/edge-candidate-agent

Ideation Loop (Human + LLM)

Use this reference when generating hints before running auto detection.

Hint File Schema (hints.yaml)

hints:
  - title: "AI leaders breaking out after shallow pullback"
    observation: "Large-cap semis recovered above key moving averages"
    preferred_entry_family: "pivot_breakout"
    symbols: ["NVDA", "AVGO", "SMCI"]
    regime_bias: "RiskOn"
    mechanism_tag: "behavior"

LLM CLI Integration Contract

auto_detect_candidates.py --llm-ideas-cmd "<command>" passes JSON to stdin:

{
  "as_of": "YYYY-MM-DD",
  "market_summary": {...},
  "anomalies": [...],
  "instruction": "Generate concise edge hints ..."
}

The command must print either:

• [{...}, {...}] (YAML/JSON list), or
• {"hints": [{...}, {...}]}.

Each hint supports:

• title
• observation
• preferred_entry_family (pivot_breakout or gap_up_continuation)
• symbols (optional)
• regime_bias (optional)
• mechanism_tag (optional)

Suggested Daily Loop

1. Run detector without LLM hints and review anomalies.json.
2. Generate or edit hints.yaml from observed market behavior.
3. Re-run detector with --hints (and optionally --llm-ideas-cmd).
4. Compare exported ticket quality over time and keep only stable ideas.