Feature: Pendle Stablecoin Yield Agent (Web AG-UI)

[vseb1]

Overview

• Add a Pendle-focused agent in web-ag-ui that discovers stablecoin pools, compares APY, and rebalances PT holdings to the best available yield.
• Follow the mock agent workflow structure as the primary scaffold; use agent-clmm patterns only for transaction planning/execution and runtime checks.
• Execute all Pendle actions through onchain-actions (local worktree preferred; production fallback if needed).

Goals

• Discover Pendle stablecoin PT yields and rebalance into the best yield when thresholds are met.
• Run autonomously after one-time setup (starting token + amount + delegations).
• Emit AG-UI artifacts and progress updates aligned to mock agent patterns.
• Include compounding gains and automatic Pendle rollovers in final scope.

Non-Goals

• Non-stablecoin pools or volatile assets.
• Leverage strategies or complex multi-hop yield optimization.
• New pricing/oracle systems.
• Custodial wallet management.

Functional Requirements

Pool discovery

• Fetch Pendle stablecoin pools from onchain-actions.
• Filter by stablecoin whitelist only (no liquidity/maturity filters).
• Normalize APY fields and rank consistently; deterministic ordering for ties.
• Fetch swappable tokens/pairs from onchain-actions and ignore pools that cannot be swapped to/from.

Portfolio assessment

• Identify current PT holdings and associated pools.
• Compare current APY vs best available APY.
• Rebalance only when a configured threshold is exceeded (default: 0.5% APY delta).

Rebalance planning

• Build plan to exit current PT and enter target PT.
• Keep planning details internal; do not surface plan artifacts to UI.
• Full scope: starting token → underlying → target PT.

Execution

• Execute plan via onchain-actions (discovery, swaps, PT acquisition).
• Track lifecycle with AG-UI events (planned → pending signature → submitted → confirmed/failed).
• On failure, surface a clear rollback/next-step message; no silent retries.
• Initial setup: starting token → highest-APY underlying → PT.
• Rebalance flow: old PT → old underlying → new underlying → new PT.

Workflow UX

• Only surface mock-agent-aligned artifacts: pendle-telemetry and pendle-summary.
• UI shows latest trade, performance accounting metrics, and activity indicators only.
• Do not expose pool lists, holdings, or rebalance plans.
• Delegations collected during setup; no additional approvals mid-cycle.

Technical Requirements

• Implement as a dedicated agent workflow inside web-ag-ui/apps, not by extending mock agent directly.
• Use mock agent command routing/state machine patterns.
• Use agent-clmm only for transaction execution flow, onchain confirmations, and wallet/network readiness checks.
• Validate all external API inputs/outputs at client boundaries with Zod.
• No mocks for missing env vars; local runs require explicit configuration.

Default Stablecoin Whitelist

• USDai, reUSD, NUSD, rUSD, yzUSD, ysUSDC, upUSDC, USD3, jrUSDe, iUSD, syrupUSDC, syrupUSDT, USDe

Dependencies

• onchain-actions service (local worktree preferred; production fallback allowed).
• Pendle pool discovery/analytics support in onchain-actions.
• web-ag-ui runtime + artifact rendering.

Acceptance Criteria

• Agent registered/discoverable in web-ag-ui and gated by feature flag.
• Pools filtered by whitelist; APY ranking is normalized and deterministic.
• Rebalance decisioning respects threshold and is internal-only.
• Execution uses onchain-actions and reports lifecycle events and telemetry/summary artifacts.
• Setup flow captures starting token/amount + delegations, then runs autonomously.
• Tests include unit coverage for filtering/ranking/decision logic and at least one integration test for discovery → decisioning.

Test Plan

• Unit tests: pool filtering, APY normalization/ranking, rebalance threshold logic.
• Integration test: mock pool list + holdings → internal execution decision logged.
• Optional E2E: full transaction flow with funded test wallet.

Rollout

• Implement behind feature flag in web-ag-ui.
• Validate locally against a local onchain-actions instance.
• Enable in staging after integration tests pass.