Blockchain Skills Agent

Portable markdown skills + a Python autonomous loop on Base — discover, plan, validate, execute, verify, log. Built for Synthesis (Protocol Labs, Open Track, Uniswap).

Project cover — Blockchain Skills Agent

Repository

Source, tests, agent.json, skills/, and runbook.

GitHub →

Mainnet proof

Real Base swap via Uniswap Trading API (see agent_log.json in repo).

BaseScan tx →

Try locally

git clone https://github.com/CuongTranXuan/blockchain-skills-agent.git
cd blockchain-skills-agent
python3 -m venv .venv && .venv/bin/pip install -r requirements.txt
# configure .env from .env.example — never commit secrets
.venv/bin/python -m agent.main --scenario happy-path --dry-run
.venv/bin/python -m pytest tests/ -q
Autonomous loop diagram: Discover through Log

Demo scenarios

Sharp terminal mockups (HTML/CSS) — no upscaled PNGs. happy-path reflects the real Base mainnet proof; blocked-path / failure-path show guardrails and error handling.

happy-path · mainnet
python -m agent.main --network mainnet --scenario happy-path
[OK] discover → portfolio loaded
[OK] plan → ETH → USDC (Uniswap quote)
[OK] validate → limits & chain checks passed
[OK] execute → tx broadcast
[OK] verify → receipt matches intent
[OK] log → agent_log.json
SUCCESS · tx confirmed
0.005 ETH → USDC · Base
tx 0x5061…a8db → BaseScan
happy-path Full autonomous loop with real mainnet execution and structured receipts. Use the BaseScan link above to verify.
blocked-path · dry-run
python -m agent.main --scenario blocked-path --dry-run
STATUS simulation · validation gate
VALIDATION plan rejected
reason: exceeds safety limits (slippage / notional / allow-list)
action: execution skipped — no transaction broadcast
Guardrails worked as intended.
blocked-path Deterministic validate stops unsafe plans before signing. Rehearse with --dry-run.
failure-path · demo
python -m agent.main --scenario failure-path
[OK] validate passed (intent allowed)
[ERR] execute → simulated_failure
[LOG] agent_log.json updated (step + error)
[NOTIFY] human escalation recommended
skill-debug: classify · summarize · next steps
failure-path Execution fault path with structured logging and skill-debug-style human notification (demo scenario).