From zama-skills
One-command end-to-end pipeline. Runs design → init → contract → test → audit → deploy → frontend in sequence, pausing only at safety gates (audit findings, the disable-model-invocation deploy step, frontend smoke test). Best for first-time users who want a working confidential dApp without thinking about which skill comes next.
How this skill is triggered — by the user, by Claude, or both
Slash command
/zama-skills:autonomousThis skill is limited to the following tools:
The summary Claude sees in its skill listing — used to decide when to auto-load this skill
`/zama-autonomous` is the **end-to-end orchestrator**. One command, six follow-on skills, two safety gates. Best for:
/zama-autonomous is the end-to-end orchestrator. One command, six follow-on skills, two safety gates. Best for:
It does NOT replace the individual skills — it just sequences them and asks the user before any irreversible step.
This skill talks to two MCP servers. The first is required; the second is recommended for higher-quality UI output.
| MCP | Status | Why | Install |
|---|---|---|---|
context7 | REQUIRED | Live Zama / OpenZeppelin Confidential / fhEVM docs (anti-hallucination guarantee) | claude mcp add context7 -- npx -y @upstash/context7-mcp |
magic (21st.dev) | RECOMMENDED for /zama-frontend and /zama-design | Production-grade UI component scaffolding (shadcn-flavored, design-system-aware) | claude mcp add magic -- npx -y @21st-dev/magic (sign-in required) |
Before invoking any mcp__context7__* or mcp__magic__* tool, verify the tool is available. If a context7 call would fail (tool not found / not in the available tool list):
STOP. Do NOT generate any code or write any file.
Tell the user (verbatim, do not paraphrase):
This skill requires the context7 MCP server to fetch live Zama documentation.
It does not appear to be installed.
Install it (one-time setup):
claude mcp add context7 -- npx -y @upstash/context7-mcp
After install, restart Claude Code (or run /mcp to verify) and re-run this skill.
Wait for the user to confirm install. Re-attempt the call. If it still fails, tell the user to run /zama-doctor for a full diagnostic.
If a magic call would fail (only relevant for /zama-frontend and /zama-design UI generation):
Do NOT stop — magic is optional. Continue with hand-authored shadcn components.
Tell the user (once, near the start of UI generation):
Magic MCP (21st.dev) is not installed. UI components will be hand-authored
using shadcn primitives. For higher-quality, design-system-aware components,
install Magic:
claude mcp add magic -- npx -y @21st-dev/magic
Then restart Claude Code and re-run this skill.
Continue without magic.
context7 is hard-required. Every Zama / OpenZeppelin / fhEVM API the skill emits is verified against /zama-ai/fhevm (1772 HIGH-reputation snippets) and /websites/openzeppelin_confidential-contracts (354 snippets). A WebFetch fallback would weaken the anti-hallucination guarantee — if context7 is unavailable, the right answer is to fix the setup, not to silently degrade.
Run /zama-doctor — it lists every required and recommended MCP/tool with install commands and a status check.
Single source of truth. Every SKILL.md in this plugin transcludes this block via a
@sync:shared:context7-querymarker pair. Edit here, runpnpm sync, and all skills update.
Zama Protocol package surfaces (@fhevm/solidity, @zama-fhe/relayer-sdk, @fhevm/hardhat-plugin, @openzeppelin/confidential-contracts) evolve faster than any LLM training cut-off. To avoid hallucinated APIs, always query context7 for live documentation before emitting Zama-related code.
When this skill activates, BEFORE generating any Solidity, TypeScript, or config file that touches fhEVM, perform the following calls in order:
Resolve the primary fhEVM library
mcp__context7__resolve-library-idlibraryName: "fhevm" → expect /zama-ai/fhevm (HIGH reputation, ~1772 snippets).Fetch topic-scoped fhEVM docs for the operation in progress
mcp__context7__get-library-docscontext7CompatibleLibraryId: "/zama-ai/fhevm"topic: narrowed to the user's question (e.g., "acl", "decryption", "euint", "relayer", "auction").Fetch hardhat-template scaffolding docs when generating build/test infra
context7CompatibleLibraryId: "/zama-ai/fhevm-hardhat-template"topic: such as "deploy", "test", "mock".Fetch OpenZeppelin Confidential Contracts when generating tokens, governance, or vesting
context7CompatibleLibraryId: "/websites/openzeppelin_confidential-contracts"topic: such as "ERC7984", "VotesConfidential", "FHESafeMath".Fallback narrowing — if returned docs are too broad, re-call get-library-docs with a tighter topic: parameter focused on the user's exact question.
fhevmjs or fhevm (root package). Both were officially deprecated 2025-07-10. Replacements are @zama-fhe/relayer-sdk and @fhevm/solidity respectively.pinned-versions.json, the JSON wins. That file was npm-registry-verified on 2026-05-03 and reflects peer-dep alignment across the Zama stack. Treat docs as API guidance; treat the JSON as version truth.https://docs.zama.org/protocol/protocol-apps/addresses/testnet/sepolia at skill runtime.| Library ID | Snippets | Use For |
|---|---|---|
/zama-ai/fhevm | 1772 | Solidity FHE primitives, ACL, decryption, encrypted types |
/zama-ai/fhevm-hardhat-template | 43 | Hardhat config, deploy scripts, mock testing |
/websites/openzeppelin_confidential-contracts | 354 | ERC-7984, governance, FHESafeMath |
[/zama-doctor] ← preflight (auto, exits if context7 missing)
↓
[/zama-design] ← describe your idea, get DESIGN.md + UI-WIREFRAME.md
↓
── pause ── "Continue with this design?" Y/N
↓
[/zama-init] ← scaffold the monorepo
↓
[/zama-contract] ← write the confidential contract (reads DESIGN.md)
↓
[/zama-test] ← generate mock + Sepolia tests
↓
[/zama-audit] ← ACL / cleartext / HCU / deprecation review
↓
── gate ── audit clean (exit 0)? yes → continue. warn/critical → ask user.
↓
[/zama-deploy] ← Sepolia deploy (manual confirm — skill is disable-model-invocation)
↓
[/zama-frontend] ← wire the UI
↓
── pause ── "Ready to test? Open the frontend with `pnpm dev`."
↓
DONE
You are running /zama-autonomous. Follow this exact sequence. Never skip a gate.
Check for an existing run in .planning/v1-autonomous/state.json. If present:
{
"useCase": "private-voting",
"completedSteps": ["doctor", "design", "init"],
"currentStep": "contract",
"startedAt": "2026-05-04T12:00:00Z"
}
Ask the user: "Resume from <currentStep> or start fresh?" If resume, jump to the named step. If fresh, delete the state file and start from step 2.
If no state file exists, create one with completedSteps: [], currentStep: "doctor".
/zama-doctor inline)Run the diagnostic script directly:
bash ${CLAUDE_SKILL_DIR}/../doctor/scripts/diagnose.sh
If exit code is non-zero (required check failed), STOP. Tell the user:
Setup not ready. Run the suggested fix commands above, then re-run
/zama-autonomousto continue.
Update state: currentStep: "design". Save.
/zama-designTell the user explicitly:
Starting
/zama-design. I'll ask you a few questions about your use-case, then generateDESIGN.mdandUI-WIREFRAME.md.
Invoke the design skill (Claude Code will route automatically because the description matches). When it completes, read the generated .planning/v1-design/<slug>/DESIGN.md to confirm output is non-empty.
Update state: completedSteps: [..., "design"], currentStep: "init". Save.
GATE 1 — Design review:
Show the user the first ~30 lines of DESIGN.md, then ask via AskUserQuestion:
If "Edit": pause, tell user to edit and re-run with --resume.
If "Cancel": exit cleanly.
/zama-initTell the user:
Starting
/zama-init. Use-case will be auto-suggested from DESIGN.md.
Invoke init. Wait for completion (closing summary contains "Next: /zama-contract").
Update state: completedSteps: [..., "init"], currentStep: "contract". Save.
/zama-contractTell the user:
Starting
/zama-contract. Reading DESIGN.md for state schema and ACL plan.
Invoke contract. Wait for closing summary.
Update state: completedSteps: [..., "contract"], currentStep: "test". Save.
/zama-testTell the user:
Starting
/zama-test. Will generate mock + Sepolia tests for the new contract.
Invoke test. Run pnpm test after to confirm the mock tests pass. If they fail, STOP and tell user to inspect.
Update state: completedSteps: [..., "test"], currentStep: "audit". Save.
/zama-auditInvoke audit on packages/contracts/contracts/. Capture exit code.
GATE 2 — Audit verdict:
/zama-debug or /zama-contract --regenerate)" → Yes / Stop--resume." Save state with currentStep: "audit" (so resume re-runs audit)./zama-deploy — MANUAL confirmThis skill has disable-model-invocation: true. You CANNOT auto-invoke it. Tell the user explicitly:
Ready to deploy to Sepolia. Because deploy is irreversible (real network, real testnet ETH), Claude Code requires you to explicitly type
/zama-deployto invoke it. I will pause here.
Wait for user to invoke /zama-deploy themselves. After they do, watch for the closing summary ("✅ /zama-deploy complete — Sepolia"). Update state.
/zama-frontendTell the user:
Starting
/zama-frontend. Will wire fhe.ts, useDecrypted hook, and EncryptedInput to the deployed contract address.
Invoke frontend. Wait for closing summary.
Update state: completedSteps: [..., "frontend"], currentStep: "done". Save.
Print:
🎉 Pipeline complete.
Use-case: <useCase>
Contract: <address> (<etherscan-url>)
Frontend: run `pnpm dev` from packages/frontend, then open localhost:3000
Files of interest:
.planning/v1-design/<slug>/DESIGN.md
.planning/v1-design/<slug>/UI-WIREFRAME.md
packages/contracts/contracts/<name>.sol
packages/contracts/deployments/sepolia/<name>.json
packages/frontend/src/components/
Next: connect MetaMask (Sepolia network), interact, see encrypted balances decrypt.
Delete .planning/v1-autonomous/state.json (run is complete).
/zama-deploy — it's disable-model-invocation: true for a reason. The user must type the command themselves.package.json already exists in cwd, ask the user before continuing past /zama-init..planning/v1-autonomous/state.json and is the only file this skill writes outside of the downstream skills' outputs.npm publish or git push — those are out of scope./zama-debug — if a step fails, it asks the user to handle it.{{useCase}}{{stepsCompleted}}/{{stepsTotal}}{{startedAt}}{{completedAt}}{{contractAddress}}pnpm dev from packages/frontend → open http://localhost:3000{{designPath}}{{wireframePath}}{{contractPath}}{{deploymentPath}}/zama-debug.examples/confidential-token/VERCEL.md.npx claudepluginhub kocaemre/zama-skills --plugin zama-skillsProvides UI/UX resources: 50+ styles, color palettes, font pairings, guidelines, charts for web/mobile across React, Next.js, Vue, Svelte, Tailwind, React Native, Flutter. Aids planning, building, reviewing interfaces.
Fetches up-to-date documentation from Context7 for libraries and frameworks like React, Next.js, Prisma. Use for setup questions, API references, and code examples.