--- name: reflexivity-trader type: strategist description: > Soros-style reflexive position trader. Heavily composes the reflexivity-sentiment director to identify adoption-validation feedback loops in real-time narratives (AI capex, obesity drugs, etc.), then sizes by the director's reported conviction. Weekly cadence, ≤30 day holds. Use when user asks "is X a reflexive setup", "narrative trade ideas", or "what does the reflexivity trader see". Default review-only. do_not_promote: true kill_reason: uncontracted_signal_max_reflexive_cohort kill_reason_long: "Load-bearing on reflexivity-sentiment, which is a router not a typed contract — JSON-mode prompts may fail 5-10% of ticks without defensive parsing. No published rule-based reflexivity system exists (Agent A: Confirmed absence). The allowlist is the most-reflexive-downside cohort (SMCI/MSTR/PLTR/COIN/etc.), so a misclassified validation→exhaustion turn is asymmetric loss. Per the v0.2 dive (see _shared/v0.2-improvement-spec.md). SKILL.md retained for analyst-mode thesis exploration; this strategist must not be promoted to mode: live." metadata: author: nirav version: "0.1" compatibility: Hermes Agent + Claude Code allowed-tools: Read Skill Bash WebSearch strategist: mode: review asset_class: equity account_lock: "619508153" time_window: "10:00-15:55 America/New_York" cron_hint: "0 15 * * 3" allowlist: [NVDA, AMD, AVGO, SMCI, ARM, TSLA, LLY, NVO, VST, CEG, TLN, COIN, MSTR, PLTR, META] max_position_pct: 33 max_concurrent_positions: 2 capability_requires: [] profile_compatibility: [autotrader] requires_directors: [reflexivity-sentiment] director_response_schema: required_fields: [symbol, direction, confidence, phase, invalidation] phases: [emergence, validation, acceleration, exhaustion, reversal] max_hold_days: 30 exit_on_phase: [exhaustion, reversal] --- # Reflexivity Trader Trades positions where price action *validates* the narrative, which in turn attracts more flows, which further validates — Soros's reflexive feedback. The hard part is detection. This strategist delegates detection entirely to the `reflexivity-sentiment` director and acts on its structured response. > Required: `_shared/intent-schema.md`. See `references/decision-rules.md` and `references/example-ticks.md`. The director's existing `reflexivity-sentiment/SKILL.md` and its sub-skills are the authoritative source for the reflexivity framework. ## Strategy summary - **Universe:** 15 narrative-heavy large/mid-caps spanning AI, obesity, AI capex, crypto-equity, defense compute. - **Signal source:** the `reflexivity-sentiment` director's structured response (see below). - **Entry:** director identifies a `validation` or `acceleration` phase with confidence ≥ 0.5 on a symbol in allowlist. - **Exit:** director returns `exhaustion` or `reversal` phase on the held symbol, OR 30-day hold cap, OR position drops below the director's stated `invalidation` price. - **Sizing:** `notional = buying_power * 0.33 * confidence`. - **Cadence:** Weekly, Wednesday 11am ET. ## 6-phase protocol ### Phase 1 — Pre-flight Read state file. If a previous tick this week has already evaluated → skip. ### Phase 2 — Compose context (the load-bearing phase) Invoke `reflexivity-sentiment` with this exact prompt: > "Identify the single highest-conviction reflexive setup currently active in equities. Respond ONLY with a JSON object matching this schema: > > ```json > { > "symbol": "", > "direction": "long|short", > "phase": "emergence|validation|acceleration|exhaustion|reversal", > "confidence": , > "invalidation": {"price": , "reason": ""}, > "narrative": "" > } > ``` > > If no setup meets ≥0.5 confidence, respond with `{"symbol": null}`. > > Also check the held positions: `{state.open_positions}`. For each, report current `phase`. Append as `{held_status: [{symbol, current_phase}]}`." Parse the response. If `symbol` is null → skip phase 4 entries. ### Phase 3 — Signal gather For the director's symbol (if any) + all held symbols: - `get_equity_quotes(symbols=[...])` — current prices and prior closes. - `get_equity_fundamentals(symbols=[...])` — market cap, P/E (for sanity, not for entry). ### Phase 4 — Decide **Exits** (held positions): ``` for pos in open_positions: held_phase = director.held_status[pos.symbol] current_price = quotes[pos.symbol].last_trade_price age_days = today - pos.opened_at if held_phase in exit_on_phase: emit sell; reason: f"phase_{held_phase}" elif age_days >= max_hold_days: emit sell; reason: "max_hold_days" elif current_price <= pos.invalidation_price (long) or >= invalidation (short): emit sell; reason: "invalidation_hit" ``` **Entries** (director identified a setup): ``` if director.symbol and director.confidence >= 0.5 and director.phase in {"validation", "acceleration"}: if director.symbol in allowlist and not held: target_notional = buying_power * 0.33 * director.confidence qty = floor(target_notional / quote.last_trade_price) if qty == 0: skip if director.direction == "short": skip (cash account, no shorts) emit buy intent (qty, limit_price=current*1.002, reason: director.narrative) record director.invalidation.price in state ``` ### Phase 5 — Hand off to executor ### Phase 6 — Emit & persist State file records `open_positions` with the director's stated `invalidation_price`, `narrative`, `entry_phase`. Future exit logic uses these. ## Interactive mode - "What does the reflexivity trader see?" — run phases 1-4 in review mode, surface the director's full response in `notes`. - "Is NVDA a reflexive setup?" — invoke `reflexivity-sentiment` directly with NVDA-specific prompt; report director's view. ## What this strategist will never do - Short. Cash account. - Trade without a director call. The signal source is the director, period. - Override director invalidation. If price hits invalidation, exit. No "but the story is still good." - Trade outside the 15-symbol allowlist (director may flag others; strategist ignores them). - Hold past 30 days. Reflexive themes have shelf life. - Pyramid — single entry per symbol per cycle.