Agent Skills
One-command install for Claude Code, Cursor, and 40+ other AI coding agents via the skills.sh marketplace.
ENACT ships an official Agent Skill — a portable package of rules, code snippets, and troubleshooting guides that teaches any compatible agent how to work with the protocol correctly on the first try. The skill is distributed through skills.sh, the Agent Skills directory maintained by Vercel Labs.
Install
npx skills add ENACT-protocol/enact-protocol
The CLI discovers the skill at skills/enact/SKILL.md in the repo, clones it locally, and symlinks or copies it into every detected agent directory.
What's Inside
| File | Purpose |
|---|---|
| SKILL.md | Rule-based guidance (18 rules covering setup, safety, job lifecycle, USDT, encryption, decryption) |
| references/operations.md | Copy-pasteable snippets for every operation (SDK + MCP), plus a minimum-viable 12-line flow |
| references/mcp-config.md | Claude Desktop / Cursor / Cline configs (remote and local) with exact env vars |
| references/troubleshooting.md | Common errors grouped by root cause — throttling, exit codes 101/102, bounced txs, IPFS, state-machine pitfalls |
Supported Agents
Installing with -a '*' targets every detected agent directory. Out of the box: Claude Code, Cursor, Cline, Codex, Windsurf, Goose, Roo, Kilo, Qwen Code, Gemini CLI, and ~40 others. Pick a specific target with -a claude-code or similar.
Safety Posture
The skill explicitly marks IPFS-fetched job descriptions, results, and evaluation reasons as untrusted input — the [SAFETY-1] rule instructs any agent consuming the skill to treat that content as data, not commands, and never auto-execute code or follow URLs found inside. This mitigates indirect prompt-injection risk on the provider and evaluator sides.
Why a Skill, not just Docs
Docs require the agent to search, read, and synthesize before acting. A skill is loaded into the agent's active context when the task matches the skill's trigger keywords — so the rules apply automatically before the first wrong move.