compose-nav-oracle-setup

name

compose-nav-oracle-setup

description

Configure and deploy this Compose nav-oracle example under the user's own Goldsky account. A cron task (default every 5 min) fetches a NAV bundle from a custodian endpoint and publishes it to ReserveAggregator contracts on Base Sepolia and Arbitrum Sepolia. Walks a new user through CLI install, deploying two instances of ReserveAggregator.sol (one per chain) with the Compose-managed publisher wallet, wiring addresses into the task, swapping the custodian URL, optionally publishing to a new GitHub repo, and the final deploy + log-tailing smoke test. Use when a user has just cloned this example or asks to set up / deploy / configure the nav-oracle app.

Setup: Compose nav-oracle

This skill makes the cloned nav-oracle example runnable under the user's own Goldsky account. On a cron (default every 5 minutes) the app fetches a NAV (net asset value) bundle from a custodian HTTP endpoint, scales the values to 18-decimal fixed-point, and writes them to ReserveAggregator contracts on two chains in parallel via a single Compose-managed wallet.

Assume the user has never used Goldsky Compose before. Do not skip preflight.

Non-negotiables

• Never run forge create, goldsky compose deploy, git push, or gh repo create without showing the exact command first and getting explicit confirmation.
• The publisher constructor arg on ReserveAggregator must be the Compose-managed wallet address. Anything else reverts every write with OnlyPublisher(). This wallet is created by Compose on first deploy — you cannot pick it; you must deploy Compose first to find out what it is.
• The SALT-like detail here is the custodian's JSON schema. If the custodian response doesn't match the CustodianResponse interface in src/tasks/nav-oracle.ts exactly (accountName, asOf, cash, tbills, repo, totalNav, ripcord), publishes silently drop or throw on parse.
• Do not touch src/lib/scaling.ts. The 18-decimal scaling is coupled to the contract's decimals() return value (also hardcoded at 18).

Variable handling for agents

When this skill says $FOO, capture the literal value from the prior command's output and substitute it directly into the next command. Do not rely on shell variables persisting between separate Bash tool invocations — each invocation gets a fresh shell with no env carryover from earlier commands.

Preflight

1. goldsky CLI — run goldsky --version. If missing, install per https://docs.goldsky.com/reference/cli.
2. goldsky authenticated — run goldsky project list. If it errors, stop and tell the user: "Please run goldsky login in your terminal — browser flow. Tell me to continue when you see the success message." Do not spawn goldsky login from Bash — it needs an interactive browser. Alternative: the user passes --token <token> on each command.
3. deno — run deno --version. Install with curl -fsSL https://deno.land/install.sh | sh if missing.
4. foundry — run forge --version. Contract deployment needs it. Install: curl -L https://foundry.paradigm.xyz | bash && foundryup.

Step 1 — Configuration interview

1. "App name?" (default: nav-oracle) — becomes the top-level name: field in compose.yaml and the deploy URL path.
2. "Which two chains?" (default: baseSepolia + arbitrumSepolia) — this example is hardcoded to publish to two chains. If the user wants a single chain or different chains, note that in Step 3 you'll also need to edit the two BASE_SEPOLIA_AGGREGATOR/ARBITRUM_SEPOLIA_AGGREGATOR constants and the corresponding wallet.writeContract(evm.chains.*, ...) calls in the Promise.allSettled([...]) block of src/tasks/nav-oracle.ts.
3. "Custodian endpoint URL?" — the HTTP endpoint that serves the NAV JSON. Default is the mock at https://raw.githubusercontent.com/goldsky-io/documentation-examples/main/compose/nav-oracle/mock-custodian.json. If the user is just demoing, leave it; otherwise ask for their own URL.
4. "Feed description string for each chain?" — human-readable label stored on-chain (e.g. "Example RWA Fund I NAV / USD").
5. "How often should the cron run?" (default: */5 * * * *) — custodian-dependent. Hourly (0 * * * *) is common for real PoR feeds.
6. "Publish to a new GitHub repo?" — optional.

Step 2 — Provision the publisher wallet

The publisher wallet is named nav-oracle-publisher (matches the name: in the evm.wallet({ name: "nav-oracle-publisher", sponsorGas: true }) call inside src/tasks/nav-oracle.ts). Provision it and print its address without needing to deploy first:

goldsky compose wallet create nav-oracle-publisher

Save the printed address — call it $PUBLISHER.

Note: you do not need to fund this wallet. The task passes sponsorGas: true to evm.wallet({...}) in src/tasks/nav-oracle.ts, so Goldsky covers gas on both chains.

Step 3 — Deploy ReserveAggregator on both chains

Output these two commands. Don't execute — the user runs them with their own funded EOA on each chain (they only need gas for the deploy tx, not for runtime).

# Base Sepolia
forge create contracts/ReserveAggregator.sol:ReserveAggregator \
  --rpc-url https://sepolia.base.org \
  --private-key $PRIVATE_KEY \
  --broadcast \
  --constructor-args $PUBLISHER "<feed description>"

# Arbitrum Sepolia
forge create contracts/ReserveAggregator.sol:ReserveAggregator \
  --rpc-url https://sepolia-rollup.arbitrum.io/rpc \
  --private-key $PRIVATE_KEY \
  --broadcast \
  --constructor-args $PUBLISHER "<feed description>"

Each prints Deployed to: 0x.... Capture both — call them $BASE_AGG and $ARB_AGG.

Step 4 — Wire aggregator addresses into the task

Edit src/tasks/nav-oracle.ts — use grep anchors, line numbers will drift over time:

• Find const BASE_SEPOLIA_AGGREGATOR and replace its address with "$BASE_AGG"
• Find const ARBITRUM_SEPOLIA_AGGREGATOR and replace its address with "$ARB_AGG"

If the user picked a custom custodian URL in Step 1:

• Replace the CUSTODIAN_URL string near the top of the file with theirs.

If the user picked different chains in Step 1, also edit:

• The two wallet.writeContract(evm.chains.baseSepolia, ...) and wallet.writeContract(evm.chains.arbitrumSepolia, ...) calls inside the Promise.allSettled([...]) block — swap chains and (optionally) rename the constants to match.

If the user wants a custom cron cadence, edit the expression: under the cron trigger in compose.yaml.

Step 5 — Optional: publish to a new GitHub repo

git init
git add .
git ls-files --cached | grep -iE '(keypair\.json|\.env|private[._-]?key|\.pem|id_rsa)' && \
  { echo "ABORT: secret-shaped file staged"; exit 1; }
git commit -m "Initial commit: Compose nav-oracle"
gh repo create <user's repo name> --<public|private> --source=. --push

Step 6 — Redeploy Compose

goldsky compose deploy

Step 7 — Smoke test

Tail logs and wait for the next cron fire (up to 5 min):

goldsky compose logs

Good output:

Published <accountName> NAV=$<totalNav> — base:ok, arb:ok

Verify on chain explorers — each aggregator contract should show a NavUpdated event:

• https://sepolia.basescan.org/address/$BASE_AGG#events
• https://sepolia.arbiscan.io/address/$ARB_AGG#events

Sanity-test the ripcord: if the user controls the custodian endpoint, flip "ripcord": true in the response JSON. Next cron fire should log Ripcord engaged … and skip cleanly (not an error).

Troubleshooting

• Edits to compose.yaml or source files don't take effect after redeploy. The local .compose/ bundle cache is stale. Run rm -rf .compose/ and redeploy.
• OnlyPublisher() revert. $PUBLISHER you wired into the forge constructor doesn't match the Compose wallet. Only the address currently holding the publisher role can rotate it (setPublisher reverts unless msg.sender == publisher). If the wrong address went in at constructor time, redeploy is the only recovery path — the deployer EOA has no privileged role on the contract.
• Task keeps logging skipped: unconfigured. One of BASE_SEPOLIA_AGGREGATOR / ARBITRUM_SEPOLIA_AGGREGATOR is still the zero address. Re-check src/tasks/nav-oracle.ts lines 22–23.
• Custodian fetch fails with JSON parse error. The custodian response does not match the CustodianResponse shape. All seven fields are required; cash/tbills/repo/totalNav must be JSON numbers (not strings), asOf must be ISO 8601, ripcord must be boolean.
• Both baseSepolia and arbitrumSepolia writes fail. The task throws and the retry config kicks in (2 attempts, 2s/4s backoff). If it persists, check https://sepolia.basescan.org/address/$BASE_AGG for recent state — the contract may be self-destructed, paused by a bad setPublisher, or on the wrong chain.
• One chain's write succeeds, the other fails. This is tolerated by design (Promise.allSettled). Next cron cycle reconciles. Only worry if failures are persistent on one side — likely an RPC or aggregator-specific issue.

What you should NOT do

• Keep sponsorGas: true on the evm.wallet({...}) call. Managed (Privy) wallets default to true already; the explicit setting is for clarity. Do not flip it to false — the publisher would then need native gas on both chains.
• Do not change toScaled18 in src/lib/scaling.ts or the decimals() return on the contract. They're coupled.
• Do not add historical round support to ReserveAggregator.sol as part of setup — the README notes this is a demo simplification. If the user asks for it, that's a separate change, not setup.
• Do not add a PRIVATE_KEY secret. The publisher is a Compose-managed wallet with gas sponsorship; the user's EOA is only used once, for forge create.