Skip to content

Open-source platform for 3D AI agents. Turn text or a photo into a rigged, animated GLB avatar, give it an LLM brain, memory and a wallet, and embed it anywhere with one web component. Remote MCP server, x402 payments, Three.js, browser-native.

docs(finish): trim order 918 to the owner-gated arming and capital steps plus four unshipped social ideas

nirholasd28ecffb6f9b
diff --git a/prompts/finish/918-roadmap-pumpfun-trading-arena.md b/prompts/finish/918-roadmap-pumpfun-trading-arena.md
index 8dca21529..8e52f6c7b 100644
--- a/prompts/finish/918-roadmap-pumpfun-trading-arena.md
+++ b/prompts/finish/918-roadmap-pumpfun-trading-arena.md
@@ -1,459 +1,32 @@
-# The Pump.fun Trading Arena — Master Plan
-
-**Status:** Strategy / roadmap. Owner-facing. Last updated 2026-09-29.
-**Scope:** ONLY pump.fun trading, deploying, agent monetization, and copy-trading. Nothing else.
-
-> **Build status (verified 2026-09-03).** Phases 0 to 6 are shipped, on surfaces that
-> settled on different names than this plan proposed: the Arena front door is
-> [/play/arena](https://three.ws/play/arena) (the tournament surface took `/arena`), the proof page is
-> `/trader/:id`, strategy authoring is `/strategy-lab` plus `POST /api/sniper/compile`,
-> and the copy loop runs as two engines (custodial `/mirror`, non-custodial
-> `/dashboard/copy`) over `copy_subscriptions` / `copy_executions` rather than the
-> `agent_copy_followers` schema sketched in section 8. Prompts A, B, C, E and F all
-> have shipped equivalents (`workers/agent-sniper/llm-judge.js`,
-> `workers/agent-sniper/journal.js` plus `/clip-director`, `api/_lib/strategy-compiler.js`,
-> the pump launcher, and `api/_lib/meta-allocator.js`).
->
-> Section 4's anti-gaming layer was the last real gap and closed on 2026-09-03
-> (`api/_lib/copy-eligibility.js`): the copyable-status sybil bar (4.4), the follower
-> drawdown circuit breaker (4.5), and self-copy / self-follow exclusion (4.6).
-> Section 9's named blocker, AMM exits on graduated positions, is written,
-> committed and running in production (`workers/agent-sniper/amm-exit.js`,
-> imported by `executor.js`, `positions.js` and `graduation-ride.js`; the fleet
-> rolled onto HEAD on 2026-09-29, see the owner paragraph below).
->
-> **Prompt D, the adversarial Risk Officer (section 6), is now BUILT**
-> (`workers/agent-sniper/risk-officer.js`, migration
-> `20260904020000_sniper_risk_officer.sql`, tests `tests/sniper-risk-officer.test.js`).
-> It is the independent second opinion the executor's other gates (Mayhem
-> exclusion, the oracle gate, the token-intel gate, market-realness, the trade
-> firewall's buy/sell round-trip, budgets and concurrency) cannot be: it runs last
-> in `executeBuy`, sees the real price impact and the real firewall verdict plus
-> the judge's own thesis, and is told to assume the trade is bad until the facts
-> prove otherwise. It carries the `size_adjustment` this section specifies, which
-> may only ever shrink a trade.
->
-> **It ships DISARMED, and that resolves the gate rather than deferring it.** The
-> level defaults to `shadow` on every strategy: the review runs fire-and-forget,
-> records into `sniper_risk_reviews`, and changes nothing about what the live
-> fleet buys, so shipping it is not a spend decision. Shadow rows join to the
-> positions that actually opened, so their realized P&L is the evidence for or
-> against arming it (the query is in `workers/agent-sniper/README.md`).
->
-> **Owner: one owner-gated step remains, plus a capital top-up. Step (1), the
-> agent-sniper deploy, is DONE (2026-09-29, owner approval "finish them all").**
-> Measured, not assumed:
->
-> - **What the old image lacked.** Revision `agent-sniper-00033-rg6` ran image
-> `sha256:9946506e…`, Cloud Build `1deea978` from a working-tree upload of
-> 2026-08-09 09:47 UTC (the revision itself was rolled 2026-08-11). Since that
-> upload, 15 commits touched `workers/agent-sniper/` (9 files, +996/-66):
-> `risk-officer.js` is new, and the executor, config, journal, intel learner,
-> reconcile and radar changed. `amm-exit.js` was NOT missing, contrary to the
-> earlier wording here: it has been in the image since June, and only its SDK-2
-> pool fields (`quoteMint`, `isMayhemMode`, `creatorFeeBps`) were behind.
-> - **The first attempt was refused, and that exposed a real bug.** Revision
-> `agent-sniper-00034-7zp` crashed on boot with `ENOENT /app/data/plans.json`:
-> since 2026-09-25 `api/payments/_config.js` reads that file at load, the sniper
-> reaches it through the trade-fee module, and the Dockerfile never copied
-> `data/`. Cloud Run kept 100% on 00033, so the fleet never stopped. Fixed in
-> `b17456355` (the Dockerfile copies `data/plans.json`, and `npm run
-> deploy:sniper` now boots `index.js` from exactly the Dockerfile COPY set
-> before it spends a Cloud Build, which reproduces that crash in 10 seconds).
-> - **What runs now.** `agent-sniper-00035-krc`, image `sha256:5b115be9…` built
-> from commit `b17456355`, serving 100%, worker booted 07:07:13 UTC. The env is
-> byte-identical to the old revision (26 keys, `SNIPER_MODE=live`,
-> `SNIPER_RISK_OFFICER` unset so the default is `shadow`), and the old instance
-> drained cleanly ("bye" at 07:07:21). All 31 sniper-related test files
-> (504 tests) passed at that commit first. No new ERROR-severity logs on 00035;
-> the only error-level line is the pre-existing "funding master cannot cover
-> refills".
-> - **Shadow rows exist from the fleet.** In the 32 minutes watched after the
-> boot the fleet opened one live position (07:14:35 UTC, `oracle_crossing`,
-> 0.002 SOL), and `sniper_risk_reviews` recorded its shadow review 1.6 seconds
-> later (severity `none`, 1,079 ms, answered by the platform chain's
-> `nvidia/nemotron-3-super-120b-a12b` rung), joined by `position_id`. That
-> position then exited through the new image at 07:30:01 UTC
-> (`liquidity_decay`, realized -0.000355 SOL). The table now holds 8 rows: the
-> 7 from the 2026-09-04 local verification plus the first fleet row.
->
-> **What remains, with owners.** (2) Arm the Risk Officer to `enforce` (gate 1:
-> it changes what real SOL buys), fleet-wide with `SNIPER_RISK_OFFICER=enforce`
-> via `--update-env-vars` (never `--set-env-vars`) or per strategy via
-> `risk_officer_level`, and only once the shadow-P&L query in
-> `workers/agent-sniper/README.md` has enough closed, joined rows to show the
-> `block` (and `caution`) severities losing more than `none`. One row is not evidence; at
-> the current rate that takes weeks. (3) An owner spend, not a code task: the
-> fleet is capital-starved. `/api/healthz` reads "7/11 armed wallets can still
-> trade (4 starved, 1 shrunk). Refilling them needs 0.2397 SOL. The funding
-> master holds only 0.0250 SOL", and the buy rate (26 opens in the 24 hours
-> before the deploy) and therefore the shadow-evidence rate scale with that
-> refill. Both are row 19 of
-> [production-100-OWNER-ACTIONS.md](_context/production-100-OWNER-ACTIONS.md).
-> Everything else in this plan is written, tested, deployed and applied to the
-> production schema.
->
-> One product gap in section 3.2 closed on 2026-09-24: "why it traded" was only
-> visible to the owner (the private `/api/sniper/journal`). Every public trade
-> now carries a receipt of the evidence its gates recorded before the entry
-> (`api/_lib/trade-receipt.js`, `GET /api/sniper/receipt`, the "why" drawer on
-> `/trader/:id`, the "Why it traded" section on `/trade/:id`, the `trade_receipt`
-> MCP tool; doc `docs/trade-receipts.md`).
->
-> **Prove the deploy state from the database, not only from `gcloud`.** A shadow
-> review is recorded fire-and-forget on EVERY buy (`reviewBuy` in `executor.js`,
-> `insert into sniper_risk_reviews` in `risk-officer.js`), so fleet buys with no
-> matching review rows mean the running image predates `risk-officer.js`. Before
-> 2026-09-29 that was 41 buys against 0 fleet rows; after it, every new
-> position carries a review row, which is how a future regression to an old
-> image shows up.
-**The only coin this platform promotes is `$THREE` (`FeMbDoX7R1Psc4GEcvJdsbNbZA3bfztcyDCatJVJpump`).** Pump.fun coins traded/launched through the platform are user runtime data, never endorsements.
-
----
-
-## 0. The one-sentence pitch
-
-> **three.ws is where AI agents trade pump.fun live, on-chain, with verifiable track records — and anyone can copy the winners with one tap while the agent's creator earns a cut.**
-
-It's a Twitch + a hedge fund + a 3D game, for memecoin trading. The traders are AI agents with faces (3D avatars), they trade real money, you watch them win or lose in real time, and you can ride along.
-
----
-
-## 1. Why we win (what already exists)
-
-Most "AI trading" products are vaporware: a prompt box and a fake equity curve. We have the opposite problem — we have the *engine* but haven't packaged it as a product anyone can feel.
-
-| Capability | Where it lives | State |
-|---|---|---|
-| Autonomous on-chain pump.fun execution from agent custodial wallets | `workers/agent-sniper/` | **Live (mainnet)** |
-| Strategy config (budget, concurrency, scoring, stop-loss/take-profit/trailing/timeout) | `agent_sniper_strategies`, `workers/agent-sniper/scorer.js` | **Live** |
-| Realized-P&L leaderboard | `api/sniper/leaderboard.js` | **Live** |
-| Live activity SSE stream | `api/sniper/stream.js` | **Live** |
-| Simulate → live mode (paper trade then graduate) | `workers/agent-sniper/config.js` (`mode`) | **Live** |
-| PumpPortal real-time feed + graduation detection | `api/_lib/pumpfun-ws-feed.js`, `services/pump-graduations/` | **Live** |
-| 3D avatar that reacts to the live feed | `pump-fun-skills/reactive/` | **Live** |
-| Buy/sell/launch (user-signed + agent-signed) | `api/pump/[action].js` | **Live** |
-| Anonymous paid launch (x402) | `api/x402/pump-launch.js` | **Live** |
-| Read-only intelligence MCP (25+ tools) | `packages/pumpfun-mcp/` | **Published** |
-| Agent reputation + on-chain attestations | `api/x402/agent-reputation.js`, `api/_lib/solana-attestations.js` | **Live** |
-| Monetization rail (x402, fee-sharing ledger, withdrawals, $THREE tiers, referrals) | `api/x402/*`, `2026-04-30-agent-monetization.sql`, `src/three-economy.js` | **Live** |
-| Per-agent custodial EVM+Solana wallets | `api/_lib/agent-wallet.js` | **Live** |
-
-**Conclusion:** the engine is real and trades real money today. The work is *productization, trust, fun, and the copy-trade money loop* — not core trading.
-
----
-
-## 2. The flywheel (the only diagram that matters)
-
-```
- ┌──────────────────────────────────────────────────────┐
- │ │
- ▼ │
- More AGENTS compete ──► Public ARENA (live P&L, 3D faces) │
- ▲ │ │
- │ ▼ │
- Creators promote Spectators COPY the winners (1 tap) │
- their agents │ │
- ▲ ▼ │
- │ Followers earn ──► Leaders earn fees ─────┘
- │ │
- └──── $THREE demand ◄──────┘ (fee discounts, holder rev-share, tiers)
-```
-
-Every arrow already has infrastructure behind it except **"Spectators copy the winners"** and **"Leaders earn fees."** That's the build.
-
----
-
-## 3. The product, surface by surface
-
-### 3.1 The Arena (`/arena`) — the front door
-The thing people screenshot. A live, public board of AI agents trading pump.fun **right now**.
-
-- **Grid of agent cards**, each with: 3D avatar (idle-animated, reacts to wins), live P&L today, open positions ticker, win rate, follower count, "COPY" button.
-- **Sort/filter**: 24h ROI, 7d ROI, win rate, followers, "hot streak", "new & rising", risk level.
-- **Live tape** down the side: real-time buys/sells from all agents (`api/sniper/stream.js` already streams this). Each event is a mini-card: avatar + "🟢 ApedFox bought 0.4 SOL of $XXXX — mcap $42k."
-- **Spectator energy**: the avatar that just took profit does a celebration emote (reactive skill already drives avatar gestures from the feed). Big wins trigger confetti + a toast everyone watching sees.
-- **Empty/loading/error**: skeleton cards; empty = "No agents live yet — be the first, deploy a trader in 60s →"; error = retry + status link.
-
-> **Outside-the-box:** make the Arena feel like a *spectator sport*. Round-based "seasons." A live "biggest win of the hour" banner. A "rekt cam" that (tastefully, opt-in) shows the biggest drawdown so it's not just survivorship theater — honesty builds more trust than a wall of green.
-
-### 3.2 The Agent Profile (`/arena/agent/:id`) — the proof page
-Trust is the entire game. This page exists to make "is this real?" answerable in 5 seconds.
-
-- **Equity curve** from `agent_sniper_positions` (real, on-chain, every trade links to Solscan).
-- **Stat block**: realized P&L, win rate, avg hold time, best/worst trade, max drawdown, Sharpe-ish consistency score, # closed positions, $ assets traded.
-- **Every trade is public + verifiable**: entry/exit, mcap at entry, exit reason (take-profit / stop-loss / trailing / timeout / graduation), tx links. No cherry-picking — the full ledger.
-- **"Why it traded"**: each trade has an AI-written one-line rationale (see Prompt B). Transparency is content.
-- **Track-record badges**: `Verified Profitable (30d)`, `100+ trades`, `Survived a -40% market`, `Paper→Live graduate`. On-chain-attested where possible (`solana_attestations`).
-- **COPY panel**: allocation input, risk cap, fee terms shown plainly, one tap to start.
-- **Followers tab**: who's copying, aggregate copied volume, leader's lifetime fee earnings.
-
-### 3.3 Deploy-a-Trader (`/arena/new`) — the on-ramp
-Turn a normal user into an agent owner in 60 seconds. No code.
-
-- **Plain-English strategy**: "Snipe fresh launches under $30k mcap with safe creators, take profit at 2x, stop at -30%, max 0.2 SOL per trade." → Strategy Author prompt (Prompt C) compiles it to an `agent_sniper_strategies` row, shown back as editable knobs.
-- **Pick a face**: choose/fork a 3D avatar (avatar fork schema exists).
-- **Starts in PAPER mode** automatically. Builds a real, timestamped track record with zero risk. This is the trust unlock *and* the cold-start solution.
-- **Graduate to LIVE**: once it clears a bar (e.g. 7 days + 20 trades + positive P&L), one click funds the agent wallet and flips `mode=live`. Now it's copyable and earns its owner fees.
-
-### 3.4 Copy-Trade (the money loop) — **the core new build**
-- Follower picks an allocation and a hard cap. Funds an **isolated copy-wallet** (agent-derived, withdrawable anytime — never a blind custodial deposit).
-- When the leader opens a position, the follower's wallet mirrors it **proportionally to allocation**, with the follower's own guardrails (max per trade, daily cap, global stop) layered on top — never blindly.
-- On a **profitable closed position**, a **performance fee** is taken with a **high-water mark** (you only pay on new profit, never twice). Split: **leader's owner gets the majority, platform takes a cut, $THREE holders get rev-share.**
-- All settled through the existing `agent_revenue_events` ledger + `agent_withdrawals`.
-
-**Fee model (proposed default — tune later):**
-- **Performance fee: 15%** of net new profit (high-water mark). → ~10% to leader owner, ~4% platform, ~1% to $THREE holder rev-share pool.
-- **No management fee.** Memecoin traders hate flat drag; charge only on wins. This is itself a marketing line: *"You only pay your agent when it makes you money."*
-- **$THREE holders get a discount** on the performance fee (e.g. 15% → 12% for top tier) — ties trading demand to the coin.
-- **Optional subscription tier** for premium agents (priority signal latency, private strategies) priced in $THREE via the existing tier system.
-
-### 3.5 Creator earnings dashboard (`/arena/earnings`)
-Owners see: followers, copied volume, fees earned (pending/withdrawable), per-agent breakdown, payout history. Withdraw to Solana/EVM. This page is what makes creators *promote their own agents for us* — the growth engine.
-
----
-
-## 4. Trust & anti-gaming (without this, nothing else matters)
-
-Copy-trading dies the moment people suspect fake track records. Defenses:
-
-1. **On-chain or it didn't happen.** Every stat derives from `agent_sniper_positions` / on-chain trades. Every number links to Solscan. No off-chain "reported returns."
-2. **Full ledger, no cherry-picking.** Profiles show *every* closed trade, including losers and max drawdown. We surface losses on purpose.
-3. **Paper trades are clearly labeled** and never mixed into live track record. The "graduate" badge marks the live transition date.
-4. **Sybil resistance on copyable status.** To become copyable: minimum live-trade count, minimum live age, minimum real capital deployed (so a "leader" can't fake-trade dust to farm a curve). Attestations recorded on-chain.
-5. **Drawdown circuit breakers for followers.** Copy stops automatically if the leader breaches a follower-set max drawdown — followers are never strapped to a runaway.
-6. **Wash-trade / self-follow detection.** A leader can't profit from copying themselves; flag clustered wallets.
-7. **Honest consistency score**, not just total P&L — rewards steady winners over one-lucky-100x, which is the failure mode of every copy-trade platform.
-8. **Withdraw-anytime copy wallets.** Followers keep custody control; reduces "rug the followers" risk to near zero.
-
----
-
-## 5. The "fun" / social layer (why they tell their friends)
-
-The engine is rational; the *growth* is emotional. Memecoin culture is loud, fast, and meme-driven. Lean in.
-
-- **3D avatars with personality.** The avatar is the trader's brand. It dances on a win, slumps on a loss (reactive skill already maps the feed to emotes). A profitable agent's avatar gets visibly *flashier* (cosmetics from the existing shop). Status is visual.
-- **Shareable trade cards.** Auto-generated image for every notable trade: avatar + "$XXXX +312% in 4 min 🚀" + verifiable link. Built for X. One tap to post. This is the organic acquisition loop.
-- **Seasons + leaderboards.** Weekly/monthly seasons, prizes (in $THREE / fee rebates), badges, streak counters. Resets keep newcomers competitive — no permanent incumbents.
-- **Hot-streak FOMO.** "🔥 ApedFox: 7 green trades in a row" surfaced on the Arena and pushed to followers.
-- **Live spectator chat** per agent during active trading. Spectators react; some convert to copiers in the moment of a big win.
-- **"Clone the champ."** Top agents are forkable as templates — clone the strategy, tweak it, paper-trade your variant, climb the board. Turns spectators into creators.
-- **Referral loop (exists).** Invite a friend; if their copied agent profits, both get fee rebates. Viral coefficient baked into the money.
-- **Agent vs agent.** Head-to-head matchups, bracket tournaments ("Pump Wars"), same starting capital, same window, public. Built-in narrative + content.
-
-> **Wildcard ideas worth prototyping:**
-> - **"Copy the consensus."** A meta-index that auto-allocates across the top N agents — an AI-managed pump.fun ETF. One tap, diversified, rebalanced. Most users want "just make me money," not picking a trader.
-> - **Live narration mode.** The agent's LLM rationale read aloud (Edge TTS already in the stack) over the Arena — sports-commentary energy for a memecoin trading stream.
-> - **"Backtest battle."** Submit a strategy, it forward-tests in paper against the live feed for 24h, ranked publicly. Low-stakes funnel into deploying.
-
----
-
-## 6. The prompts (the agent brains)
-
-These are real, shippable system prompts grounded in the actual strategy schema (`scoreMint`, market-cap bands, stop-loss/take-profit/trailing/timeout, budget/concurrency caps). They run through the free-first LLM policy (`llmComplete`, never a single provider).
-
-### Prompt A — Pump.fun Trading Agent (the decision brain)
-```
-You are {AGENT_NAME}, an autonomous pump.fun trader on Solana. You trade REAL money
-from your own wallet. Capital preservation outranks upside: a missed trade costs nothing,
-a bad trade costs real SOL.
-
-YOUR MANDATE (set by your owner, do not exceed):
-- Max per trade: {MAX_SOL} SOL
-- Daily budget: {DAILY_BUDGET} SOL | Spent today: {SPENT_TODAY}
-- Max concurrent open positions: {MAX_CONCURRENT} | Currently open: {OPEN_COUNT}
-- Risk profile: {RISK_PROFILE} (conservative | balanced | degen)
-- Stop-loss: {STOP_LOSS_PCT}% Take-profit: {TAKE_PROFIT_PCT}% Trailing: {TRAIL_PCT}%
- Max hold: {MAX_HOLD_MIN} min
-
-A NEW LAUNCH JUST APPEARED. Facts (all on-chain, do not invent any others):
-- Mint: {MINT} Name/symbol: {NAME}/{SYMBOL}
-- Market cap: ${MCAP} SOL in curve: {SOL_RESERVES} Age: {AGE_SECONDS}s
-- Creator: {CREATOR} | Creator history: {CREATOR_PRIOR_LAUNCHES} prior launches,
- {CREATOR_RUG_COUNT} rugged, {CREATOR_GRAD_COUNT} graduated
-- Holders: {HOLDER_COUNT} Top-10 concentration: {TOP10_PCT}%
-- Social signals: {SOCIAL_SUMMARY} | Buy quote (price impact at {MAX_SOL} SOL): {PRICE_IMPACT}%
-
-DECIDE. Output STRICT JSON only:
-{
- "action": "buy" | "skip",
- "size_sol": <= MAX_SOL and <= remaining daily budget; 0 if skip,
- "confidence": 0.0-1.0,
- "stop_loss_pct": within mandate,
- "take_profit_pct": within mandate,
- "reason": "one sentence, plain English, the single biggest driver of this call",
- "red_flags": ["..."] // empty if none
-}
-
-HARD RULES:
-- If creator has rugged before, or top-10 concentration > 50%, or price impact > {RISK_IMPACT_CAP}%,
- default to SKIP unless a strong, specific offsetting signal exists — name it in "reason".
-- Never exceed the mandate. If a buy would breach budget/concurrency, SKIP.
-- You cannot see the future. Do not claim a token "will" pump. Trade probabilities, not certainties.
-- If signals are thin or contradictory, SKIP. Skipping is a valid, frequent, correct outcome.
-```
-
-### Prompt B — Trade Rationale / Narrator (transparency + content)
-```
-You are the voice of {AGENT_NAME}. Explain ONE trade to a human spectator in ≤ 240 chars.
-Confident but never hype. No price predictions. No financial advice. Plain language a
-first-time trader understands.
-
-Trade: {ACTION} {SIZE_SOL} SOL of {SYMBOL} at ${ENTRY_MCAP} mcap.
-Drivers: {STRUCTURED_REASON}
-Outcome (if closed): {PNL_PCT}% via {EXIT_REASON}.
-
-Write the single line that would go on a shareable trade card. Make it true, make it human.
-```
-
-### Prompt C — Strategy Author (plain English → config)
-```
-A user described how they want their pump.fun trading agent to behave. Convert it into a
-strict strategy config. Do NOT invent risk they didn't ask for; when they're vague, choose
-SAFE defaults and list every assumption you made.
-
-User said: "{USER_TEXT}"
-
-Output STRICT JSON matching agent_sniper_strategies:
-{
- "max_sol_per_trade": number,
- "daily_budget_sol": number,
- "max_concurrent_positions": integer,
- "mcap_min_usd": number, "mcap_max_usd": number,
- "min_holders": integer, "max_top10_pct": number,
- "require_safe_creator": boolean,
- "stop_loss_pct": number, "take_profit_pct": number,
- "trailing_stop_pct": number|null, "max_hold_minutes": integer,
- "risk_profile": "conservative"|"balanced"|"degen",
- "assumptions": ["plain-English note for each default you chose"]
-}
-
-Guardrails: stop_loss_pct is MANDATORY (never null). daily_budget_sol >= max_sol_per_trade.
-If the user asks for something reckless (no stop-loss, all-in sizing), include it but add a
-loud assumption explaining the risk so the UI can warn them.
-```
-
-### Prompt D — Risk Officer (adversarial pre-trade check)
-```
-You are an independent risk reviewer. The trading agent wants to BUY. Your ONLY job is to
-catch what it missed. Assume the trade is bad until the facts prove otherwise.
-
-Proposed: BUY {SIZE_SOL} SOL of {SYMBOL}. Agent's reason: "{AGENT_REASON}".
-Facts: {SAME_ONCHAIN_FACTS_AS_PROMPT_A}
-Mandate remaining: budget {BUDGET_LEFT} SOL, open slots {SLOTS_LEFT}.
-
-Output STRICT JSON:
-{ "veto": true|false,
- "severity": "none"|"caution"|"block",
- "reasons": ["specific, fact-based; no generic 'crypto is risky'"],
- "size_adjustment": number|null // suggest a smaller size instead of a full veto
-}
-Veto (block) only for concrete, nameable danger: known-rug creator, extreme concentration,
-price impact that eats the edge, budget breach. Default to NOT vetoing on thin-but-clean
-setups — the agent already chose to skip the obvious junk.
-```
-
-### Prompt E — Launch / Deploy Agent (deploy a pump.fun coin)
-```
-You help a user deploy a pump.fun coin through three.ws. Be a careful operator, not a hype man.
-
-User intent: "{USER_TEXT}"
-Confirm and structure ONLY what they asked. Output STRICT JSON:
-{ "name": "", "symbol": "", "description": "",
- "initial_buy_sol": number, // 0 if they didn't ask to ape their own launch
- "image_action": "use_provided"|"generate"|"none",
- "vanity_suffix": string|null, // only if requested
- "front_run_protection": boolean,
- "missing": ["anything required but not provided — ask the user for these, do not guess"] }
-
-NEVER recommend or name any token other than what the user is launching. Do not promote the
-coin. Do not promise returns. If name/symbol/image is missing, put it in "missing" — do not
-fabricate branding the user didn't give you.
-```
-
-### Prompt F — Copy-Trade Concierge (helps a spectator choose)
-```
-A user wants to copy an AI trader but doesn't know which. Recommend from the REAL data below —
-never invent agents or stats. Match their risk appetite, not the flashiest number.
-
-User said: "{USER_TEXT}"
-Candidate agents (live, verifiable): {AGENTS_JSON} // P&L, win rate, max drawdown, trades, consistency, followers, risk_profile
-
-Output STRICT JSON:
-{ "recommendations": [
- { "agent_id": "", "why": "one honest sentence tied to their stated goal",
- "suggested_allocation_sol": number, "suggested_max_drawdown_pct": number } ],
- "caution": "one plain-English risk reminder — copying can lose money, past performance isn't future results" }
-Prefer consistent, lower-drawdown agents for cautious users; only surface high-variance agents
-to users who explicitly want degen risk. Always include the caution. Never guarantee profit.
-```
-
----
-
-## 7. Build order (phased, each phase ships something usable)
-
-**Phase 0 — Make the engine visible (1 surface).** Build `/arena`: render live agents from `api/sniper/leaderboard.js` + `api/sniper/stream.js` with 3D avatars + reactive emotes. No new backend. *This alone is demo-able and screenshot-worthy.* Add a changelog entry.
-
-**Phase 1 — Proof pages.** Agent profile (`/arena/agent/:id`) with the full on-chain ledger, equity curve, badges, trade rationales (Prompt B wired into the sniper's trade record). Trust layer first, before any money moves.
-
-**Phase 2 — Deploy-a-Trader, paper mode.** `/arena/new` + Strategy Author (Prompt C). New agents start in `mode=simulate`. Solves cold start: people build track records risk-free and the Arena fills up.
-
-**Phase 3 — Graduate to live.** Funding flow + `mode=live` flip + copyable eligibility bar + on-chain attestation of the graduation. Now there are real, verifiable, live track records.
-
-**Phase 4 — Copy-trade (the money loop).** New tables (`agent_copy_followers`, `agent_copy_trades`), mirror logic in the sniper worker (proportional sizing + follower guardrails + drawdown breaker), high-water-mark performance fee through `agent_revenue_events`, isolated withdraw-anytime copy wallets. Creator earnings dashboard.
-
-**Phase 5 — Social/viral layer.** Shareable trade cards, seasons + prizes, hot-streak push, spectator chat, "clone the champ," referral fee rebates.
-
-**Phase 6 — Meta products.** Copy-the-consensus index, agent-vs-agent tournaments, live narration mode.
-
-Each phase is independently launchable and each one is a changelog event $THREE holders see.
-
----
-
-## 8. New data model (minimal — extends what exists)
-
-```sql
--- Who copies whom, on what terms
-agent_copy_followers (
- id, leader_agent_id, follower_agent_id, follower_user_id,
- allocation_sol numeric, max_drawdown_pct numeric, max_per_trade_sol numeric,
- performance_fee_bps int, -- snapshot of fee at follow time
- high_water_mark_lamports bigint, -- for fair perf-fee accounting
- status text, -- active | paused | stopped
- copy_wallet_pubkey text, created_at, updated_at )
-
--- Each mirrored trade, linked back to the leader's position
-agent_copy_trades (
- id, leader_position_id, follower_agent_id, follower_position_id,
- signal_at, executed_at, size_sol numeric, realized_pnl_lamports bigint,
- perf_fee_lamports bigint, revenue_event_id )
-
--- Materialized leader stats per window (or compute at query time first)
-leader_performance_stats (
- agent_id, period, win_count, loss_count, avg_pnl_pct, max_drawdown_pct,
- consistency_score, total_followers, copied_volume_sol,
- fee_revenue_lamports, refreshed_at )
-```
-Everything else (positions, revenue events, withdrawals, wallets, reputation) already exists.
-
----
-
-## 9. Risks & how we hold the line
-
-- **Fake track records** → on-chain-only stats, full ledger, paper labeling, sybil bar for copyable status. (§4)
-- **Leader rugs followers** → isolated withdraw-anytime copy wallets + follower drawdown breaker; leader never custodies follower funds.
-- **Regulatory optics** → performance-fee-only, no "guaranteed returns" copy anywhere, mandatory risk caution in every copy flow (baked into Prompt F), "past performance ≠ future results" everywhere.
-- **RPC / fee spikes during volatility** → existing guardrails (price-impact breaker, idempotency lock, budget caps, global kill switch) extend to copy mirroring.
-- **Coin-policy compliance** → the Arena renders user-launched coins as runtime data only; $THREE is the only promoted coin, everywhere, always.
-- **AMM exits gap** (known) → finish graduated-position AMM exits before copy goes live, or followers get stuck in graduated positions.
-
----
-
-## 10. The north star
-
-A new user lands on `/arena`, sees a 3D fox avatar do a backflip because it just 5x'd a launch, watches three more agents trade live in the tape, taps COPY on the steadiest one, funds 0.3 SOL, and 90 seconds later owns a position an AI picked — while the agent's creator quietly earns a cut. They screenshot the trade card, post it, and their friend signs up.
-
-That loop, done so well people can't not talk about it, is the whole plan.
+# The Pump.fun Trading Arena: what remains
+
+**Trimmed 2026-10-23.** Phases 0 to 6 of the original master plan are shipped and verified
+(Arena `/play/arena`, proof page `/trader/:id`, `/strategy-lab`, custodial `/mirror` and
+non-custodial `/dashboard/copy`, anti-gaming in `api/_lib/copy-eligibility.js`, Risk Officer in
+`workers/agent-sniper/risk-officer.js`, public trade receipts, tournaments, meta-allocator).
+Re-verified this session: `npx vitest run` on the risk officer, copy eligibility, copy engine,
+copy earnings, tournament scoring and arena tournament API suites, 121 tests passing.
+Hot-streak badges on the Arena board shipped in `d06feb19b` (`api/_lib/win-streak.js`).
+The full original text is in git history (`git log -- prompts/finish/918-roadmap-pumpfun-trading-arena.md`).
+
+**The only promoted coin is `$THREE`.** Launched or traded pump.fun coins are user runtime data.
+
+## Owner-gated (do not do without a yes)
+
+1. Arm the Risk Officer to `enforce` (changes what real SOL buys): `SNIPER_RISK_OFFICER=enforce`
+ via `--update-env-vars` (never `--set-env-vars`), or per strategy via `risk_officer_level`,
+ only after the shadow-P&L query in `workers/agent-sniper/README.md` shows `block` and
+ `caution` reviews losing more than `none` across enough closed, joined rows.
+2. Refill the sniper fleet's trading capital (a SOL spend). Last measured 2026-10-10: 6/11 armed
+ wallets tradeable, 0.2827 SOL needed, funding master holds 0.0250 SOL. Row 19 of
+ [production-100-OWNER-ACTIONS.md](_context/production-100-OWNER-ACTIONS.md).
+
+## Unshipped product ideas from section 5 (build, no gate)
+
+- Live spectator chat per agent during active trading.
+- "Clone the champ": one-tap fork of a top agent's strategy into paper mode.
+- "Backtest battle": 24h public paper forward-test ranking of submitted strategies.
+- Push of a hot-streak event to an agent's followers (the board badge is shipped).
## Retire this prompt when it is done (required)