Imported from napindc/os_poly_bots (AGENTS.md). Install upstream with npx skills add napindc/os_poly_bots. Copyright stays with the author.

EVPOLY Agent Notes

Mandatory Update Rule

• If any strategy logic, checkpoint schedule, entry/exit behavior, risk gate, sizing, or strategy env defaults are changed, update strategy-changelog.md in the same task.
• Add a dated entry under ## Change Log describing:
  • what changed,
  • which strategy/timeframe/symbol it affects,
  • key env or code path touched.

Scope

Strategies include (minimum):

• premarket_v1
• endgame_sweep_v1
• evcurve_v1
• sessionband_v1
• evsnipe_v1
• mm_rewards_v1
• mm_sport_v1

Details: Low-Level AI Operating Guide

1) Mission

• Ship safe, minimal diffs.
• Preserve runtime behavior unless explicitly requested to change.
• Keep docs and env examples aligned with runtime.

2) Repo Map

• Main runtime: src/main.rs
• Premarket scheduler: src/strategy_decider.rs
• Strategy configs:
  • src/config.rs (endgame/global)
  • src/evcurve.rs
  • src/sessionband.rs
  • src/evsnipe.rs
  • src/mm/mod.rs
• Shared sizing policy: src/size_policy.rs
• Manual HTTP API: src/bin/manual_bot.rs
• Alpha service binary: src/bin/alpha_service.rs
• Env templates: .env.example, .env.full.example
• Strategy behavior history: strategy-changelog.md

3) Current Runtime Defaults

• Strategy toggles:
  • ON: premarket, endgame, evcurve, sessionband, evsnipe
  • OFF: mm_rewards, mm_sport
• Default symbols:
  • premarket/sessionband: BTC, ETH, SOL, XRP
  • endgame/evcurve/evsnipe: BTC, ETH, SOL, XRP, DOGE, BNB, HYPE
• MM mode default: auto

4) Known Combo Constraints

• mm_rewards_v1 should run alone, except optional pairing with mm_sport_v1.
• endgame_sweep_v1 and sessionband_v1 should not run together.
• See docs/strategy_combos.md.

Runtime Control Rule

• Use ./ev for runtime lifecycle (start/restart/stop/status/logs/autorestart) and do not run the live main bot using direct cargo run.

5) Remote Dependency Facts

• Shared timeframe discovery: remote-first with local fallback.
• EVSnipe discovery: remote-first (alpha host failover) with local fallback.
• Premarket alpha:
  • remote-first,
  • transport/timeout/connect errors fallback to local 50/50,
  • other alpha errors fail-closed.
• Endgame/EVcurve/SessionBand alpha:
  • remote decision path,
  • unavailable/invalid responses skip decision (strategy-specific skip behavior).
• MM rewards:
  • remote alpha for selection only,
  • preflight is local in bot runtime,
  • market discovery path is local.

6) Required Work Sequence for Any Task

1. Run git status --short.
2. Inspect target code paths with rg and focused reads.
3. Implement only requested changes.
4. If strategy behavior/defaults changed, update strategy-changelog.md.
5. Run local CI-equivalent checks before push, in this order:
   • cargo fmt --all -- --check
   • cargo check --all-targets
   • cargo test --all-targets --quiet
   • ./scripts/security_audit.sh
6. Re-check git status --short and diff.
7. Commit with clear scope message.

7) Editing Rules

• Do not revert unrelated user changes.
• Do not use destructive git commands.
• Keep comments concise and useful.
• Prefer consistency with existing architecture.

8) Manual Endpoint Facts

• Binary: manual_bot
• Auth headers:
  • x-evpoly-manual-token
  • x-evpoly-admin-token
• Open/close semantics:
  • /manual/open is BUY path
  • /manual/close is reduce-only SELL path
  • tp_price allowed for open only
• Full route guide: docs/manual_endpoint_guide.md

9) Definition of Done

• Requested changes implemented.
• Full local CI-equivalent suite passes.
• Docs updated when behavior/surface changed.
• Commit and push completed.