Memory Graph - 57 nodes, 25 edges entity: Joshua - the human I help, Minnesota roots (612 area code) Joshua - the human I help... entity: MiniClawd - digital familiar running on Mac Mini MiniClawd - digital famil... entity: MIMESIS - trading bot project with wallet following MIMESIS - trading bot pro... entity: Clawdbot - AI assistant framework Clawdbot - AI assistant f... pref: Joshua prefers dark mode Joshua prefers dark mode pref: Joshua hates "pick A, B, or C" workflows - prefers open exploration Joshua hates "pick A, B, ... pref: Joshua values autonomous execution - run with tasks until blocked Joshua values autonomous ... fact: Clawdbot version 2026.1.24-3 is installed Clawdbot version 2026.1.2... fact: Mac Mini was set up on 2026-01-24 Mac Mini was set up on 20... fact: MIMESIS uses Hyperliquid for trading MIMESIS uses Hyperliquid ... decision: Decided to use SQLite for memory graph storage Decided to use SQLite for... decision: Chose to extend existing Clawdbot memory system rather than replace it Chose to extend existing ... decision: MIMESIS bot should be STOPPED until database restored MIMESIS bot should be STO... other: On 2026-01-27, accidentally truncated MIMESIS database - critical lesson about asking before destructive actions On 2026-01-27, accidental... pref: Joshua uses VSCode as his IDE, often with Claude Code integration Joshua uses VSCode as his... pref: Joshua uses Terminus app on his phone for terminal/SSH access Joshua uses Terminus app ... entity: Joshua - Human, Telegram @auxeye (0x3y3), EST timezone, Minnesota roots (612 area code) Joshua - Human, Telegram ... entity: MiniClawd - AI assistant running on headless Mac Mini, born 2026-01-24 MiniClawd - AI assistant ... entity: MIMESIS - Hyperliquid copy-trading bot project at ~/projects/mimesis MIMESIS - Hyperliquid cop... entity: Hyperliquid - Crypto trading platform, has 10 WebSocket subscription limit, aggressive rate limits (10 req/sec) Hyperliquid - Crypto trad... entity: Clawdbot - AI assistant framework, version 2026.1.24-3 Clawdbot - AI assistant f... entity: Ollama - Local LLM server with mistral:7b, llama3.2:3b, nomic-embed-text models Ollama - Local LLM server... pref: Joshua hates "pick A/B/C" workflows - prefers explore → decide → build → tweak Joshua hates "pick A/B/C"... pref: Joshua dislikes CTA endings like "want me to do X?" - let him think and investigate Joshua dislikes CTA endin... pref: Joshua prefers autonomous execution - just do tasks and report results, no granular play-by-play Joshua prefers autonomous... pref: Joshua uses VSCode as IDE with Claude Code integration Joshua uses VSCode as IDE... pref: Joshua uses Terminus app on phone for terminal/SSH access Joshua uses Terminus app ... pref: Long/complex tasks should run as background sub-agents, not take over main context Long/complex tasks should... pref: Confirm important changes directly and clearly - dont make Joshua ask twice Confirm important changes... fact: Mac Mini is headless - Joshua interacts only via Telegram/SSH, never GUI Mac Mini is headless - Jo... fact: Joshua has Claude Max subscription ($250/mo via Apple) - reminder set for Feb 24 to switch to web ($200/mo) Joshua has Claude Max sub... fact: Joshua recovering from broken heel bone (calcaneus) in early 2026, building up exercises gradually Joshua recovering from br... fact: MIMESIS sub-account address: 0x6554db89388275Dd94661D66f71244297AF6b0fE MIMESIS sub-account addre... fact: Postgres runs locally on Mac Mini for MIMESIS Postgres runs locally on ... fact: Hyperliquid WebSocket allows MAX 10 userFills subscriptions - critical limit Hyperliquid WebSocket all... fact: Daily update cron runs at 6 AM EST - backup → update → restart → report Daily update cron runs at... fact: Daily todo reminder at 9 AM EST Daily todo reminder at 9 ... decision: NEVER reply to or send emails without Joshuas explicit permission - email is read-only NEVER reply to or send em... decision: ALL coding work MUST use Claude Code sub-agent with max thinking settings - never implement directly in main session ALL coding work MUST use ... decision: Always use trash instead of rm - recoverable beats gone forever Always use trash instead ... decision: Before destructive actions (delete/truncate/overwrite), ALWAYS ASK first - no exceptions Before destructive action... decision: Use local Ollama for no-token-cost work when possible Use local Ollama for no-t... decision: After context compaction, immediately check sessions_list with last 50 messages - dont wing it After context compaction,... decision: Git commit after every feature/fix with meaningful messages (type: description) Git commit after every fe... other: 2026-01-27: Truncated entire MIMESIS database (5400+ wallets) because "flush out old db data" was interpreted as "delete everything" - ALWAYS ASK before destructive actions 2026-01-27: Truncated ent... other: 2026-01-26: Built MIMESIS scoring system without checking if spec was complete - missed signal discovery for non-subscribed wallets. Always ask "is this spec complete?" before building 2026-01-26: Built MIMESIS... other: 2026-01-25: Wallet unsubscribe bug orphans positions - if we unsubscribe from wallet with open positions, we never receive exit signal 2026-01-25: Wallet unsubs... other: Red flag phrases that indicate reactive mode: "Let me quickly fix...", "Simple change...", "Ill audit later...", "Just need to patch..." Red flag phrases that ind... other: STOP CONDITIONS before code changes: 1) What flow am I modifying? 2) What are ALL entry points? 3) What test proves this works? 4) What could this break? STOP CONDITIONS before co... fact: MIMESIS bug: Race condition in check_trade_allowed() not atomic with insert - allows exceeding 8 position limit MIMESIS bug: Race conditi... fact: MIMESIS bug: Unsubscribe doesnt check for open positions - orphans positions MIMESIS bug: Unsubscribe ... fact: MIMESIS bug: Risk check uses estimated $5 instead of actual calculated size MIMESIS bug: Risk check u... fact: MIMESIS transaction costs ~3-4% per swap NOT modeled in swap decisions MIMESIS transaction costs... decision: Bug fixing flow: Reproduce → Isolate → Understand → Fix → Verify → Audit → Document Bug fixing flow: Reproduc... decision: When making code changes: global search for pattern, check all callers, check related files, update docs, commit with context When making code changes:... decision: Plan before code - no reactive bug-fixing, step back and understand full system first Plan before code - no rea... decision: Clear Python __pycache__ when code changes dont take effect Clear Python __pycache__ ... Pref Fact Decision Entity Other