# AVAX Skills

Independent Avalanche development library inspired by [ETHSkills](https://ethskills.com/). Original guidance and design; no affiliation with Ava Labs, the Avalanche Foundation, or ETHSkills.

## Included

- **31 skills:** one router and 30 focused topics, including architecture, standards, audit, orchestration, QA and Noir.
- **Seven protocol recipes:** Pharaoh, Aave V4, LFJ, Blackhole, Silo, Euler and BENQI, with source-pinned manifests, runtime checks, unsigned plans and successful representative local-fork operations.
- Runnable contract/browser/indexer reference app, typed-authorization and agent-payment validators, and actual Noir proof → generated Solidity verifier integration.
- Searchable static site, raw Markdown, exact-source pages, machine manifest, evidence browser and complete executable download.

Source review: **2026-09-06**. Observations are dated, not live quotes or evergreen deployment guarantees. See [capability matrix and limits](PARITY.md).

## Build the website

Python 3.11+ is sufficient for the static site; no npm dependency is required for that build.

```sh
python3 scripts/build.py
python3 scripts/serve.py --port 4179
```

Open the loopback URL printed by the server. `dist/` is the website release directory. No public deployment, domain, repository publication or wallet transaction is implied. Publish only a reviewed `dist/`, never the workspace/project root.

## Run the examples

Demonstrated platform: Linux x86_64, Node22+, Python3.11+, and browser runtime system dependencies. Use the environment's permitted npm command and keep its supply-chain protections enabled.

```sh
npm ci
npm run setup:browser
python3 scripts/setup_native_tools.py
python3 scripts/setup_native_tools.py --check
npm test
npm run test:full
npm run test:forks
```

`npm test` runs protocol, standards, indexer and wallet-state tests. `test:full` adds Solidity/fuzz, real browser/local-EVM and real Noir/Solidity verification. `test:forks` performs public read-only inspections and writes only to disposable loopback forks. It requires an available Avalanche RPC and current matching protocol deployments.

The native installer verifies pinned official downloads and extracted binary hashes, installs only under `.tooling`, and supports Linux x86_64. Other platforms need compatible tools matching `config/native-tools.json`; executable-path overrides are listed in the examples. No production keys are required. The JavaScript lockfile pins ethers6.16.0, Playwright1.58.2 and a patched ws8.21.3 override accepted by the guarded installer.

Interactive educational application:

```sh
node examples/core/demo.mjs
```

The demo uses **local Anvil chain43113, not public Fuji**. Its controlled wallet adapter is test infrastructure, not a real browser wallet extension. The custody ledger is not yield or ERC4626. Assets must have exact transfers and stable balances; arbitrary out-of-band rebases are unsupported, not automatically detected.

## Agent use

In a source checkout, give the agent `content/SKILL.md`; executable commands run from the project root. In the download, root `SKILL.md` routes the same topics and code lives beside it. The duplicated `content/` copies allow rebuilding the website from the downloaded bundle.

Use the site's copied prompt only with its actual served URL. There is no claimed public package-install command or registered domain.

## Evidence and limitations

- [Protocol fork swaps](research/parity-dexes/fork-summary.json)
- [Aave/BENQI execution](research/parity-aave-benqi/fork-proof.json)
- [Silo/Euler execution](research/parity-silo-euler/fork-evidence.json)
- [Core browser/EVM/indexer](research/parity-core/e2e-evidence.json)
- [Noir proof/EVM verification](research/parity-privacy/evidence.json)
- [Standards live observations](research/parity-standards/live-evidence.json)
- [Independent agent build results](research/parity-agent/summary.json)
- [Read-only contract review](research/parity-audit/read-only-review.json)

The baseline and skill-assisted agents each passed the same24 independent LFJ checks. This demonstrates a bounded usable workflow, **not measured improvement**, statistical superiority or a direct benchmark against ETHSkills.

An optional seeded-exploit evaluation was blocked by automatic cyber review and stopped. It is not counted as passing evidence; unexecuted seeded fixtures are excluded from the download. The separate read-only core review led to documented compatibility/conservation regressions.

Funded Silo/Euler borrowing, mature sAVAX redemption, protocol governance/claims/LP operations, actual x402 settlement, real wallet extensions, full L1/ICM infrastructure and public deployment are outside demonstrated execution. This is not a third-party protocol audit.

## Source fidelity

Durable skill content is authored through Skill Workshop, then exported unchanged from its private revision. `catalog.json` controls site presentation, not guide content. Builds preserve hashes of raw guides and bundled source/evidence. The private authoring/workflow receipts are excluded from distribution.

`python3 scripts/verify_build.py` validates the static release and reproducibility. Browser checks use `node scripts/qa.mjs` and `node scripts/qa_final.mjs`; start separate previews on4179 and4180 with `--mount /preview/` on4180 first. All interactive results and limitations remain separately labeled.