | Sonnet ×3 | context + the full analysis digest |
+| 3 — synthesis | portfolio-manager · chief-of-staff | **Opus** (PM), Sonnet | context + the full debate (analysis + risk) |
+
+The `researcher` (cross-cutting) is excluded from v1 — it is a lookup helper, not
+a debate seat. Risk agents run parallel *within* phase 2 (each only needs the
+analysis output, not each other's) — an intentional refinement over the catalog's
+`step_2_sequential`.
+
+## How the model split works
+
+The runner (`src/swarm.ts::defaultClaudeRunner`) spawns `claude -p <prompt> --model <id>`.
+Each task's `model` is the agent's `recommended_model` from the catalog, so
+analysis/risk run Sonnet-class, `technical` runs Haiku, and `portfolio-manager`
+runs Opus. Override the binary with `STARLIGHT_CLAUDE_BIN`.
+
+## Boundary (non-negotiable)
+
+The council produces **analysis + a pending decision brief** — nothing else. Every
+prompt carries the R5 non-advisory clause and the no-execution line
+(`tools_denied` includes `execution-platforms` + `credential-stores`). **Trades
+execute only through the local trade-gate MCP** (`verticals/investment-intelligence/mcp/trade-gate`)
+**with a human approval token.** The swarm has no path to a broker.
+
+## Audit
+
+Each `--live` run appends a JSONL record (context, timestamps, per-phase agent
+pass/fail) to `private/voice-operator/logs/swarm-council.jsonl` next to the
+existing swarm audit log.
+
+## Hosted twin
+
+The same catalog drives the Vercel-hosted council on frankx.ai
+(`/admin/swarm` + `app/api/swarm/run` via the Vercel AI Gateway) — that surface
+also emits pending decisions only. See `docs/architecture/sis-swarm-hosted.md`
+in the FrankX repo.
+
+**Built on SIP** — /siswarm · investment-intelligence council · v0.1

REGISTRY.md

@@ -85,6 +85,22 @@ attestation: built_on_sip
 - **requires:** none
 - **license:** MIT
 
+### trade-gate-mcp
+- **sip_version:** 1.1.1
+- **owner:** Frank Riemer / Starlight Holding BV
+- **status:** active (v0.1 — paper broker only; live adapters NOT_WIRED by design)
+- **repo:** `frankxai/Starlight-Intelligence-System` (`verticals/investment-intelligence/mcp/trade-gate/`)
+- **endpoint:** stdio (local)
+- **provides:**
+  - `propose_trade` — TradeIntent in → verdict: auto-approved (DCA whitelist, capped) | pending-approval | rejected
+  - `request_approval` — human issues a single-use approval token for a pending intent
+  - `list_pending` — pending-approval queue
+  - `execute_approved` — approved intent → broker adapter (paper default; live throws NOT_WIRED in-repo)
+  - `read_audit` — append-only JSONL audit trail
+- **requires:** none (broker MCPs — Alpaca / IBKR / Coinbase — are operator-wired downstream, never bundled)
+- **license:** MIT
+- **note:** Fail-closed by doctrine and by test (`e2e.test.ts` red case: live execute without approval must throw). Every tool output carries the R5 non-advisory footer. Pattern lineage: `payment-intelligence-system/mcp`. Board record: `docs/boards/2026-07-02-investment-intelligence-vertical-spawn.md`.
+
 ## Claw Registry
 
 SIS Claws are installable operational units built on this substrate. Each Claw declares its permission surface, skill requirements, and MCP dependencies in a `CLAW.md` contract.

VERTICALS.md

@@ -196,7 +196,7 @@ Domain sub-stacks are forkable reference verticals. Unlike the sovereign vertica
 - **Class:** sovereign domain sub-stack (reference vertical) — third reference, validates the **Houses-as-sub-systems** decomposition primitive AND first composition-instance under Wealth IS composition layer (per `STACK.md` § Composition Layer, declared 2026-05-17).
 - **Domain:** Cryptocurrency practiced as the architecture of sovereignty + protocol-mechanism literacy + cycle-position pattern recognition. Six Houses (On-Chain · Macro · DeFi · Sovereignty · Research · Allocation) decomposed as archetypal stances.
 - **Owner:** open reference (forkable by sovereign crypto practitioners). Composes under Wealth IS as first reference instance of Composition Layer substrate primitive.
-- **Status:** `v0.1 proof-of-pattern — House of On-Chain scaffolded (5 commands + agent + skill + knowledge); 5 Houses gated on v0.1-proof-pass per Board R4 close-out`. Investment IS held until proof-pass.
+- **Status:** `v0.2 full scaffold — all 6 Houses shipped` (R4 proof-of-pattern passed via House of On-Chain). Investment IS gate released — spawned 2026-07-02, see § Investment Intelligence below.
 - **Primary repo:** `verticals/crypto-intelligence/` in `frankxai/Starlight-Intelligence-System` (v0.2+ extraction target: `github.com/frankxai/crypto-intelligence-system` per Board open-question (c) export-hook landed via vertical-local ATTESTATIONS.md + MCP-shape in SKILL.md).
 - **Houses (v0.1):** `onchain` (scaffolded). **Houses (v0.2 gated):** `macro` · `defi` · `sov` · `res` · `alloc`.
 - **Canon:** declines (canon-free per substrate register).
@@ -207,6 +207,20 @@ Domain sub-stacks are forkable reference verticals. Unlike the sovereign vertica
 - **Board verdict:** REVISE close-out 2026-05-17 — `docs/boards/2026-05-17-crypto-investment-spawn.md` (5 items closed same-session per v7.5.1 recovery precedent).
 - **Genius prerequisite gate:** CLOSED via `genius/profile-frankx.md` + `genius/freedom-path-frankx.md` (Path A in-repo corpus excavation, 2026-05-17).
 
+### Investment Intelligence
+- **Class:** sovereign domain sub-stack (reference vertical) — sibling of Crypto Intelligence under the Wealth IS ACL manifest (Wealth collapsed to ACL-only per 2026-06-16 falsifier default, recorded `docs/boards/2026-07-02-investment-intelligence-vertical-spawn.md`).
+- **Domain:** Cross-asset capital allocation practiced as system, not impulse — research → debate → risk-gate → recommend → human-approve → execute (paper-first) → learn, across stocks/ETFs/crypto/DeFi + time/labor allocation via DPI framing.
+- **Owner:** open reference (forkable by sovereign operators). Engine promoted (copied, not moved) from the private IIS operator instance — proven material, not fresh scaffold.
+- **Status:** `v0.1 — engine promoted (11-agent catalog · 5 schemas · 4 commands · 3 adapters · 2 sanitized archetypes), trade-gate MCP shipped + tested; runtime tiers 1-2 gated on operator wiring per RUNBOOK.md`.
+- **Primary repo:** `verticals/investment-intelligence/` in `frankxai/Starlight-Intelligence-System` (extraction target: `github.com/frankxai/investment-intelligence-system` per Board open-question (c) export-hook pattern — extraction is a copy, not a refactor).
+- **Layers:** `analysis` (5 agents, blind-parallel) · `risk` (3 agents, veto-on-size-not-direction) · `synthesis` (portfolio-manager Opus + chief-of-staff) · cross-cutting `researcher`.
+- **Canon:** declines (canon-free per substrate register); composes SIP + payment-intelligence-system doctrine (pattern-level).
+- **Compounds:** trajectory ReasoningBank (thesis→action→outcome→lesson) + per-agent calibration + practitioners forking + attestation graph.
+- **SIP commands:** `/invest-strategy` · `/invest-thesis-debate` · `/invest-snapshot` · `/invest-retro` (daily-5 completed by `/wealth-dpi`, R2 discipline declared at spawn).
+- **MCP:** `mcp/trade-gate/` — fail-closed approve/execute (propose_trade · request_approval · list_pending · execute_approved · read_audit); paper-broker default, live adapters NOT_WIRED stubs, human token above DCA whitelist.
+- **Non-waivable:** R5 non-advisory clause inline on every artifact; no broker credentials in repo; paper-first ladder; trajectory write-back.
+- **Board verdict:** PROCEED — `docs/boards/2026-07-02-investment-intelligence-vertical-spawn.md`.
+
 ---
 
 ## Registry rules