AI Agent Skills for Crane

Crane ships with a rich library of skills under .claude/skills/. These enable Claude Code, Bankr agents, OpenClaw, Cursor, and other compatible tools to author correct, secure, gas-efficient Diamond contracts using Crane patterns.

Public skills live under .claude/skills/. The tree is curated for Crane product + protocol architecture + Foundry + borderline TS tooling. Personal/Bankr-ecosystem bazaar skills are not tracked in this repository.

Core Crane Skills (Start Here)

  • crane-architecture — Facet-Target-Repo, storage slots, guard functions, AwareRepo, Service, DFPkg rules.
  • crane-deployment — CREATE3, DiamondPackageCallBackFactory, FactoryService, Init*Service, salt conventions.
  • crane-testing — TestBase, Behavior libraries, handlers, invariants, comparators.
  • crane-adversarial-testing — Abuse/attack catalogs for diamonds/vaults.
  • crane-code-style — Headers, naming (_layoutStruct, param_), no viaIR, struct patterns for stack.
  • crane-natspec — Full documentation requirements with include-tags and custom selectors. Values: CENTRALLY_COMPUTED_NATSPEC_VALUES.md.
  • crane-access — Operable, ERC8023 MultiStepOwnable, reentrancy.
  • crane-tokens — ERC20/2612/4626 native implementations + DFPkgs + Permit2 aware.
  • crane-utilities — Math (ConstProdUtils), sets, EIP712, cryptography, pagination.
  • crane-porting — How to vendor protocols into contracts/external + contracts/protocols with shared transitive deps (no private OZ clones).
  • crane-porting-verification — Hermetic/fork tests, Behaviors, and definition-of-done gates for ports.
  • docs-to-skills — Crawl full documentation sites/trees; inventory every page; emit multi-skill families with coverage reports.
  • skill-authoring — Progressive disclosure, description triggers, compartmentalized references/, quality checklists for SKILL.md.

Agent identities

  • crane-porter (.claude/agents/crane-porter.md) — end-to-end protocol porting sessions (vendor, remap, wrap, verify).
  • docs-skill-scribe (.claude/agents/docs-skill-scribe.md) — documentation scrape → progressive-disclosure skill families.

Protocol Skills (Reusable Ports)

Extensive high-quality ports with dedicated skills:

  • Balancer V3 (all pool types: weighted, stable, gyro, ReClamm, COW, hooks, vault ops)
  • Uniswap V2 / V3 / V4 (pools, positions, swaps, hooks, flash accounting)
  • Aerodrome + Slipstream (pools, gauges, CL math, rewards, voter)
  • Aave v3 + v4 Hub/Spoke (full architecture, config, liquidation, tokens, position mgr)
  • Euler (EVC, EVault, risk, oracles, periphery)
  • pons (Robinhood Chain launchpad): pons-architecture, pons-operations, pons-integration (+ pons-family coverage/sources) — docs-derived end-user and integrator skills for v1 (live Uniswap V3) and v2 (curve → V4; addresses pending)
  • And many more (Pendle, Frax ecosystem, Reliquary, Resupply, Comet, Permit2, Chainlink VRF, Olympus, Morpho, Reactive, etc.)

Each protocol skill teaches both the external protocol and how Crane wraps it with services, aware repos, DFPkgs, and test infrastructure.

How Agents Should Use Skills

  1. When the user asks for a feature, call the relevant skill(s) first.
  2. Generate code that follows the documented patterns exactly.
  3. Add NatSpec + tags.
  4. Write accompanying tests using the testing skill.
  5. After delivery, consider contributing an updated or new skill so the knowledge persists for all agents.

Maintaining Skills

Skills live in this repo under .claude/skills/<name>/SKILL.md.

  • Keep them concise but example-rich.
  • Include "references/" sub-files for long examples.
  • Update when the underlying contracts or best practices evolve.
  • Remove or archive stale "copy" directories.

See the root AGENTS.md for broader instructions.

Installable marketplaces

For agents that load skills via Claude Code / Codex / Grok / OpenCode marketplaces:

MarketplaceAudienceInstall
cyotee/cyotee-claude-pluginsDevelopers building on Crane and DeFi protocols/plugin marketplace add cyotee/cyotee-claude-plugins then /plugin install crane@cyotee
cyotee/defi-agent-skillsAgents operating on-chain (cast/Bankr runbooks)/plugin marketplace add cyotee/defi-agent-skills
  • Forge skills (forge-testing, forge-fuzz-testing, forge-deployment)
  • Optional TS/JS agent tooling (tevm, voltaire-effect, wagmi) via the developer marketplace

Using these skills together gives agents a structured way to build and ship modular on-chain software with Crane patterns.