Combine volatility-driven + config-driven LP strategy (merge sync)#83
Open
agushudono wants to merge 9 commits into
Open
Combine volatility-driven + config-driven LP strategy (merge sync)#83agushudono wants to merge 9 commits into
agushudono wants to merge 9 commits into
Conversation
AbortError from fetchWithTimeout has no HTTP status, so isRetryableStatus(0) was false and a single 10s per-attempt timeout threw immediately — never using the 30s maxElapsedMs retry budget. Treat timeouts/aborts as retryable so transient relay slowness retries before falling back to direct LPAgent fetch. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
FALLBACK_MODEL was "stepfun/step-3.5-flash:free", an OpenRouter model name that 404s against the configured Anthropic endpoint — so a transient 502/503/529 on the primary model would switch to a fallback that itself errors. Point it at claude-haiku-4-5-20251001 (fast/cheap, available on this API key). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Source-level fallbacks still pointed at openrouter/minimax model names (healer-alpha, hunter-alpha, minimax-m2.x) which 404 against the Anthropic-compatible endpoint. Align agent.js, config.js and the example config with the live user-config: default to claude-opus-4-7. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Default model claude-opus-4-7, fallback claude-haiku-4-5-20251001, default base URL api.anthropic.com; clarify OPENROUTER_API_KEY is a legacy fallback (live key comes from user-config llmApiKey). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
fetchMyPositions called the Agent Meridian relay (and its direct LPAgent fallback) before the portfolio lookup, but the result is only consumed in the per-position loop. With an empty wallet that loop never runs, so every management/screening cycle awaited the relay for its full ~30s retry budget and logged a misleading "raw relay failed; falling back" warning — 77x today, and on every `npm run dev`. Gate the relay + direct fetch on pools.length > 0 so an empty wallet skips them entirely. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
closePosition: disable the relay zap-out attempt. It failed 42/42 closes (the OKX zap-out order embeds an owner SOL transfer that the safety guard assertNoUnsafeSystemTransfer rejects) and always fell back to the local close + Jupiter autoswap. The doomed attempt added sign→simulate→throw latency to every close, worsening fills during fast dumps, and requested a 50% (slippageBps 5000) swap. Go straight to the safer, working local path; the guard logic is untouched and the relay is easily re-enabled via the flag. deriveLpAgentPnlPct: add collectedFee to the PnL term so already-claimed fees aren't dropped. Omitting it understated PnL by the claimed amount after every claim_fees, producing a constant reported-vs-derived gap and false "suspicious PnL" flags. Risk-exit tuning (stopLossPct -10→-6, trailingTriggerPct 3→5) lives in the gitignored user-config.json and is applied there separately. Activates on restart; the live process is a stale 2026-06-06 in-memory snapshot. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Symmetric downside handling for single-side SOL positions, which catch a falling knife when price dumps through the entry range (the main source of the -16/-21% tail losses). - index.js getDeterministicCloseRule: add rule 6 (close when active bin is more than outOfRangeBinsToClose below the range — mirror of rule 3) and rule 7 (OOR-below past outOfRangeWaitMinutes — mirror of rule 4). - index.js computeStrategy(volatility): pick LP shape by volatility — bid_ask at >= 4, else spot — and use it on the programmatic deploy path and in the screener prompt (single source of truth). Refuses non-finite/ non-positive volatility. - tools/definitions.js: align deploy_position strategy guidance with the volatility-driven rule. - lessons.js: detect "dumped below range" closes, emit a specific falling-knife AVOID lesson (downsize / require momentum for that volatility tier), and count the dump as negative evidence. Activates on restart; the live process is a stale 2026-06-06 snapshot. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Newer Opus models deprecate the temperature parameter; the request 400s with a "temperature ... unsupported/deprecated" error. Detect that case and retry the same step once with temperature omitted (same pattern as the existing thinking-mode tool_choice fallback), latching omitTemperature for the run. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Resolve diverged main (local 8 ahead / 2 behind). Conflicts in
index.js, tools/definitions.js, tools/dlmm.js, CLAUDE.md.
Strategy selection now combines both approaches:
- config.strategy.strategy === "auto" → volatility-driven shape
(bid_ask if volatility >= 4, else spot), resolved in deployPosition
- any other value ("bid_ask"/"spot") → explicit config override
- default + example now "auto"
Also realign CLAUDE.md model docs to actual Anthropic code defaults
(claude-opus-4-7 / claude-haiku-4-5 fallback) after the remote rewrite.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
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
1 participant
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.
Syncs local
mainwithorigin/mainand resolves the diverged history (local was 8 ahead / 2 behind).Strategy selection — combined approach
config.strategy.strategy === "auto"→ volatility-driven LP shape (bid_askif volatility >= 4, elsespot), resolved intools/dlmm.js#deployPosition."bid_ask"/"spot") → used as-is (config override).user-config.example.jsonset to"auto".Merge conflicts resolved
index.js— strategy block reflects auto vs explicit.tools/definitions.js— deploy tool description updated for combined logic.tools/dlmm.js—"auto"resolution indeployPosition.CLAUDE.md— took remote rewrite; realigned model docs to actual Anthropic code defaults (claude-opus-4-7/claude-haiku-4-5-20251001fallback) and theautostrategy default.Syntax checked (
node --check) on all touched JS; example JSON validated.🤖 Generated with Claude Code