[new-plugin] otto-kol-follow v0.1.1

[Vb6Z]

Plugin: otto-kol-follow v0.1.1

Resubmitting after Google Form submission (https://forms.gle/gFCef1Y4qCv49L2z7). Original PR (#368) was bulk-closed in the procedural sweep on 2026-04-28; this is the v0.1.1 follow-up.

What this Plugin does

Otto KOL Follow mirrors aggregated sentiment from the top 50 crypto KOLs on X into a Hyperliquid perp. Trades only when the cohort hits statistically meaningful consensus (sample size + confidence + direction gates).

Required files

• .claude-plugin/plugin.json ✓
• plugin.yaml ✓ (with dependent_plugin / risk_level: advanced / supported_venues per FOR-DEVELOPERS.md "Submitting Strategy Plugins")
• SKILL.md ✓ (Overview / Pre-flight Checks / Commands / Error Handling / Security Notices)
• SUMMARY.md ✓ (English Overview / Prerequisites / Quick Start)
• LICENSE ✓ (MIT)
• scripts/bot.py + scripts/config.py — optional autonomous poller; SKILL.md reactive flow is primary

Advanced-tier compliance

• ✅ Dry-run default (DRY_RUN = True in config.py)
• ✅ Stop-loss mechanism (SL_PCT = 0.02)
• ✅ Maximum amount limits (MAX_SIZE_USD, MAX_POSITION_PCT_EQUITY)
• ✅ Risk disclaimer in SKILL.md and config.py
• ✅ --strategy-id otto-kol-follow on every write op (autonomous bot.py and reactive SKILL.md flows)

Stress test

All 3 Otto strategy plugins exercised end-to-end against the live Otto signal feed at https://signals.useotto.xyz with real hyperliquid-plugin order and polymarket-plugin list-markets previews. Decision-logic harness lives at services/otto-skills/stress_test.py in the source monorepo. Production telemetry runner shipped at services/otto-skills-stress/ (Phase 1: dry-run logging; Phase 2: live trading gated behind STRESS_LIVE_MODE env var).

Threshold tunes (v0.1.1)

Calibrated against observed live data 2026-04-30:

• MIN_KOL_COUNT: 40 → 25 (real distinct-author counts: 30–40)
• MIN_CONFIDENCE_KOL: 0.70 → 0.50 (sentiment_score ≥75 OR ≤25 = real consensus)
• MIN_SCORE: 0.65 → 0.50 (token-alpha confidence range observed)
• MAX_SIGNAL_AGE_SEC: 900 / 600 → 4500 (covers full hourly producer cadence)
• MIN_EDGE_PCT (mispricing only): 0.08 → 0.05

Happy to address any review feedback. Ping us at @useOttoAI on X or via the email on the Google Form.

[Vb6Z]

Thanks for the review window. We've pushed a v0.1.2 hardening update across the three Otto Skills based on findings from our internal stress-testing infrastructure (24h dry-run telemetry against live signals on Railway).

Key changes — all backward-compatible, no breaking interface changes:

• Per-coin cooldown (4h on alpha-sniper, 6h on kol-follow) prevents repeated fires on the same asset within a short window
• Signal freshness gate added to alpha-sniper (kol-follow already had one) — stale producer feeds no longer fire trades
• Tightened sizing semantics — size_usd is now unambiguously the position notional, with margin = notional / leverage. Comment + code are consistent
• Polymarket failure modes categorized (geoblock / auth / plugin-missing / timeout) instead of generic "unreachable"
• Cleaner abort taxonomy in our telemetry pipeline so signal-quality issues vs execution issues are easy to distinguish

Same SKILL.md interface. Same strategy modes (trending / kol-follow / funding-fade / mispricing). Just safer defaults and clearer telemetry.

Happy to walk through any specific change if helpful.

[Vb6Z]

v0.1.3 hardening update — second pass after additional internal review surfaced more opportunities for safety improvements. All v0.1.2 backward-compatible, no interface changes.

Consumer-facing changes:

• TP/SL bracket no longer passes --strategy-id (leaderboard attribution lives on the parent order; tpsl is per-position)
• Loud warning log when bracket fails — position is OPEN without TP/SL, manual intervention required so operators see it immediately in their stdout
• SKILL.md size formula clarified: size_tokens = size_usd / mark_px — size_usd is the position notional; --leverage controls margin posted (margin = notional / leverage). Previous wording could be misread.

Internal stress-runner changes (not user-facing):

• Plugin response parsing now uses LAST JSON block (the authoritative success/failure record) rather than first (which is always a preview echo in --confirm mode)
• Bracket response positively validated — empty {} and unknown shapes now fail-closed
• Mark-price fail-hard: refuses to send a trade if mark unavailable (no more raw size_usd fallback)

Same SKILL.md protocol, same CLI commands, same strategy modes. Just safer defaults and clearer telemetry.