Optimize agent state path and worker timing

.gitignore

@@ -1,5 +1,6 @@
 .DS_Store
 
+data/
 data/store.e2e.json
 data/store.json
 data/store*.sqlite

.gitmodules

@@ -0,0 +1,8 @@
+[submodule "vendors/openclaw-lite-main/vendor/openclaw-main"]
+	path = vendors/openclaw-lite-main/vendor/openclaw-main
+	url = https://github.com/openclaw/openclaw.git
+	branch = main
+
+[submodule "vendors/agent0-ts"]
+	path = vendors/agent0-ts
+	url = https://github.com/Agent-Town/agent0-ts.git

AGENTS.md

@@ -8,6 +8,7 @@ This repo is a **minimal** Agent Town landing page.
 2. **Human + agent co-op** — the unlock flow requires both participants.
 3. **Session-token identity** — do not add external identity providers.
 4. **Deterministic testability** — every milestone must be verifiable with Playwright.
+5. **Wallet-first identity** — each user is represented by their connected wallet; wallet continuity drives session continuity.
 
 ## Non-goals / constraints
 
@@ -110,6 +111,7 @@ When changing worker behavior required by skill files (`skill.md` / `SKILL.md`):
 
 ### 6) Session and identity guardrails
 
+- The user identity is the connected wallet (or wallets), not a transient browser credential.
 - Team Code is a session token/routing token and should stay hidden from cluttered UX surfaces.
 - Team/session identity should be stable across polling/refresh for a live session; avoid regressions that rotate it unexpectedly.
 

README.md

@@ -94,6 +94,31 @@ npm run setup:sepolia-wallet -- --no-faucet
 - Co-op actions: `/api/agent/canvas/paint`, `/api/agent/house/*`.
 - House API auth and ceremony details are documented in `specs/02_api_contract.md`.
 
+## Trainer namespace toggle
+You can enable/disable `trainer.*` tools with either URL params or localStorage.
+
+- URL query param:
+  - `?trainerNamespace=1` enables
+  - `?trainerNamespace=0` disables
+  - aliases also supported: `trainer_namespace`, `trainer-tools`, `trainerTools`
+- localStorage override key: `agentTown:feature:trainerNamespace`
+  - set `"true"` to enable
+  - set `"false"` to disable
+  - remove the key to clear the override
+
+Browser console examples:
+
+```js
+localStorage.setItem('agentTown:feature:trainerNamespace', 'true');  // enable
+localStorage.setItem('agentTown:feature:trainerNamespace', 'false'); // disable
+localStorage.removeItem('agentTown:feature:trainerNamespace');       // clear override
+```
+
+Resolution precedence:
+- localStorage override
+- URL query param
+- default behavior
+
 ## Key routes
 - `/` — onboarding, Team Code, token check, reconnect.
 - `/start` — start page (logo/video/welcome + Enter -> Privy login).
@@ -148,3 +173,4 @@ Unlocking a house in the UI is gated by a Privy-backed Solana wallet signature.
 - API contract: `specs/02_api_contract.md`
 - Experience flow: `specs/01_experience_flow.md`
 - TDD milestones: `specs/04_tdd_milestones.md`
+- District map + storefront: `specs/11_district_map_storefront_spec.md`

docs/getting-started.md

@@ -65,6 +65,23 @@ Mind settings are included in agent state backup/restore in the house flow:
 - If you switched providers, verify `Model ID` updated to a model supported by that provider.
 - `test-local` is for deterministic tests and not real model inference.
 
+### Trainer namespace toggle
+
+Use either query params or localStorage:
+
+- URL: `?trainerNamespace=1` (enable) or `?trainerNamespace=0` (disable)
+- localStorage key: `agentTown:feature:trainerNamespace` with value `"true"` or `"false"`
+
+Console helper:
+
+```js
+localStorage.setItem('agentTown:feature:trainerNamespace', 'true');
+localStorage.setItem('agentTown:feature:trainerNamespace', 'false');
+localStorage.removeItem('agentTown:feature:trainerNamespace');
+```
+
+Precedence: localStorage override, then URL param, then default.
+
 ## Next
 
 - Decision guide: [Which Provider Should I Pick?](/docs/which-provider.md)

docs/internal-skill-testline.md

@@ -2,7 +2,7 @@
 
 Status: Active  
 Audience: Engineering only
-Last updated: 2026-02-18
+Last updated: 2026-02-22
 
 ## Goal
 
@@ -77,9 +77,28 @@ Keep `skill.md` evolution testable as we:
 | Session restore path works when cookie is missing but `x-team-code-hint` is available | `public/app.js`, `public/house.js`, `server/index.js` | client auto-sends team-code hint header and server rebinds same team session identity | `e2e/57_phase3_onboarding_wallet_llm_persist.spec.js` (`state endpoint restores session via team code hint when cookie is missing`) |
 | Truthful local “agent active” readiness | `public/app.js` + worker `skillState` bridge | OpenClaw Lite suppresses “ready” when skill import state is failed; `/skill.md` auto-import is attempted after local connect and failures are surfaced | `e2e/57_phase3_onboarding_wallet_llm_persist.spec.js` (`agent readiness status tracks skill import failure and recovery`) |
 | House-page OpenClaw export compatibility | `public/house.js` | download delegates to local worker `gateway.command.exportZip` (`openclaw-lite-export.zip`); upload accepts OpenClaw export and legacy house backup zip formats | `e2e/54_agent_state_backup_restore.spec.js` (`house backup stores encrypted state and supports ZIP download/upload restore`) |
+| Trainer namespace discovery + feature flag gating | `public/trainer_namespace_plugin.js`, `public/trainer.js`, `public/app.js`, `server/index.js` | plugin-constrained `trainer.*` registry is visible only when `featureFlags.trainerNamespace` resolves true | `e2e/98_trainer_namespace_contract_harness.spec.js` (`trainer namespace tools are discoverable when enabled and hidden when disabled`) |
+| Trainer namespace read-only introspection tools | `public/trainer_namespace_plugin.js`, `public/trainer.js` | `trainer.list_runs`, `trainer.get_run`, `trainer.get_event`, `trainer.get_session_context` return deterministic payloads | `e2e/99_trainer_namespace_read_tools.spec.js` (`trainer namespace read tools return deterministic run, event, and session context payloads`) |
+| Trainer namespace dynamic action catalog bridge | `public/trainer_namespace_plugin.js`, `public/trainer.js`, `public/skill_actions_plugin.js` | `trainer.list_actions` reflects active skill action extraction and switches atomically across skill changes | `e2e/100_trainer_namespace_action_catalog.spec.js` (`trainer.list_actions reflects active skill and updates after skill switch`) |
+| Trainer namespace action invocation bridge | `public/trainer_namespace_plugin.js`, `public/trainer.js`, `public/skill_actions_plugin.js` | `trainer.invoke_action` executes skill action requests with provided params and deterministic failure codes | `e2e/101_trainer_namespace_invoke_action.spec.js` (`trainer.invoke_action validates inputs and executes action requests with provided params`) |
+| Trainer namespace evidence freshness + expiry loop | `public/trainer_namespace_plugin.js`, `public/trainer.js` | `trainer.list_evidence` supports deterministic freshness filtering and expiry-window evaluation | `e2e/102_trainer_namespace_evidence_loop.spec.js` (`trainer.list_evidence supports deterministic freshness and expiry windows after invoke_action`) |
+| Trainer namespace transcript integrity + not-used diagnostics | `public/trainer_namespace_plugin.js`, `public/trainer.js`, `public/app.js` | `trainer.get_transcript_integrity` and `trainer.explain_not_used` surface reason codes and align with Session Context diagnostics | `e2e/103_trainer_namespace_diagnostics.spec.js` (`trainer diagnostics tools expose transcript integrity + not-used reasons and align with Session Context diagnostics`) |
+| Trainer namespace approval-gated destructive tools | `public/trainer_namespace_plugin.js`, `public/trainer.js` | `trainer.delete_trace` / `trainer.clear_traces` require one-time approval tokens with TTL and deterministic failure paths | `e2e/104_trainer_namespace_approval_gate.spec.js` (`trainer destructive tools require approval token, allow one operation, and expire deterministically`) |
+| Trainer namespace budgets + recursion guards | `public/trainer_namespace_plugin.js` | per-turn/per-window budgets and recursion blocking enforce deterministic `TRAINER_RATE_LIMITED` and `TRAINER_RECURSION_BLOCKED` outcomes | `e2e/105_trainer_namespace_rate_limit_recursion.spec.js` (`trainer namespace enforces rate limits and blocks recursive dispatch attempts deterministically`) |
+| Trainer namespace redaction in diagnostics and debug panes | `public/trainer_namespace_plugin.js`, `public/app.js` | secret-like values are masked in trainer namespace outputs/audit snapshots and agent debug surfaces | `e2e/106_trainer_namespace_redaction.spec.js` (`trainer namespace redacts secret-like values from diagnostics and avoids leaking raw secrets into debug panes`) |
+| Trainer namespace human-agent coop verification loop | `public/trainer_namespace_plugin.js`, `public/trainer.js` | builder demonstration + repeat invocation + evidence-backed verification flow remains deterministic in trainer tooling | `e2e/107_trainer_namespace_coop_canvas.spec.js` (`trainer namespace supports a deterministic human-agent coop loop for canvas verification`) |
 
 ## Progress Log
 
+### 2026-02-22
+
+- Added plugin-constrained `trainer.*` namespace coverage for discovery, read tools, dynamic action catalogs, and action invocation (`e2e/98` to `e2e/101`).
+- Added deterministic trainer namespace evidence lifecycle coverage (`freshOnly` and expiry windows) via `e2e/102_trainer_namespace_evidence_loop.spec.js`.
+- Added transcript-integrity and not-used diagnostics parity coverage between trainer tools and Session Context tab (`e2e/103_trainer_namespace_diagnostics.spec.js`).
+- Added approval-gated destructive trainer tool coverage with one-time token + TTL semantics (`e2e/104_trainer_namespace_approval_gate.spec.js`).
+- Added trainer namespace policy guard coverage for per-turn/window rate limits and recursion blocks (`e2e/105_trainer_namespace_rate_limit_recursion.spec.js`).
+- Added redaction coverage for trainer diagnostics/debug panes and cooperative canvas verification loop coverage (`e2e/106_trainer_namespace_redaction.spec.js`, `e2e/107_trainer_namespace_coop_canvas.spec.js`).
+
 ### 2026-02-18
 
 - Added deterministic PKCE OAuth contract coverage for OpenAI Codex (`start`/`status`/`exchange`) and state-rebind recovery when `attemptId` is stale.

docs/rate-limits.md

@@ -0,0 +1,87 @@
+# Rate Limits (Current State)
+
+## Summary
+
+- `/api/state` is **not** rate-limited by middleware.
+- `/api/state` no longer performs `readStore()` and no longer returns signup/public-team stats.
+- Most explicit request throttling is in `server/index.js` middleware and Pony inbox flow.
+- There is also a worker-side tool limiter for `http_request` fallback proxy calls.
+
+## `/api/state` hot path (current)
+
+Route:
+- `/Users/robin/.codex/worktrees/d83e/Portal/server/index.js` (`app.get('/api/state', ...)`)
+
+Current behavior:
+- Uses only in-memory session state (`ensureHumanSession`, lite/onboarding/ceremony/experience snapshots).
+- Does not read global store tables per request.
+- Does not compute `stats.signups` or `stats.publicTeams`.
+
+Implication:
+- `/api/state` latency should now reflect normal request handling, not full-store JSON deserialization.
+
+Note:
+- `/api/session` and `/api/session/reset` still read store stats today and are separate from `/api/state`.
+
+## Active server rate limits
+
+Middleware implementation:
+- `/Users/robin/.codex/worktrees/d83e/Portal/server/index.js` (`rateLimit({ windowMs, max, keyFn })`)
+
+Headers when active:
+- `X-RateLimit-Limit`
+- `X-RateLimit-Remaining`
+- `X-RateLimit-Reset`
+- On reject: `Retry-After`, response `429 { ok: false, error: 'RATE_LIMITED' }`
+
+Configured middleware limits:
+
+| Scope | Window | Max | Key |
+|---|---:|---:|---|
+| `/api/agent` | 60s | 1200 | `agent:${req.ip}` |
+| `/api/house` | 60s | 180 | `house:${req.ip}` |
+| `/api/share/create` | 60s | 60 | `share:${req.ip}` |
+| `/api/token` | 60s | 30 | `token:${req.ip}` |
+| `/api/wallet` | 60s | 30 | `wallet:${req.ip}` |
+| `/api/house/init` | 60s | 20 | `house-init:${req.ip}` |
+
+## Pony message rate limit
+
+Implementation:
+- `/Users/robin/.codex/worktrees/d83e/Portal/server/index.js`
+- `PONY_RATE_WINDOW_MS = 60_000`
+- `PONY_RATE_MAX_PER_PAIR = 20`
+- `checkPonyRateLimit({ senderKey, toHouseId })`
+
+Applied in:
+- `/api/pony/send`
+
+Behavior:
+- Keyed by sender-target pair: `${senderKey}->${toHouseId}`
+- On reject: `429 { ok: false, error: 'RATE_LIMITED_PONY', retryAfter }`
+
+## Worker/runtime-side rate limit (OpenClaw)
+
+Implementation:
+- `/Users/robin/.codex/worktrees/d83e/Portal/vendors/openclaw-lite-main/src/openclaw-lite/worker.js`
+- `HTTP_RATE_LIMIT_WINDOW_MS = 1000`
+- `HTTP_RATE_LIMIT_MAX = 50`
+- `consumeHttpRateLimit(url)`
+
+Behavior:
+- Applies to `http_request`/`web_fetch` proxy fallback path by origin.
+- Effective threshold: up to 50 fallback requests per origin per 1-second window.
+- Same-origin requests that succeed directly do not hit this limiter.
+- Returns tool error code `RATE_LIMIT` with `retryAfterMs`.
+
+Synced runtime artifact:
+- `/Users/robin/.codex/worktrees/d83e/Portal/public/openclaw-lite/worker.js`
+
+## Endpoints not currently middleware-rate-limited
+
+Examples:
+- `/api/state`
+- `/api/session`
+- `/api/atlas/*`
+
+These may still be slow due to CPU/IO hot paths (not throttling).

e2e/01_home.spec.js

@@ -1,4 +1,5 @@
 const { test, expect } = require('@playwright/test');
+const { enterHatch } = require('./helpers/phase2');
 
 const resetToken = process.env.TEST_RESET_TOKEN || 'test-reset';
 
@@ -7,9 +8,7 @@ test.beforeEach(async ({ request }) => {
 });
 
 test('home loads, shows team code and skill link', async ({ page, request }) => {
-  await page.goto('/');
-  await page.getByTestId('auth-signin').click();
-  await expect(page.getByTestId('hatch-panel')).toBeVisible();
+  await enterHatch(page, 'signin');
 
   await expect(page.getByTestId('team-code')).toHaveText(/TEAM-[A-Z0-9]{4}-[A-Z0-9]{4}/);
   await expect(page.getByTestId('skill-link')).toBeVisible();

e2e/03_create_share_leaderboard.spec.js

@@ -12,23 +12,6 @@ function ssePayload(chunks) {
   return chunks.map((chunk) => `data: ${JSON.stringify(chunk)}\n\n`).join('') + 'data: [DONE]\n\n';
 }
 
-test('co-op open -> co-create -> generate house -> unlock with wallet signature', async ({ page, request }) => {
-  // Mock a Solana wallet (Phantom-style) for Playwright.
-  await page.addInitScript(() => {
-    // Minimal mock matching usage in create.js/house.js
-    const sig = new Uint8Array(64);
-    // Deterministic but non-zero
-    for (let i = 0; i < sig.length; i++) sig[i] = (i * 13) & 0xff;
-    window.__PRIVY_WALLET_BRIDGE__ = {
-      connectSolana: async () => ({ address: 'So1anaMock111111111111111111111111111111111' }),
-      disconnectSolana: async () => {},
-      signSolanaMessage: async () => ({ signature: sig, publicKey: { toString: () => 'So1anaMock111111111111111111111111111111111' } })
-    };
-  });
-
-  await page.goto('/');
-  const teamCode = (await page.getByTestId('team-code').innerText()).trim();
-
 function makeToolChunks({ id, model, toolName, args = {}, callId }) {
   const created = Math.floor(Date.now() / 1000);
   return [

e2e/05_erc8004_mint.spec.js

@@ -8,7 +8,7 @@ test.beforeEach(async ({ request }) => {
   await request.post('/__test__/reset', { headers: { 'x-test-reset': resetToken } });
 });
 
-test('ERC-8004 UI stays hidden on house page', async ({ page, request }) => {
+test('ERC-8004 UI stays hidden on house page', async ({ page }) => {
   // Mock Solana wallet + EVM wallet + ag0 SDK
   await page.addInitScript(() => {
     // Solana mock
@@ -47,28 +47,6 @@ test('ERC-8004 UI stays hidden on house page', async ({ page, request }) => {
   });
 
   await page.goto('/');
-  const teamCode = (await page.getByTestId('team-code').innerText()).trim();
-
-  // Connect agent
-  await request.post('/api/agent/connect', { data: { teamCode, agentName: 'ClawTest' } });
-
-  // Match
-  await page.getByTestId('sigil-key').click();
-  await request.post('/api/agent/select', { data: { teamCode, elementId: 'key' } });
-
-  // Press open
-  await page.getByTestId('open-btn').click();
-  await request.post('/api/agent/open/press', { data: { teamCode } });
-  await page.waitForURL('**/create');
-
-  // Agent ceremony
-  // Use randomness to avoid deterministic houseId collisions when tests run in parallel workers.
-  const ra = crypto.randomBytes(32);
-  const agentRevealPair = makeCeremonyRevealPair();
-  const raCommit = crypto.createHash('sha256').update(ra).digest('base64');
-  const commitResp = await request.post('/api/agent/house/commit', {
-    data: { teamCode, commit: raCommit, revealPub: agentRevealPair.publicKeyB64 }
-  });
   await reachCreateViaLite(page);
 
   await page.getByTestId('px-0-0').click();

e2e/06_agent_house_append.spec.js

@@ -8,34 +8,7 @@ test.beforeEach(async ({ request }) => {
   await request.post('/__test__/reset', { headers: { 'x-test-reset': resetToken } });
 });
 
-function sha256(buf) {
-  return crypto.createHash('sha256').update(buf).digest();
-}
-
-function hkdf(ikm, info, len = 32) {
-  // Node >= 15
-  return crypto.hkdfSync('sha256', ikm, Buffer.alloc(0), Buffer.from(info, 'utf8'), len);
-}
-
-function houseAuthHeaders(houseId, method, path, body, key) {
-  const ts = String(Date.now());
-  const bodyHash = crypto.createHash('sha256').update(body || '').digest('base64');
-  const msg = `${houseId}.${ts}.${method}.${path}.${bodyHash}`;
-  const auth = crypto.createHmac('sha256', key).update(msg).digest('base64');
-  return { 'x-house-ts': ts, 'x-house-auth': auth };
-}
-
-function aesGcmEncrypt(key32, plaintext, aad) {
-  const iv = crypto.randomBytes(12);
-  const cipher = crypto.createCipheriv('aes-256-gcm', key32, iv);
-  if (aad) cipher.setAAD(aad);
-  const ct = Buffer.concat([cipher.update(plaintext), cipher.final()]);
-  const tag = cipher.getAuthTag();
-  // WebCrypto AES-GCM returns ciphertext||tag as one buffer; mirror that.
-  return { iv, ct: Buffer.concat([ct, tag]) };
-}
-
-test('agent derives ceremony key and appends; human can decrypt in house UI', async ({ page, request }) => {
+test('agent derives ceremony key and appends; human can decrypt in house UI', async ({ page }) => {
   // Mock Solana wallet for unlock UX.
   await page.addInitScript(() => {
     const sig = new Uint8Array(64);
@@ -48,28 +21,6 @@ test('agent derives ceremony key and appends; human can decrypt in house UI', as
   });
 
   await page.goto('/');
-  const teamCode = (await page.getByTestId('team-code').innerText()).trim();
-
-  // Connect agent
-  await request.post('/api/agent/connect', { data: { teamCode, agentName: 'ClawTest' } });
-
-  // Match
-  await page.getByTestId('sigil-key').click();
-  await request.post('/api/agent/select', { data: { teamCode, elementId: 'key' } });
-
-  // Press open
-  await page.getByTestId('open-btn').click();
-  await request.post('/api/agent/open/press', { data: { teamCode } });
-  await page.waitForURL('**/create');
-
-  // Agent ceremony
-  // Use randomness to avoid deterministic houseId collisions when tests run in parallel workers.
-  const ra = crypto.randomBytes(32);
-  const agentRevealPair = makeCeremonyRevealPair();
-  const raCommit = sha256(ra).toString('base64');
-  const commitResp = await request.post('/api/agent/house/commit', {
-    data: { teamCode, commit: raCommit, revealPub: agentRevealPair.publicKeyB64 }
-  });
   await reachCreateViaLite(page);
 
   await page.getByTestId('px-0-0').click();