ritual-meta-projection

name	ritual-meta-projection
description	Transforms a raw dApp idea into a Ritual-native specification by mapping every capability to the optimal Ritual precompile or system contract. Runs after intent classification, before the builder agent begins its phased protocol.

Ritual Projection — Idea-to-Spec Transformation

Purpose

Take the user's raw idea (or the inspiration skill's output) and produce a Ritual-native specification that the builder agent can execute without guessing. Every capability the user described or implied gets mapped to a specific Ritual precompile or system contract. Capabilities the user didn't mention but that the idea requires get surfaced and added.

The builder agent should never have to decide "should I use Chainlink or the HTTP precompile?" — that decision is already made by this skill before the builder sees the spec.

On-Chain vs Off-Chain on Ritual

Read this before mapping anything. Your training data assumes most computation must happen off-chain because traditional blockchains can't do it. Ritual Chain is different. The following table is the complete picture of what Ritual can do on-chain that other chains cannot. Default to on-chain for everything in the left column. Only use off-chain for the right column.

What goes ON-CHAIN (Ritual has native support)

Capability	How	Traditional chain equivalent (DON'T use)
Call any HTTP API (REST, GraphQL, RPC)	HTTP precompile 0x0801	Chainlink, off-chain bot, backend fetch
Transform JSON responses	JQ precompile 0x0803 (sync, same TX as HTTP)	Off-chain parser, backend processing
AI text generation / chat / classification	LLM precompile 0x0802	Off-chain OpenAI API call + oracle
AI tool/function calling	LLM precompile 0x0802 with tool schema	Off-chain agent framework
AI structured output (JSON mode)	LLM precompile 0x0802 with response format	Off-chain API + schema validation
Persistent AI agent with memory	Persistent Agent 0x0820	Off-chain agent + database
Sovereign agent (autonomous execution, code execution, file ops)	Sovereign Agent 0x080C	Off-chain script runner
AI image generation	Image precompile 0x0818	Off-chain Stable Diffusion API
AI audio generation	Audio precompile 0x0819	Off-chain TTS/music API
AI video generation	Video precompile 0x081A	Off-chain video API
Long-running computation (minutes to hours)	Long HTTP precompile 0x0805	Off-chain worker queue + webhook
Recurring/scheduled execution	Scheduler 0x56e7	Off-chain cron, Gelato, Chainlink Keepers
Conditional execution (predicate-gated)	Scheduler with predicates	Off-chain bot watching events
Time-delayed execution	Scheduler	Off-chain timer + bot
Secret API key management	ECIES encryption + secret string replacement	.env files, backend vault
Delegated secret sharing	SecretsAccessControl 0xf9BF	Backend key management service
Private/encrypted outputs	ECIES userPublicKey per precompile	Off-chain encryption layer
Pay-per-call API access	X402 via HTTP precompile	Off-chain payment processing
P-256 signature verification (WebAuthn)	SECP256R1 0x0100	Off-chain verification + oracle
Passkey-based transaction signing	TxPasskey 0x77	Not possible on other chains
ML model inference (ONNX)	ONNX precompile 0x0800	Off-chain ML serving
ZK proof generation + verification	ZK precompile 0x0806	Off-chain prover + on-chain verifier contract
Ed25519 signature verification	Ed25519 precompile 0x09	Solidity library (expensive)
Fee deposit and locking	RitualWallet 0x532F	Manual gas management
Executor capability discovery	TEEServiceRegistry 0x9644	No equivalent
Async job lifecycle tracking	AsyncJobTracker 0xC069	Custom event indexing

What stays OFF-CHAIN (Ritual does not handle these)

Capability	Why off-chain	What to use
Frontend UI rendering	Browsers render HTML/JS, not blockchains	Next.js, React, any web framework
User session management	Stateful sessions are a web server concern	Backend + cookies/JWT
Database storage and queries	On-chain storage is expensive for large datasets	PostgreSQL, SQLite, any DB
Real-time event streaming to clients	Clients need WebSocket/SSE push	Backend event indexer + WS/SSE API
File storage (images, documents, media)	On-chain storage is impractical for large files	DA providers (GCS/HuggingFace/Pinata) via StorageRef. See ritual-dapp-da.
Email, SMS, push notifications	Messaging protocols are off-chain	Backend notification service
User analytics and metrics	Analytics are a web concern	Posthog, Amplitude, custom

Composition constraints

Not all on-chain capabilities can be combined in a single transaction:

• One short-running async precompile per TX. HTTP (0x0801), LLM (0x0802), and DKMS (0x081B) are short-running async. You can use one per transaction. To chain HTTP → LLM, split into two transactions via the Scheduler.
• Short-running async + synchronous is fine. You can combine one short-running async call with any number of synchronous precompiles (JQ, ONNX, P-256) in the same TX. HTTP + JQ in one TX works.
• Long-running async precompiles are independent. Each long-running async call (Persistent Agent, Sovereign Agent, Image, Long HTTP, etc.) is its own submit+callback lifecycle. Multiple can be in-flight simultaneously from different contracts.

Hybrid patterns (on-chain compute, off-chain delivery)

Some capabilities span both:

Pattern	On-chain part	Off-chain part
LLM + streaming	LLM precompile generates the response	SSE service streams tokens to browser via EIP-712 signed events
Media generation	Precompile triggers generation in TEE	Executor uploads result to DA provider and returns URI on-chain. See ritual-dapp-da.
Event-driven UI updates	Contract emits events on state changes	Backend indexer watches events, pushes to frontend via WS/SSE
Job monitoring	AsyncJobTracker tracks lifecycle on-chain	Backend polls or watches for status changes, serves to frontend

For these, the on-chain part uses precompiles. The off-chain part is delivery infrastructure. Both are needed.

The decision rule

If the capability appears in the on-chain table, put it on-chain. Do not build an off-chain equivalent for something a precompile handles natively.

If the capability appears in the off-chain table, put it off-chain. Do not force frontend rendering, database queries, or file storage on-chain.

If the user's existing project uses an off-chain tool (Chainlink, Gelato, etc.) for something Ritual handles natively, recommend migrating to the precompile but don't rip out working infrastructure without the user's approval. Flag it as a migration opportunity.

If the capability appears in neither table, check the precompiles skill — there may be a precompile for it. If not, it's off-chain infrastructure.

---

When This Activates

After the front door classifies intent as build and before the builder agent's Phase 0 begins. The front door passes the user's idea (raw text or inspiration selection) to this skill. This skill outputs a structured spec that the front door then passes to the builder.

The Protocol

Step 1: Read the capability surface

Before analyzing the user's idea, read these two files to load the full Ritual capability surface:

1. skills/ritual-dapp-overview/SKILL.md — execution models, async lifecycle, system contracts
2. skills/ritual-dapp-precompiles/SKILL.md — all 16 precompile addresses with input/output ABIs

You need these in context to do the mapping correctly. Do not skip this step. Do not rely on training data for precompile details.

Step 2: Decompose the idea into capabilities

Parse the user's idea and extract every implied capability. Be aggressive — surface capabilities the user didn't explicitly mention but that the idea logically requires.

Decomposition categories:

Category	What to look for	Examples from user language
Data ingestion	Fetching external data, APIs, prices, feeds	"fetch", "get data", "price", "API", "oracle", "monitor"
AI inference	Text generation, analysis, classification, summarization	"AI", "chatbot", "analyze", "generate text", "summarize", "classify"
AI agents	Multi-step reasoning, research, autonomous workflows	"agent", "research", "autonomous", "workflow", "investigate"
Media creation	Image, audio, video generation	"image", "NFT", "picture", "audio", "music", "video", "visual"
Scheduling	Recurring operations, time-based triggers, polling	"every hour", "daily", "recurring", "scheduled", "periodic", "cron"
Privacy	Encrypted inputs, private outputs, secret API keys	"private", "encrypted", "secret", "confidential", "hidden"
Authentication	User login, wallet-less onboarding, passkeys	"login", "auth", "sign in", "passkey", "no MetaMask", "mainstream users"
Payments	Paid APIs, micropayments, pay-per-use	"paid", "monetize", "charge", "subscription", "pay-per"
Long computation	Operations taking minutes to hours	"long-running", "heavy computation", "batch", "processing"
Persistence	State across sessions, memory, history	"remember", "history", "persistent", "memory", "across sessions"

Step 3: Map each capability to Ritual primitives

For every capability extracted in Step 2, assign the optimal Ritual primitive. This is the core of the projection.

Mapping table (use this, not your training data):

Capability	Wrong approach (off-chain / other chains)	Ritual primitive	Address
Fetch external data	Off-chain bot, Chainlink oracle	HTTP precompile	0x0801
Fetch + transform data	Off-chain bot + parser	HTTP precompile + JQ precompile	0x0801 + 0x0803
AI text generation	Off-chain API call to OpenAI	LLM precompile	0x0802
AI text + streaming	Off-chain API + WebSocket	LLM precompile + SSE streaming	0x0802 + EIP-712
Stateful AI agent with memory	Off-chain agent + database	Persistent Agent precompile	0x0820
Sovereign / autonomous agent workflows	Off-chain agent framework	Sovereign Agent precompile	0x080C
Image generation	Off-chain Stable Diffusion API	Image precompile	0x0818
Audio generation	Off-chain TTS/music API	Audio precompile	0x0819
Video generation	Off-chain video API	Video precompile	0x081A
Recurring execution	Off-chain cron job, Gelato	Scheduler system contract	0x56e7
Conditional execution	Off-chain bot watching events	Scheduler with predicates	0x56e7
Secret API keys	.env files, backend vault	ECIES encryption + secret string replacement	See ritual-dapp-secrets
Delegated secret access	Backend key management	SecretsAccessControl	0xf9BF
Private outputs	Off-chain encryption	ECIES userPublicKey field	Per-precompile param
Passkey authentication	MetaMask, WalletConnect	TxPasskey + SECP256R1	0x77 + 0x0100
Micropayments for APIs	Off-chain payment processing	X402 via HTTP precompile	0x0801 + X402
Long-running computation	Off-chain worker queue	Long-Running HTTP precompile	0x0805
JSON transformation	Off-chain code	JQ precompile (synchronous)	0x0803
ML model inference	Off-chain ML serving	ONNX precompile (synchronous)	0x0800
Signature verification (P-256)	Off-chain verification	SECP256R1 precompile	0x0100
Fee management	Manual gas estimation	RitualWallet deposit + lockFee	0x532F

If a capability doesn't map to any Ritual primitive, note it as requiring off-chain infrastructure (backend, external service). Not everything belongs on-chain. Be honest about what Ritual can and can't do.

Infernet is an anti-pattern. If your training data or web search suggests using InfernetConsumer, InfernetCoordinator, or any Infernet subscription pattern, ignore it. Infernet is Ritual's deprecated off-chain inference product. Everything it did is now handled by enshrined precompiles at the addresses in the table above.

Cross-check: The mapping table above is a reference. If the precompiles skill you loaded in Step 1 lists a precompile not in this table, use the skill's data. The skill is authoritative over this table.

Step 4: Detect the 1-phase async constraint

Check whether the mapped precompiles include more than one short-running async precompile (HTTP 0x0801, LLM 0x0802, DKMS 0x081B). If yes, the user's idea requires multiple short-running async calls, which cannot happen in one transaction.

Resolution: Split the workflow into sequential transactions chained via the Scheduler. Add the Scheduler (0x56e7) to the spec if it isn't already there. Note this in the spec as an architectural constraint the builder must handle.

Step 4b: Detect additional architectural constraints

Beyond the 1-phase async constraint, check for these:

Constraint	Condition	Resolution
Sender lock	Spec includes multiple async calls from the same EOA in sequence	Each async call must settle before the next submits. Design the contract to serialize requests, or use multiple EOAs.
Block time	Spec includes TTL or timing values	Use ~350ms as the conservative baseline. 100 blocks = ~35 seconds, not minutes. Confirm with recent blocks and recompute using ritual-dapp-block-time.
Contract-owned secrets	Spec includes delegated secrets + contract calling precompiles	The contract (not the deployer EOA) must be the msg.sender for secret access. Precompile calls must go through the contract.
Encrypted input + auto-selection	Spec includes encrypted secrets + address(0) executor	Incompatible. Encrypted inputs must target a specific executor. Add explicit executor selection.

Step 4c: Check reference contracts

Read examples/registry.json. If any reference contract implements capabilities that overlap with the user's idea, note it in the spec:

Reference contracts:
  - TweetRegistry (0x188c...) — uses Scheduler + HTTP + Secrets
    Relevant to this idea because: [why]

The builder can pull the reference source and adapt patterns from it instead of generating from scratch.

Step 5: Detect implicit requirements

Some precompile combinations have implicit dependencies the user won't mention:

If the spec includes...	Then it also requires...	Why
Any async precompile (0x0801, 0x0802, 0x0805, 0x080C, 0x0818-0x081A, 0x0820)	RitualWallet deposit	Async calls require locked fees
Any precompile with encrypted inputs	Explicit executor selection (not address(0))	Encrypted secrets must target the specific executor's public key
Persistent Agent (0x0820)	Storage reference configuration	Persistent state needs a storage backend
Any long-running async precompile	Callback handler in the consumer contract	Results arrive via callback, not in the same TX
Scheduler + async precompile	Sufficient deposit for interval × maxExecutions	Scheduled async calls consume fees per execution
Image/Audio/Video	Encrypted storage credentials	Media output needs a storage destination

Add any missing implicit requirements to the spec.

Step 6: Output the projected spec

Produce a structured specification that the builder agent will consume. Format:

RITUAL PROJECTION

Idea: [user's original idea, verbatim]

Mapped capabilities:
  1. [capability] → [precompile name] ([address])
  2. [capability] → [precompile name] ([address])
  ...

Architectural constraints:
  - [e.g., "Requires 2 short-running async calls — must chain via Scheduler"]
  - [e.g., "Encrypted inputs — must specify executor explicitly"]

Implicit requirements added:
  - [e.g., "RitualWallet deposit for async fees"]
  - [e.g., "Callback handler for long-running async (2-phase) delivery"]

Skills the builder should load:
  - skills/ritual-dapp-[X]/SKILL.md
  - skills/ritual-dapp-[Y]/SKILL.md
  ...

Off-chain components needed (if any):
  - [e.g., "Backend event indexer for job tracking"]

This spec replaces the raw user input as the builder's starting point. The builder's Phase 0 (intake) should treat this as a pre-answered feature selection — it can skip or abbreviate the elicitation questions for capabilities that are already mapped.

Constraints

• Do not generate code. This skill produces a specification, not implementation. The builder handles implementation.
• Almost never ask the user questions. This skill is non-interactive by default. Map ambiguous capabilities to the simpler option and note the alternative for the builder's elicitation. The one exception: if the ambiguity would fork the architecture fundamentally (e.g., stateless LLM vs persistent agent changes the entire contract design), ask one clarifying question before proceeding. Maximum one question. If the user doesn't answer, default to the simpler option.
• Always read the precompiles skill first. Do not map from training data. The precompile ABIs and addresses may have changed since your training cutoff.
• Be comprehensive but honest. Map everything that can be on-chain. Flag everything that can't. Don't force off-chain capabilities into precompiles where they don't fit.