Arete documentation
Prompt Cookbook
For the complete documentation index optimized for AI agents, see llms.txt or llms-full.txt. A markdown version of this page is available by appending.mdto the URL or sendingAccept: text/markdown.
For AI agents: the documentation index is at llms.txt (full corpus: llms-full.txt). A markdown source for this page is /agent-skills/prompts.md.
Copy a prompt, replace the parts in [BRACKETS] with your own details, and
paste it into your agent. Each one describes the result you want and leaves the
installed skills to look up current packages, schemas, and endpoints.
These prompts assume setup is finished, meaning a4 init -y and
a4 doctor --json have both completed. If not, start with
Choose and Set Up an Agent.
Explore what is happening now
Section titled “Explore what is happening now”Use this when you are curious about live activity and don’t yet know which program or view holds the answer.
Use Arete to investigate current activity related to [TOPIC]. Search by intent, identify relevant programs and subscribe-ready stacks, inspect exact view schemas, and connect to the smallest useful view. Gather a bounded sample, answer the question with the program and view provenance, state any limitations, and disconnect. Do not invent a view when none is available.
Survey a protocol
Section titled “Survey a protocol”Use this to get the lay of the land for a protocol: which programs are involved, and what Arete can do with each today.
Use the Arete catalog and knowledge layer to survey [PROTOCOL OR INTENT]. Tell me which programs are involved and which capabilities are currently available for each: typed reads, transaction building, and live subscriptions. Inspect exact descriptors for the best candidates and distinguish catalog knowledge from installable or hosted delivery.
Inspect a program before integrating it
Section titled “Inspect a program before integrating it”Use this before you commit to a program, to see everything its SDK would give you.
Find the Arete program package for [PROGRAM OR PROTOCOL]. Inspect its exact descriptor and summarize its account readers, PDA helpers, raw instructions, higher-level operations, flows, SDK targets, authentication, and installation command. Use current generated or descriptor names; do not infer them from an IDL example.
Add a Program SDK
Section titled “Add a Program SDK”Use this when you know what you want to do and are ready for the agent to install the SDK and write code against it.
Find the read- and build-ready Arete program package that matches [INTENT]. Add it to this project for [TYPESCRIPT, RUST, OR PYTHON] using the descriptor’s install command. Preserve arete.toml and arete.lock, inspect the generated surface, and implement [READ OR OPERATION] using generated types. Do not edit generated files.
Prepare without submitting
Section titled “Prepare without submitting”Use this to see exactly what a transaction would do. Nothing gets signed or sent.
Using the installed Arete Program SDK, prepare and inspect a transaction for [ACTION]. Show the derived accounts, instructions, required signers, estimated fee, simulation result, and program error metadata. Do not request a wallet signature and do not submit anything.
Build a live application
Section titled “Build a live application”Use this to go from an idea to a working interface on top of live data.
Find a currently subscribe-ready Arete stack for [INTENT], inspect its exact views, and install it into this [FRAMEWORK] project. Build a UI using the generated SDK. Handle connecting, ready-empty, ready-with-data, refreshing, validation-error, disconnected, and retry states. Resolve all view and field names from generated types rather than copying an unrelated example.
Turn an MCP exploration into code
Section titled “Turn an MCP exploration into code”Use this after an exploration session, when you want your app to read the same data your agent just looked at.
We used Arete MCP to validate [STACK AND VIEW] for [PURPOSE]. Inspect the same exact stack descriptor, add it as a project dependency, generate the [LANGUAGE] SDK, and implement the equivalent query or subscription in application code. Confirm that the installed view identity and fields match the explored surface.
Identify a missing live view
Section titled “Identify a missing live view”Use this when you suspect the data you need isn’t available yet. You get a written spec of the gap, and nothing is deployed.
I need to answer [QUESTION]. Search the Arete catalog and inspect relevant programs and live views. If no deployed view provides the required fields and keys, write a concise read-model specification describing the programs, entities, primary keys, sources, fields, and query windows that a custom view would need. Do not deploy anything.
Diagnose an integration
Section titled “Diagnose an integration”Use this when something that used to work has stopped, and you want the cause found before anything gets changed.
Run a4 doctor —json, validate arete.toml and arete.lock, inspect the exact installed descriptors, and reproduce [PROBLEM] with the smallest read or bounded subscription. Separate configuration, authentication, descriptor, transport, schema-validation, and application-state failures. Do not update dependencies or regenerate until you have identified the cause.
Writing your own prompts
Section titled “Writing your own prompts”A good prompt covers four things: the result you want, the scope, the time window, and how much authority the agent has. Leave out technical details you would only be guessing at, since the catalog can resolve those. Do say explicitly whether the agent may only inspect, may edit code, may compile local artifacts, or may carry out an external deployment or transaction.