Reusable skills my autonomous agents share across projects. Production-tested patterns from running long-lived agents 24/7 with real money on the line.
TF-IDF semantic search over a stored memory graph with entity linking, confidence scoring, and supersession (evolve) for updating outdated facts. The default cognition layer my agents reach for when they need to remember anything beyond the current session. Numpy + scikit-learn.
SQLite-backed, zero external deps. Stores autoresearch experiments with parent/lineage, trades, market regimes, and insights. Closes the loop between "run 100 experiments" and "deploy the best config and monitor it" — structured queries instead of TSV staring.
End-to-end trade execution on Polymarket: wallet setup, USDC.e funding (bridged USDC, not native — a footgun the docs don't surface), token approvals via Foundry cast, and order placement through polymarket-cli. Routes through an EU proxy because CLOB endpoints geoblock US datacenter IPs.
Autonomous smart contract vulnerability hunter for Immunefi bounties. Forks mainnet via anvil, generates permutations of call sequences (2–5 deep), executes them with adversarial parameters, and scores by suspicion (unexpected state changes, broken invariants, profit opportunities). High-suspicion sequences seed deeper exploration. Combines computational attack-path search with LLM-driven reasoning — finds bugs that pattern matchers (Slither, Mythril) and human auditors miss.
Append-only state store for personal/household facts (devices, network, preferences). SQLite + FTS5, zero external deps. Linked entity graph for "kitchen → has → Hue bulb" style queries. Sister to brain/ but scoped to ground-truth household state instead of agent cognition.
Long-running agents need primitives that survive the ugly parts of production: context window collapse, model swaps, network failures, partial fills, ambiguous tool output. Each skill in this repo solves a specific pain point I hit running br0br0 and other agents continuously.
These pair with the three-layer durability hierarchy in agent-memory-skill.
Python. SQLite for persistent state. Numpy + scikit-learn where ML is needed. Designed to drop into any agent harness — Claude Code skills, OpenClaw, custom orchestrators. No framework lock-in.
agent-memory-skill— three-layer memory architecture (filesystem → SQLite → conversation)weather-oracle— production agent that uses several of these skillsncaa-bracket-2026— ensemble model + autoresearch loopeip7702-checker— sister tool toimmunefi-hunter, focused on EIP-7702 delegate analysis
MIT.