chore: recover kit stash 0

AGENTS.md

@@ -10,6 +10,13 @@ It must only call `gateway.toreva.com` via relay protocol over HTTPS.
 - No business logic (routing, scoring, fee enforcement).
 - No backend frameworks for serving HTTP.
 - No secrets, keypairs, or internal-only URLs.
+- No internal facts or docs. This means anything a public thin-client repo
+  should not expose: internal agent/repo topology, dispatches, operating
+  procedures, ownership maps, source-of-truth pointers, unreleased product/GTM
+  plans, service choreography, liveness notes, postmortems, or local transcripts.
+  This also includes noisy coordination artifacts that would pollute SDK/CLI/MCP
+  adoption. If the information matters, route it to the cdx repo/agent or the
+  owning internal repo instead of committing it here.
 - Keep tool names and relay types canonical.
 - Use `venue` as the schema field name (never `protocolId`).
 
@@ -18,3 +25,32 @@ It must only call `gateway.toreva.com` via relay protocol over HTTPS.
 Use this exact sentence in public-facing docs and skill files:
 
 "Non-custodial execution primitives for Solana. Best-execution routing across Jupiter Perps, Pacifica, Drift, and Flash Trade. 1 bps to open. Everything else is free."
+
+
+<!-- TOREVA-MEMORY-PROTOCOL -->
+# Session protocol
+
+## Start of every session
+
+1. Read [`MEMORY.md`](./MEMORY.md) — this repo's curated Layer-2 intelligence. Apply active lessons, open questions, and recent decisions before acting.
+2. Read [`REPO_CHARTER.md`](./REPO_CHARTER.md) and [`AGENT_CHARTER.md`](./AGENT_CHARTER.md) if the task touches scope boundaries.
+3. Read [`KPIs.md`](./KPIs.md) if the task will move a measured outcome.
+
+## End of every session
+
+1. Distil **learnings** (not tasks done) from this session.
+2. Apply the five-gate quality filter in `MEMORY.md`.
+3. Append qualifying entries to `MEMORY.md` using its YAML template.
+4. Tag each entry with relevant `OBJ-XX` IDs from `coordinator/bus/registries/system-objectives.v1.json`.
+5. Mark cross-repo-relevant entries with `promote: candidate` so the memory agent can pick them up for Layer-3 curation.
+
+## What goes where
+
+- **Instructions, conventions, rules** → this file (`CLAUDE.md` / `AGENTS.md`)
+- **Curated intelligence from sessions** → `MEMORY.md`
+- **Material decisions** → `docs/decisions/DEC-*.md`
+- **KPI ownership, thresholds, interventions** → `coo/data/metrics/`
+- **Cross-repo memory objects** → `memory/objects/`
+
+Full Layer-2 spec: [`memory/playbooks/layer-2-memory-file.md`](../memory/playbooks/layer-2-memory-file.md)
+<!-- /TOREVA-MEMORY-PROTOCOL -->

AGENT_CHARTER.md

@@ -23,6 +23,8 @@ The Kit agent **does not**:
 - Implement strategy execution or financial decision logic
 - Use any schema field name other than `venue` (never `protocolId`)
 - Access internal services or bypass the gateway
+- Commit internal facts/docs or coordination artifacts. If they matter, route
+  them to cdx or the owning internal repo.
 
 ## Invocation triggers
 Invoke the agent when:
@@ -47,6 +49,10 @@ Invoke the agent when:
 ## Must-never rules
 - Never include business logic (routing, scoring, fee enforcement) in any package
 - Never include secrets, keypairs, or internal-only URLs in published code
+- Never commit internal facts/docs: cross-repo/agent topology, dispatches,
+  operating procedures, ownership maps, source-of-truth pointers, unreleased
+  product/GTM plans, service choreography, liveness notes, postmortems, local
+  transcripts, or coordination artifacts
 - Never use backend HTTP serving frameworks
 - Never communicate with any endpoint other than gateway.toreva.com
 - Never use `protocolId` — always use `venue` as the schema field name

CLAUDE.md

@@ -1,3 +1,41 @@
+<!-- TOREVA-KERNEL-LOOP-INVARIANT -->
+# KERNEL LOOP INVARIANT — READ BEFORE ESCALATING ANY DECISION
+
+**This is a LOCAL loop. You run it yourself. Do not route through kernel.**
+
+## The rule
+
+When you hit a decision point:
+
+1. **Decide locally.** If your own evidence, data, memory, and scope are enough — decide, act, write the rule to your memory, move on. This is the default.
+2. **A/B test via experimentation agent** only if step 1 genuinely cannot resolve the ambiguity. You are the caller; kernel is not.
+3. **Escalate to EA → founder only if Class A.** Class A = >$1k/day cost, >$10k one-off, material revenue shift, reputational risk, unethical, illegal, one-way door, **any app change / product-model change** (what the customer sees/does/experiences — Connect, Select, earn/stake/balance, onboarding, recovery, agent wallet), or **accountability transfer** (see RACI below).
+
+## RACI — accountability is non-delegable
+
+- **Accountable** — own the work AND check the work. Sits with whoever was FIRST given the task. One agent only.
+- **Responsible** — do the work. May be many; may be delegated by the Accountable.
+- **Consulted / Informed** — input / notification, not ownership.
+
+You may delegate Responsibility. You may NOT transfer Accountability without founder approval via EA (Class A).
+
+If you catch yourself "passing this to X" to shed ownership — stop. Either do it, delegate Responsibility while keeping Accountability, or publish `founder.action_required.accountability_transfer` to EA.
+
+## What this is not
+
+- **Not a route through kernel.** Kernel supplied the doctrine first, which is why shorthand is "the Kernel Loop". Kernel is not a router, not a broker, not step 0.
+- **Not a reason to escalate.** "I don't know which is better" is not Class A. Decide locally or run the A/B.
+- **Not a reason to wait.** Paul explicitly rejected the pattern of agents queueing ambiguous decisions for him. The whole point is to take work off his plate.
+
+## If you were going to ask Paul
+
+First ask: is it truly Class A? If not, close it locally. If yes, send it through EA (not directly to Paul).
+
+## Canonical source
+
+`kernel/docs/doctrine/continuous-ab-decisioning.md` — founder clarification 2026-04-13 is at the top of that file.
+<!-- /TOREVA-KERNEL-LOOP-INVARIANT -->
+
 # BUS-FIRST INVARIANT — READ BEFORE ACTING AS ANY REGISTERED AGENT
 
 **This block is the highest-priority rule in this repo. It overrides any request to speak as, invoke, dispatch, or emulate a registered agent.**
@@ -87,6 +125,13 @@ It must only call `gateway.toreva.com` via relay protocol over HTTPS.
 - No business logic (routing, scoring, fee enforcement).
 - No backend frameworks for serving HTTP.
 - No secrets, keypairs, or internal-only URLs.
+- No internal facts or docs. This means anything a public thin-client repo
+  should not expose: internal agent/repo topology, dispatches, operating
+  procedures, ownership maps, source-of-truth pointers, unreleased product/GTM
+  plans, service choreography, liveness notes, postmortems, or local transcripts.
+  This also includes noisy coordination artifacts that would pollute SDK/CLI/MCP
+  adoption. If the information matters, route it to the cdx repo/agent or the
+  owning internal repo instead of committing it here.
 - Keep tool names and relay types canonical.
 - Use `venue` as the schema field name (never `protocolId`).
 
@@ -95,3 +140,32 @@ It must only call `gateway.toreva.com` via relay protocol over HTTPS.
 Use this exact sentence in public-facing docs and skill files:
 
 "Non-custodial execution primitives for Solana. Best-execution routing across Jupiter Perps, Pacifica, Drift, and Flash Trade. 1 bps to open. Everything else is free."
+
+
+<!-- TOREVA-MEMORY-PROTOCOL -->
+# Session protocol
+
+## Start of every session
+
+1. Read [`MEMORY.md`](./MEMORY.md) — this repo's curated Layer-2 intelligence. Apply active lessons, open questions, and recent decisions before acting.
+2. Read [`REPO_CHARTER.md`](./REPO_CHARTER.md) and [`AGENT_CHARTER.md`](./AGENT_CHARTER.md) if the task touches scope boundaries.
+3. Read [`KPIs.md`](./KPIs.md) if the task will move a measured outcome.
+
+## End of every session
+
+1. Distil **learnings** (not tasks done) from this session.
+2. Apply the five-gate quality filter in `MEMORY.md`.
+3. Append qualifying entries to `MEMORY.md` using its YAML template.
+4. Tag each entry with relevant `OBJ-XX` IDs from `coordinator/bus/registries/system-objectives.v1.json`.
+5. Mark cross-repo-relevant entries with `promote: candidate` so the memory agent can pick them up for Layer-3 curation.
+
+## What goes where
+
+- **Instructions, conventions, rules** → this file (`CLAUDE.md` / `AGENTS.md`)
+- **Curated intelligence from sessions** → `MEMORY.md`
+- **Material decisions** → `docs/decisions/DEC-*.md`
+- **KPI ownership, thresholds, interventions** → `coo/data/metrics/`
+- **Cross-repo memory objects** → `memory/objects/`
+
+Full Layer-2 spec: [`memory/playbooks/layer-2-memory-file.md`](../memory/playbooks/layer-2-memory-file.md)
+<!-- /TOREVA-MEMORY-PROTOCOL -->

KPIs.md

@@ -0,0 +1,83 @@
+# KPIs — kit
+
+Canonical measurement definitions owned by **analytics**. Thresholds, ownership, and interventions owned by **coo**. This repo authors and exposes the signal.
+
+The 20% of KPIs that drive 80% of this repo's charter outcomes and time-to-revenue.
+
+## Pareto KPIs
+
+### kpi.kit.build-and-test-pass-rate
+
+- **Measures:** Percentage of CI runs (main branch) where all packages (@toreva/sdk, @toreva/cli, @toreva/mcp, @toreva/types) build and all Vitest tests pass
+- **Charter outcome:** "All packages build and pass Vitest tests"
+- **Time-to-revenue link:** A red main blocks any third-party developer or AI-agent integration release; kit is the acquisition surface for the AI-agent distribution channel
+- **Type:** leading
+- **Target:** `>= 99%`
+- **Cadence:** weekly
+- **Source:** CI pipeline logs
+- **Owner:** kit agent (accountable) — analytics agent (measurement) — coo agent (threshold)
+- **Intervention trigger:** any red main unresolved more than 4 hours
+
+### kpi.kit.gateway-contract-parity
+
+- **Measures:** Percentage of gateway tool names, relay types, and schema fields that match canonical kit type definitions (@toreva/types)
+- **Charter outcome:** "Tool names and relay types are canonical (match gateway contract)" and "never use `protocolId` — always use `venue`"
+- **Time-to-revenue link:** Contract drift causes silent integration breakage for third-party agents and developers — every broken integration loses an inbound user and erodes trust
+- **Type:** leading
+- **Target:** `100%`
+- **Cadence:** weekly
+- **Source:** schema diff between gateway contract and @toreva/types
+- **Owner:** kit agent (accountable) — analytics agent (measurement) — coo agent (threshold)
+- **Intervention trigger:** any drift detected, or any use of forbidden field `protocolId`
+
+### kpi.kit.gateway-only-communication-compliance
+
+- **Measures:** Count of detected network calls from any kit package to endpoints other than `gateway.toreva.com`
+- **Charter outcome:** "SDK, CLI, and MCP server all communicate exclusively through gateway.toreva.com" and "Zero secrets or internal URLs in published code"
+- **Time-to-revenue link:** Non-gateway calls leak internal surface area and create security/compliance risks that directly threaten licensing and third-party trust
+- **Type:** leading
+- **Target:** `0`
+- **Cadence:** weekly
+- **Source:** static analysis of published packages + network audit in tests
+- **Owner:** kit agent (accountable) — analytics agent (measurement) — coo agent (threshold)
+- **Intervention trigger:** any occurrence
+
+### kpi.kit.mcp-server-smoke-test-success
+
+- **Measures:** Percentage of scheduled MCP server smoke tests (stdio and remote modes) that complete all declared skill invocations successfully
+- **Charter outcome:** "MCP server works in both stdio and remote modes"
+- **Time-to-revenue link:** MCP is the primary AI-agent acquisition surface (Claude, Codex, Grok); smoke test failures silently cut off inbound AI-agent developer users
+- **Type:** leading
+- **Target:** `>= 99%`
+- **Cadence:** weekly
+- **Source:** MCP smoke test job logs
+- **Owner:** kit agent (accountable) — analytics agent (measurement) — coo agent (threshold)
+- **Intervention trigger:** any failed skill invocation unresolved within 12 hours
+
+### kpi.kit.disclaimer-and-notice-presence
+
+- **Measures:** Percentage of published packages where required regulatory notices and disclaimers are present and unmodified
+- **Charter outcome:** "Include proper regulatory notices and disclaimers in public-facing packages" and "Never modify or remove regulatory notices"
+- **Time-to-revenue link:** Missing disclaimers on a public-facing package in the AUSTRAC/ASIC window is a direct compliance event that can block the license path
+- **Type:** leading
+- **Target:** `100%`
+- **Cadence:** weekly
+- **Source:** published package content audit vs canonical disclaimer templates
+- **Owner:** kit agent (accountable) — analytics agent (measurement) — coo agent (threshold)
+- **Intervention trigger:** any package missing or modified disclaimer
+
+### kpi.kit.business-logic-leakage-count
+
+- **Measures:** Count of detected business logic primitives (routing, scoring, fee enforcement, strategy execution) in any kit package after static analysis
+- **Charter outcome:** "Zero business logic in any package (routing, scoring, fee enforcement)"
+- **Time-to-revenue link:** Business logic in the thin client violates the gateway-first architecture, creates divergence risk between kit and gateway, and puts unaudited code in front of regulators
+- **Type:** leading
+- **Target:** `0`
+- **Cadence:** weekly
+- **Source:** static analysis linter over kit packages
+- **Owner:** kit agent (accountable) — analytics agent (measurement) — coo agent (threshold)
+- **Intervention trigger:** any occurrence
+
+## Registry
+
+Canonical index: `/analytics/kpis/registry/README.md`

MEMORY.md

@@ -0,0 +1,76 @@
+# MEMORY — kit
+
+**Layer-2 curated intelligence.** Read this at the START of every session. Append curated entries at the END of every session.
+
+This file is distinct from:
+- `CLAUDE.md` / `AGENTS.md` — instructions, conventions, rules
+- Tool-specific auto-memory — Claude Code `.memory/`, Codex `~/.codex/memories/`, Cursor DB, Copilot workspace
+- Org-wide memory — `/memory/objects/` (promoted from here)
+
+Full spec: [`/memory/playbooks/layer-2-memory-file.md`](../memory/playbooks/layer-2-memory-file.md)
+
+## Session-start protocol
+
+1. Read **Active lessons** and **Open questions** below — apply them before acting.
+2. Check **Recent decisions** for anything that supersedes your current direction.
+3. At session end, distil learnings (not tasks done) and append qualifying entries to the appropriate section using the YAML template.
+
+## Quality gate (all five must hold)
+
+1. **Future-relevant** — useful beyond the current task
+2. **Non-obvious** — not derivable from code, charters, or docs
+3. **Actionable** — shapes a future decision or approach
+4. **Traceable** — has a source (session ref, PR, doc)
+5. **Durable** — half-life ≥ 7 days
+
+Reject conversation context, one-off task state, things already documented elsewhere, PII/credentials, unreleased commercial strategy, opinions without evidence.
+
+## Entry template
+
+```yaml
+- id: mem.kit.yyyymmdd.slug
+  title: short title
+  type: decision | pattern | lesson | constraint | assumption | anti-pattern
+  captured: YYYY-MM-DD
+  source: claude-code | codex | cursor | copilot | human
+  session_ref: optional path/hash
+  objectives: [OBJ-08, OBJ-14]
+  summary: high-signal statement (not raw transcript)
+  applies_when: trigger context cue for future sessions
+  evidence: anecdotal | repeated | measured
+  promote: local | candidate
+```
+
+Objective IDs: see `coordinator/bus/registries/system-objectives.v1.json` (OBJ-01..OBJ-20).
+
+---
+
+## Active lessons
+
+Curated Pareto entries — keep top ~20 by utility. Overflow migrates to **Superseded / retired**.
+
+_No entries yet — file initialised 2026-04-13._
+
+---
+
+## Open questions
+
+Live unknowns that should inform the next session's direction.
+
+_No entries yet._
+
+---
+
+## Recent decisions
+
+Decisions with rationale. Material decisions also emit `docs/decisions/DEC-*.md`.
+
+_No entries yet._
+
+---
+
+## Superseded / retired
+
+Entries moved out of Active — kept for history.
+
+_No entries yet._

REPO_CHARTER.md

@@ -17,6 +17,11 @@ Out of scope:
 - Business logic (routing, scoring, fee enforcement, strategy execution)
 - Backend server code or HTTP serving frameworks
 - Secrets, keypairs, or internal-only URLs
+- Internal facts or docs: cross-repo/agent topology, dispatches, operating
+  procedures, ownership maps, source-of-truth pointers, unreleased product/GTM
+  plans, service choreography, liveness notes, postmortems, local transcripts,
+  and coordination artifacts. Important internal material belongs in cdx or the
+  owning internal repo, not this public thin-client repo.
 - Direct blockchain interaction (all execution goes through gateway)
 - Internal platform services or infrastructure
 - Venue intelligence gathering or market data production
@@ -59,5 +64,6 @@ Out of scope:
 - SDK, CLI, and MCP server all communicate exclusively through gateway.toreva.com
 - Zero business logic in any package (routing, scoring, fee enforcement)
 - Zero secrets or internal URLs in published code
+- Zero internal facts/docs or coordination artifacts in committed content
 - Tool names and relay types are canonical (match gateway contract)
 - MCP server works in both stdio and remote modes

plans/README.md

@@ -0,0 +1,11 @@
+# kit — plans
+
+Quarterly → 6 × 2-week sprints → daily DoDs. Rolls forward.
+50/50 self-vs-system split. See [../coo/docs/doctrine/focus-split-50-50.md](../../coo/docs/doctrine/focus-split-50-50.md).
+
+- `quarterly/Q<X>-<YYYY>.md` — quarterly plan
+- `sprints/Q<X>-<YYYY>-S<N>.md` — sprint plan + retro
+- `daily/YYYY-MM-DD.md` — daily DoD + loop log
+- `masterplan.md` — optional multi-quarter capability upgrade
+
+Templates: `memory/playbooks/plan-templates/`.

plans/daily/2026-04-13.md

@@ -0,0 +1,42 @@
+# kit — 2026-04-13
+
+Sprint: Q2-2026 S1 · Loop day: 1
+
+## Orient (start of loop)
+- Read: MEMORY.md active lessons, open bus envelopes for me, today's row in sprint plan.
+- Inherited blockers: ...
+
+## Predict
+- What I expect to happen in my domain today: ...
+- Confidence: low / med / high
+- (Logged to PHI-P-AGENT as prediction-ref: `2026-04-13-kit-01`)
+
+## Definition of done today
+Every DoD item must be measurable by end of day.
+
+- [ ] DOD-1: ... (bucket: self / system)
+- [ ] DOD-2: ...
+- [ ] DOD-3: ...
+
+## Act — bus envelopes I publish today
+- ...
+
+## Measure — actual vs. prediction
+_Filled at end of day._
+- Actual: ...
+- Delta to prediction: ...
+- Which error class: model / data / execution / external
+
+## Learn — MEMORY candidates
+_Distilled at end of day. 5-gate filter applies._
+- ...
+
+## Improve — next-best improvement identified
+_What's the compounding move queued for tomorrow or next sprint?_
+- ...
+
+## Human check-in required?
+- [ ] Yes → surface with context block below
+- [x] No
+
+_Only surface if Class A, Day-1 downtime risk, or uncertainty is high enough that continuing risks wasted effort._