feat(ai): add Claude Opus 5 and Sonnet 5 support - #1530
Open
Hansen1018 wants to merge 78 commits into
Open
Conversation
🤖 Advisory Check ResultsThese are advisory checks to help improve code quality. They won't block your PR from being merged. 📋 PR InformationTitle Format: ✅ Good - Follows Conventional Commits 🔧 Backend ChecksGo Formatting: Files needing formattingGo Vet: ✅ Good Fix locally: go fmt ./... # Format code
go vet ./... # Check for issues
go test ./... # Run tests⚛️ Frontend ChecksBuild & Type Check: ✅ Success Fix locally: cd web
npm run build # Test build (includes type checking)📖 ResourcesQuestions? Feel free to ask in the comments! 🙏 These checks are advisory and won't block your PR from being merged. This comment is automatically generated from pr-checks-run.yml. |
🤖 Advisory Check ResultsThese are advisory checks to help improve code quality. They won't block your PR from being merged. 📋 PR InformationTitle Format: ✅ Good - Follows Conventional Commits 💡 Suggestion: This is a large PR. Consider breaking it into smaller, focused PRs for easier review. 🔧 Backend ChecksGo Formatting: Files needing formattingGo Vet: ✅ Good Fix locally: go fmt ./... # Format code
go vet ./... # Check for issues
go test ./... # Run tests⚛️ Frontend ChecksBuild & Type Check: ✅ Success Fix locally: cd web
npm run build # Test build (includes type checking)📖 ResourcesQuestions? Feel free to ask in the comments! 🙏 These checks are advisory and won't block your PR from being merged. This comment is automatically generated from pr-checks-run.yml. |
Hansen1018
changed the title
Add two new Anthropic models released in 2026: - Claude Opus 4.8 (claude-opus-4-8) — released 2026-05-28, the immediate successor to Opus 4.7. Anthropic kept 4.7's pricing; the per-call estimate here (USD 0.18/call) follows the existing repo convention from PR NoFxAiOS#1510. - Claude Fable 5 (claude-fable-5) — released 2026-06-09, the first publicly available Mythos-class model (a tier above Opus). $10 input / $50 output per 1M tokens per anthropic.com; the per-call estimate (USD 0.36/call) is roughly 2× Opus 4.8's to mirror the 2× token price. Both share 1,000,000 context, 128,000 max output, and the standard Anthropic Messages API. Reference: https://www.anthropic.com/news/claude-fable-5 https://www.anthropic.com/news/claude-opus-4-8 Files changed (5, +25 / -5): - mcp/provider/claude.go `DefaultClaudeModel` bumps to `claude-fable-5` (the newer Mythos-class flagship). - api/handler_ai_model.go the `claude` row in `supportedModels` exposes `claude-fable-5` so the UI shows the same default the runtime picks up. - mcp/payment/claw402.go `claw402ModelEndpoints` adds `claude-opus-4-8` and `claude-fable-5` x402 paths; the legacy `claude-opus` alias is kept. - store/ai_charge.go `modelPrices` gains `claude-opus-4-8 = 0.18` and `claude-fable-5 = 0.36` with comments citing the anthropic.com pricing sheet. - web/src/components/trader/model-constants.ts Both tiers enter `CLAW402_MODELS` (Fable 5 with the dragon icon \u{1F409}) and `BLOCKRUN_MODELS`; `AI_PROVIDER_CONFIG`. `claude.defaultModel` -> `claude-fable-5`. Backward compatible — existing `claude-opus`, `claude-opus-4-6` entries and prices remain unchanged and selectable. Default model policy: this PR is the single place where the Anthropic default changes (to Fable 5). All `WithModel(...)` override callers continue to pick any model they want via the existing hook.
Hansen1018
force-pushed
the
feat/add-claude-5
branch
from
July 25, 2026 09:18
548ed55 to
a8597a7
Compare
🤖 Advisory Check ResultsThese are advisory checks to help improve code quality. They won't block your PR from being merged. 📋 PR InformationTitle Format: ✅ Good - Follows Conventional Commits 🔧 Backend ChecksGo Formatting: Files needing formattingGo Vet: ✅ Good Fix locally: go fmt ./... # Format code
go vet ./... # Check for issues
go test ./... # Run tests⚛️ Frontend ChecksBuild & Type Check: ✅ Success Fix locally: cd web
npm run build # Test build (includes type checking)📖 ResourcesQuestions? Feel free to ask in the comments! 🙏 These checks are advisory and won't block your PR from being merged. This comment is automatically generated from pr-checks-run.yml. |
🤖 Advisory Check ResultsThese are advisory checks to help improve code quality. They won't block your PR from being merged. 📋 PR InformationTitle Format: ✅ Good - Follows Conventional Commits 🔧 Backend ChecksGo Formatting: Files needing formattingGo Vet: ✅ Good Fix locally: go fmt ./... # Format code
go vet ./... # Check for issues
go test ./... # Run tests⚛️ Frontend ChecksBuild & Type Check: ✅ Success Fix locally: cd web
npm run build # Test build (includes type checking)📖 ResourcesQuestions? Feel free to ask in the comments! 🙏 These checks are advisory and won't block your PR from being merged. This comment is automatically generated from pr-checks-run.yml. |
Operator choice to add aggression while keeping the hold-for-big-moves design intact: - Leverage 5x -> 10x: a wide -5% stop is now -50% of margin — still survivable, not an instant liquidation (that needed 20x). - Per-position notional 2.5x -> 5x equity; 2 positions = 10x total account notional (full margin at 10x, ~10% liquidation cushion). - Stops/targets, throttle, min-hold and noise-band settings unchanged: same wide (-5% / +10-12%) exits, doubled exposure. Live strategy in data/data.db updated to match (lev=10, ratio=5.0).
…t-privilege permissions pr-checks-comment.yml runs in the privileged workflow_run context with a write token while consuming artifacts produced by the untrusted PR workflow. Seven spots template-interpolated that untrusted data (and the fork-controlled head branch name) directly into github-script source — a crafted artifact could escape the string literal and run arbitrary JS with the privileged token (CodeQL actions/code-injection, critical). - All untrusted values now flow through env vars and process.env; the PR number is parsed and validated before use - test.yml / docker-build.yml gain workflow-level 'permissions: contents: read' (CodeQL actions/missing-workflow-permissions); publish jobs keep their job-level packages:write
Replay recorded AI decisions from decision_records under alternative risk-control and throttle parameters, then search the 15-dim space with Optuna multivariate TPE. - extract.py: parse decision cycles, intents (incl. throttled ones), 15m OHLCV and per-cycle prices out of input prompts into CSVs - simulate.py: replay engine mirroring auto_trader_throttle.go semantics (margin-based PnL%% thresholds, intra-candle SL/TP trigger orders, opens/hour + reentry + margin gates); live-config replay reproduces the real account curve (-83%% sim vs -85%% actual) - search.py: TPE search (train/holdout split) + baselines + importances Findings: train-window optima do not survive holdout (overfit); only 2/800 trials are positive across all 3 time folds. Dominant lever is min_confidence (importance 0.79) — the edge problem is decision quality, not risk parameters.
Live decomposition of the 2026-07-23 drawdown (equity 62.6 -> 31.1 in one day, 43 fills, avg hold <1h despite a 4h min hold): both exit guards compared thresholds meant as price moves against leverage-multiplied margin PnL%, so raising leverage 5x -> 10x silently halved every trigger distance: - throttle bypass: -5% 'stop' unlocked AI closes at a -0.5% price wiggle (ETH -0.5%/44min, AAVE -0.9%/45min, MU -2.4%/14min all sailed through) - drawdown monitor: 'profit > 5%' armed at +0.5% price, then market-closed winners on a 40% giveback (SNDK +0.6%/35min, NVDA +0.3%/37min) — every winner strangled at breakeven while losers realized fast Fix: evaluate both guards on price-basis PnL (margin PnL% / leverage). Throttle thresholds keep their documented price meaning at any leverage; the drawdown monitor now arms only after a real +5% price move. Peak-PnL cache stays margin-basis for prompt display consistency. Replay simulator updated to the same price-basis semantics.
The trading account could only be funded outside NOFX: no way to move
spot ('main wallet') USDC into the perp trading account, and no deposit
address/QR shown anywhere.
- API: /hyperliquid/account now also returns spot USDC (total/available)
alongside the perp summary; /hyperliquid/submit-exchange accepts a
validated usdClassTransfer action (positive plain amount, boolean
toPerp, Mainnet chain) relayed like the existing approval actions
- Web: new HyperliquidFundsPanel in the exchange config modal with a
deposit tab (wallet address QR + Arbitrum/spot routing notes) and a
transfer tab (spot<->perp, signed by the user's own wallet via EIP-712
UsdClassTransfer — user-signed actions derive the account from the
signature, so the agent key NOFX holds cannot and does not move funds)
- Shared EIP-712/provider helpers extracted from HyperliquidWalletConnect
into lib/hyperliquidWallet.ts
…ding copy The deposit tab claimed USDC sent to the user's own address on Arbitrum is auto-credited to the perp account. Wrong: sending to your own address only funds the wallet. Hyperliquid's canonical deposit is native USDC on Arbitrum One sent to the validator-controlled Bridge2 contract (0x2Df1c51E09aECF9cacB7bc98cB1742757f163dF7), which credits the SENDER's account (min 5 USDC, ~1 min) — verified against Arbiscan and HL docs. - Deposit tab now explains the two-step flow (fund wallet -> bridge in) and adds a one-click deposit: switches the wallet to Arbitrum One and sends an ERC-20 transfer of native USDC to Bridge2 - Enforces the connected wallet == configured main wallet (the bridge credits the sender, so any other wallet would fund the wrong account) - Blocks sub-minimum amounts client-side (they would be lost)
The deposit tab told users to deposit wallet USDC but never showed how much the wallet held — the 328 USDC sitting on Arbitrum was invisible, so the deposit flow looked like there was nothing to deposit. - Third balance card 'Wallet (Arbitrum)': native USDC balance + ETH gas (red warning when gas is zero), fetched directly from the public Arbitrum RPC (balanceOf + eth_getBalance), refreshed with the panel - 'Depositable' amount shown next to the bridge form + Max button - Client-side check: deposit amount cannot exceed the wallet balance
…and transfer tab Per HL's account-abstraction-modes docs, unified accounts (the default and HL's recommendation) hold ONE USDC balance that collateralizes spot, validator perps and HIP-3 perps — there is no spot/perp split and no class transfer. The panel was presenting manual-mode concepts (spot vs perp cards, a spot->perp transfer tab) to unified users: confusing and wrong. - unifiedAccount prop (from the exchange's hyperliquid_unified_account, default true): hides the transfer tab and shows a single 'Hyperliquid account' card — total / tradable / margin in use, derived from the spot balance and its hold - Manual-mode accounts keep the split cards and transfer tab - Deposit copy simplified: funds are tradable right after the bridge credit, no extra step
Per-position one-click close already existed (Action column, with confirm), but flattening the whole book took one confirm per position. Adds a 'Close All' button next to the position count: single confirm, then closes every position sequentially (parallel closes can race on exchange nonces/rate limits), refreshes via SWR and reports a partial- failure count. zh/en/id translations included.
…live' panel The one-click close only existed on the legacy dashboard's positions table; the terminal dashboard the user actually sees had a read-only positions panel. Adds a per-row 'close' button (confirm -> market close via the existing /close-position API) and a 'close all' button in the panel header (single confirm, sequential closes to avoid nonce races, partial-failure report). Hidden in demo mode and when no trader is selected; SWR keys refresh positions and account after closing.
The anti-churn exit gates (min hold, noise-close window, re-entry cooldown, bypass and noise-band thresholds) were hardcoded constants in auto_trader_throttle.go — every flat-ish position was forced to hold 8h+ and changing the pacing meant a code change and redeploy. - RiskControlConfig gains 7 exit-gate fields (minutes / signed price-%), zero = built-in default; accessor methods centralize fallbacks and are hot-reloaded from the DB like the rest of the strategy config - Throttle reads the gates from the strategy; prompt hold/exit guidance is now rendered from the same values so the AI is told exactly what the code will enforce - Defaults softened: min hold 4h -> 1.5h, noise window 8h -> 3h, re-entry 3h -> 1.5h, bypasses -5/+12 -> -3/+8, noise band -4..+6 -> -2..+3 (price-basis). A +-2-3% move is 15-20x round-trip fees — no longer 'noise' worth locking - Strategy Studio gets an 'exit pacing' row (min hold / flat-close window / re-entry cooldown); thresholds editable via strategy JSON - Live strategy updated in data/data.db with the softened values
… values Operator call: no per-strategy configurability for exit pacing — it added config fields, UI and plumbing nobody wanted. Back to plain constants in auto_trader_throttle.go, now set to the replay-validated values instead of the original guesses (4154 cycles, 3-fold robust search over only the 7 exit params, everything else fixed at the live config): - min hold 90m, noise-close window 3h, noise band -2%..+3%, bypasses -3%/+8% (all inside the searched top-20 ranges) - re-entry cooldown 4h — the one clear signal: top-20 configs cluster tightly at 3.8-4.0h; re-entering a just-closed symbol was a consistent loss source Removes the 7 RiskControlConfig fields + accessors, the Strategy Studio 'exit pacing' row, the frontend type fields, and the fields from stored strategy configs in data/data.db. Prompt guidance is static text again, matching the constants.
Clears all 7 critical Dependabot alerts (SSH subsystem: auth bypass via unenforced @Revoked status, VerifiedPublicKeyCallback permission skip, key-constraint enforcement gaps, server deadlock and infinite-loop DoS). Transitive dependency only — nothing in NOFX imports x/crypto directly.
- react-router/react-router-dom 7.17.0 -> 7.18.1 (route-matching DoS, GHSA-chx6-hx7r-mcp5) - axios 1.16.1 -> 1.18.1 (inherited-proxy leak, GHSA-gcfj-64vw-6mp9) - postcss 8.5.15 -> 8.5.23 (source-map path traversal, GHSA-r28c-9q8g-f849) - brace-expansion pinned to 1.1.16 / 2.1.2 across all nested copies via overrides (exponential-expansion DoS, GHSA-3jxr-9vmj-r5cp) Remaining GHSA-qwww-vcr4-c8h2 (react-router RSC-mode CSRF) is dismissed on GitHub as not-used: the app is a Vite SPA, RSC mode is never enabled, and the fix requires a react-router v8 major upgrade.
Hansen1018
force-pushed
the
feat/add-claude-5
branch
from
July 27, 2026 02:37
df6548b to
d9edeb4
Compare
🤖 Advisory Check ResultsThese are advisory checks to help improve code quality. They won't block your PR from being merged. 📋 PR InformationTitle Format: ✅ Good - Follows Conventional Commits 💡 Suggestion: This is a large PR. Consider breaking it into smaller, focused PRs for easier review. 🔧 Backend ChecksGo Formatting: Files needing formattingGo Vet: ✅ Good Fix locally: go fmt ./... # Format code
go vet ./... # Check for issues
go test ./... # Run tests⚛️ Frontend ChecksBuild & Type Check: ✅ Success Fix locally: cd web
npm run build # Test build (includes type checking)📖 ResourcesQuestions? Feel free to ask in the comments! 🙏 These checks are advisory and won't block your PR from being merged. This comment is automatically generated from pr-checks-run.yml. |
# Conflicts: # api/handler_ai_model.go # api/handler_hyperliquid_wallet_test.go # mcp/payment/claw402.go # store/ai_charge.go # web/src/components/common/HyperliquidFundsPanel.tsx # web/src/components/common/HyperliquidWalletConnect.tsx # web/src/components/trader/model-constants.ts # web/src/lib/hyperliquidWallet.ts
🤖 Advisory Check ResultsThese are advisory checks to help improve code quality. They won't block your PR from being merged. 📋 PR InformationTitle Format: ✅ Good - Follows Conventional Commits 🔧 Backend ChecksGo Formatting: Files needing formattingGo Vet: ✅ Good Fix locally: go fmt ./... # Format code
go vet ./... # Check for issues
go test ./... # Run tests⚛️ Frontend ChecksBuild & Type Check: ✅ Success Fix locally: cd web
npm run build # Test build (includes type checking)📖 ResourcesQuestions? Feel free to ask in the comments! 🙏 These checks are advisory and won't block your PR from being merged. This comment is automatically generated from pr-checks-run.yml. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
claude-opus-5andclaude-sonnet-5.Relationship to #1527
This PR is based on and depends on #1527, which introduces Claude Opus 4.8 and Claude Fable 5 support. Once #1527 is merged, this PR adds the remaining current Claude 5 models without changing the existing Anthropic integration architecture.
Change Type
Scope
Official References
claude-opus-5claude-sonnet-5The per-call values follow the repository’s existing approximate call-pricing convention.
Testing
go test ./api ./mcp/payment ./mcp/provider ./storego vet ./api ./mcp/payment ./mcp/provider ./storecd web && npm run test— 128 tests passedcd web && npm run buildSecurity Impact
Compatibility
claude-opus-5.Scope Boundary
This PR does not change Anthropic authentication, request/response wire formats, trading behavior, strategy logic, or database schema.
Depends on #1527.