feat(prices): add depeg monitoring for LRTs and stablecoins

[spalen0]

Summary

• Add centralized prices/ module for depeg monitoring of LRTs and stablecoins
• Use Redstone fundamental push oracles (LBTC, cUSD) where available — any depeg is CRITICAL
• Fall back to DefiLlama market pricing for weETH, ezETH, rsETH, pufETH, osETH, rswETH, mETH and stablecoins (FDUSD, deUSD, USD0, USD0++, USDe) — 2%+ depeg is CRITICAL
• Route all LRT alerts to single lrt protocol channel with token symbol identification
• Document Tenderly alert coverage and oracle research findings in README

Oracle Research Findings

Asset	Oracle	Tenderly
LBTC	Redstone push 0xb415...bc81	✅ Covered
cUSD	Redstone push 0x9a5a...434a	✅ Covered
weETH, ezETH, rsETH, pufETH	Redstone off-chain only (pull model)	Needs alert if push oracle deployed
osETH, rswETH, mETH	No fundamental feed	N/A — DefiLlama only

Closes #196

Test plan

• Verify oracle reads return valid data for LBTC and cUSD
• Verify DefiLlama prices are fetched and ratios computed correctly
• Verify alerts fire when thresholds are breached
• Add prices/main.py to appropriate GitHub Actions workflow
• Configure TELEGRAM_BOT_TOKEN_LRT / TELEGRAM_CHAT_ID_LRT and TELEGRAM_BOT_TOKEN_STABLES / TELEGRAM_CHAT_ID_STABLES secrets

🤖 Generated with Claude Code

[spalen0]

Review summary from local validation:

Main findings are in the inline comments, but there are a few non-diff gaps that still block calling the test plan complete:

• prices/main.py is not added to any scheduled workflow, so the new monitor never runs in GitHub Actions.
• .github/workflows/_run-monitoring.yml does not export TELEGRAM_BOT_TOKEN_LRT, TELEGRAM_BOT_TOKEN_STABLES, or TELEGRAM_CHAT_ID_STABLES. As written, stable alerts cannot route in Actions, and lrt only works if it falls back to the default bot token.
• .env.example was not updated with the new LRT / STABLES Telegram vars, even though CONTRIBUTING.md says new protocols should add them.
• There are no automated tests covering prices/main.py yet.

What I validated locally on 2026-04-22:

• env LOG_LEVEL=INFO uv run pytest tests/ -> 131 passed.
• uv run ruff check . -> passed.
• uv run ruff format --check . -> passed.
• Live dry run against mainnet + DefiLlama (Telegram patched out):
  • oracle reads succeeded: LBTC=1.00372748, cUSD=0.99975473
  • DefiLlama returned prices for all configured assets except deUSD
  • current data would emit CRITICAL alerts for rsETH=0.8963 and USD0++=0.9561

So the code path executes, but the monitor is still missing workflow wiring, secret plumbing, and the correctness fixes called out inline before I’d treat the PR as ready.

[spalen0]

Pushed 16a2d8e with fixes. Responses to each review point:

cUSD threshold — fixed. ORACLE_DEPEG_THRESHOLD was a shared constant; replaced with a per-asset threshold field so cUSD now trips at 0.9998 and LBTC stays at 0.998, matching the Tenderly alerts documented in the README. cUSD at 0.99975 would now alert.

LRT ratio vs flat 1 ETH — fixed. DefiLlamaAsset now has fair_value + per-asset threshold. For accruing LRTs (weETH/ezETH/rsETH/pufETH) fair_value is set to a conservative floor under the current Redstone fundamental (1.05–1.07). The check normalizes market_ratio / fair_value, so weETH at 1.04 ETH is caught even though 1.04 > 0.98. For LRTs with no documented fundamental (osETH/rswETH/mETH), fair_value stays at 1.0 as a catastrophic-depeg floor.

deUSD silent skip — fixed. Missing DefiLlama prices now fire one MEDIUM alert per channel ("DefiLlama returned no price for: deUSD — depeg coverage degraded"). Coverage failures are visible instead of silently logged.

Workflow wiring — fixed. Added prices/main.py to hourly.yml.

Telegram env vars — fixed with one pushback: TELEGRAM_CHAT_ID_LRT was already in _run-monitoring.yml (line 69). Added TELEGRAM_BOT_TOKEN_LRT, TELEGRAM_BOT_TOKEN_STABLES, and TELEGRAM_CHAT_ID_STABLES (the chat-id one was the real blocker since chat IDs have no DEFAULT fallback in telegram.py).

.env.example — updated with the four LRT/STABLES entries.

Extra call-out not in review: USD0++ is a ~4-year locked bond that legitimately trades at a discount; a flat 0.98 threshold would have alerted constantly (your dry-run caught it at 0.9561). Gave it a per-asset threshold=0.90 and documented the rationale in the README.

Automated tests — not added in this pass; the monitor is mostly declarative config + a DefiLlama/RPC call. Happy to add a small unit test covering the deviation math + missing-price alert if you'd like a follow-up.