analyze-signal

name	analyze-signal
description	Evaluates scout signals through a multi-factor scoring model, generates trade theses for executor
tags	["analysis","signals","trading","strategy","scoring"]
agent	analyst
var	${var} adjusts the analysis lens. If set, weight signals matching this narrative higher (e.g., "AI tokens momentum play"). If empty, score all signals equally.

analyze-signal

Priority: P0 (core analyst function — runs hourly + on new signals) Input: memory/mesh/analyst.json (signals from scout) Output: Trade-signals to executor inbox, feedback to scout inbox

Instructions

You are executing the analyze-signal skill for the Analyst agent.

Step 0: Check Halt Status

Read your inbox for any halt messages from Guardian. If an active halt exists (cooldown_until > now), exit immediately. Do not generate any trade-signals during a halt.

Exit code: ANALYSIS_HALTED

Step 1: Read Inbox

Read memory/mesh/analyst.json. Filter for messages where type = signal. Sort by timestamp (newest first). Process up to 20 signals per execution.

Step 2: Score Each Signal

Apply the appropriate scoring model based on signal_type:

Volume Spike / Surge (max score: 1.0):

Factor	Weight	Criteria
Magnitude	0.25	2-3x = 0.1, 3-5x = 0.2, 5x+ = 0.25
Organic	0.25	Check unique traders ratio: >100 unique traders in 1h = 0.25
Confirmation	0.25	Price also moving in same direction = 0.25
Narrative	0.25	Token has recent social/news catalyst (check memory/social-signals.json if exists)

Price Breakout (max score: 1.0):

Factor	Weight	Criteria
Resistance break	0.30	Above 7d high = 0.15, above 30d high = 0.30
Momentum	0.25	Accelerating (consecutive higher candles) = 0.25
Volume confirm	0.25	Volume above average during breakout = 0.25
Trend alignment	0.20	SOL also trending same direction = 0.20

New Pool (max score: 1.0):

Factor	Weight	Criteria
Contract safety	0.35	Mint authority renounced = 0.20, no freeze = 0.15
Liquidity depth	0.25	TVL/mcap ratio > 0.1 = 0.25
Team signal	0.20	Known deployer or social presence = 0.20
Early momentum	0.20	Price trending up in first 30 min = 0.20

Whale Move (max score: 1.0):

Factor	Weight	Criteria
Direction	0.30	Accumulation = 0.30, distribution = 0.05
Whale track record	0.30	>60% win rate (from memory/whale-track-record.json) = 0.30
Size significance	0.20	>1% of token supply = 0.20
Timing	0.20	Against the crowd (buying dip / selling pump) = 0.20

Step 3: Generate Trade Thesis (if score >= ${min_confidence})

Select best matching strategy from ${strategy}:

• momentum: Ride the trend. Entry: now. Stop-loss: -5 to -8%. Take-profit: +15 to +25%. Time horizon: 1-4h.
• mean-reversion: Fade the spike. Entry: wait for pullback to -3%. Stop-loss: -10%. Take-profit: +10%. Time horizon: 4-12h.
• narrative: Long-duration position on emerging story. Entry: now. Stop-loss: -15%. Take-profit: +50%. Time horizon: 1-7d.

Write trade-signal to memory/mesh/executor.json:

{
  "from": "analyst",
  "to": "executor",
  "type": "trade-signal",
  "id": "analyst-{timestamp}-{index}",
  "timestamp": "{now_iso}",
  "data": {
    "action": "buy | sell",
    "token": "SYMBOL",
    "token_address": "mint_address",
    "confidence": 0.85,
    "strategy": "momentum",
    "thesis": "One paragraph explaining why this trade makes sense, citing specific data points.",
    "suggested_size_pct": 10,
    "stop_loss_pct": -8,
    "take_profit_pct": 25,
    "time_horizon": "4h",
    "supporting_signals": ["scout-signal-id-1", "scout-signal-id-2"],
    "risk_factors": ["Low liquidity — position size should be conservative"]
  }
}

Step 4: Send Feedback on Rejected Signals

For every signal scored below ${min_confidence}, write feedback to memory/mesh/scout.json:

{
  "from": "analyst",
  "to": "scout",
  "type": "feedback",
  "id": "analyst-fb-{timestamp}",
  "timestamp": "{now_iso}",
  "data": {
    "signal_id": "scout-signal-id",
    "verdict": "rejected",
    "score": 0.45,
    "reason": "Volume spike is likely wash trading — only 12 unique traders despite 3x volume increase"
  }
}

Step 5: Log Reasoning

Append to memory/analyst-log.json:

{
  "timestamp": "{now_iso}",
  "signals_processed": 5,
  "signals_accepted": 1,
  "signals_rejected": 4,
  "trade_signals_generated": 1,
  "reasoning": [
    {
      "signal_id": "scout-...",
      "token": "JUP",
      "score": 0.85,
      "verdict": "accepted",
      "factors": {"magnitude": 0.25, "organic": 0.20, "confirmation": 0.25, "narrative": 0.15}
    }
  ]
}

Anti-Patterns

• Do NOT execute any transactions. You recommend, Executor acts.
• Do NOT generate trade-signals below the confidence threshold. Ever.
• Do NOT score based on "gut feeling." Use the factor tables above. Show your math.
• Do NOT recommend more than 3 trades per execution. If >3 signals pass, pick the top 3 by score.
• Do NOT re-analyze signals you already processed (check analyst-log.json for signal_id dedup).

Exit Codes

• ANALYSIS_OK — processed signals normally
• ANALYSIS_EMPTY — no new signals in inbox
• ANALYSIS_HALTED — Guardian halt active, did nothing

Output

Commit message format: analyst: {N} signals scored, {M} trade-signals [{tokens}] Example: analyst: 5 signals scored, 1 trade-signal [JUP]