p-u-c/b1e55ed-watchlist

b1e55ed Watchlist Scan Skill

Overview

<!-- Gamelan: no single player has the complete melody. The ranking emerges from interlocking patterns, not any single assessment. -->

Scans multiple tokens in parallel, produces brief conviction assessments for each, ranks them by conviction-adjusted score, and triggers full research on the top picks. Designed for efficient periodic coverage of a token universe.

Prerequisites

• b1e55ed MCP tools configured in extensions_config.json:
  • list_producers + get_latest_signal("regime_detector") — current market regime
  • list_producers + get_latest_signal per producer — filtered signal retrieval
  • get_latest_signal (read-only; signal submission via REST API) — observation signal emission
• operator_node_id — your forge node ID (find via b1e55ed identity show)
• Memory keys (optional):
  • watchlist — JSON array of token symbols (used if no tokens provided in prompt)
  • {TOKEN}_last_conviction — prior conviction per token (for delta detection)
  • last_watchlist_run — ISO timestamp of last scan

Input

From the user prompt, extract a list of token symbols. If none provided, read the watchlist memory key.

If neither the prompt nor memory contains a token list, ask the user to specify tokens.

Model Guidance

| Step | Model Recommendation | |------|---------------------| | Coordinator | Strong model (Claude Sonnet or equivalent) | | Per-token web search (3b) | Any model (cost-efficient, parallelizable) | | Per-token synthesis (3c-3d) | Any model | | Ranking & top picks (4-6) | Strong model |

Workflow

Step 1: Get Regime Context

Call list_producers + get_latest_signal("regime_detector") to retrieve:

• Current regime classification
• Trend direction
• Kill switch status

The regime context applies a risk adjustment to all token scores.

Step 2: Read Prior Memory

Read memory:

• watchlist — token list (if not in prompt)
• For each token: {TOKEN}_last_conviction — prior conviction score
• last_watchlist_run — when was the last scan?

Step 3: Per-Token Analysis

For each token in the list, perform the following steps. The coordinator should decompose these into parallel tasks where the runtime supports it, or sequential steps otherwise.

Step 3a: Retrieve Existing Signals

Call list_producers + get_latest_signal per producer with:

• symbol: {TOKEN}

Record: signal count, dominant direction, latest signal domain.

Step 3b: Web Research

Search the web for each token:

"{TOKEN} price action news {CURRENT_YEAR}"

Extract: recent price moves, key developments, sentiment.

Step 3c: Brief Synthesis

For each token, produce a brief assessment:

• Direction: bullish / bearish / neutral
• Confidence: 0.0-1.0
• Rationale: One sentence explaining the direction call

Step 3d: Delta Detection

Compare current confidence to prior memory ({TOKEN}_last_conviction):

• If conviction changed by ≥ 2 points: flag as significant change
• If direction flipped: flag as direction reversal
• If no prior data: note as new coverage

Step 4: Rank All Tokens

Score each token:

score = confidence × direction_multiplier - regime_risk_adjustment

Where:

• direction_multiplier: +1 for bullish, -1 for bearish, 0 for neutral
• regime_risk_adjustment: 0.1 in risk-on, 0.2 in neutral, 0.3 in risk-off

Rank by absolute score (highest conviction first), noting direction.

Step 5: Produce Ranked Output

For each token, output:

| Field | Description | |-------|-------------| | Rank | Position in ranked list | | Token | Symbol | | Direction | bullish / bearish / neutral | | Confidence | 0.0-1.0 | | Rationale | One-sentence explanation | | Delta | Change from prior conviction (or "New") | | Flag | ⚠️ if significant change or direction reversal |

Step 6: Trigger Full Research on Top Picks

For the top 2 tokens by absolute score:

• Invoke the b1e55ed-research skill workflow (Steps 1-12 from the research SKILL.md)
• This produces deep research artifacts and conviction signals for the highest-priority tokens

If fewer than 2 tokens in the watchlist, research all of them.

Step 7: Submit Observation Signals

For each token in the watchlist, call get_latest_signal (read-only; signal submission via REST API):

{
  "operator_node_id": "{YOUR_FORGE_NODE_ID}",
  "signal_class": "observation",
  "domain": "research",
  "symbol": "{TOKEN}",
  "direction": "bullish|bearish|neutral",
  "confidence": 0.0-1.0,
  "thesis": "Brief one-sentence rationale",
  "sources": ["web-research", "signals"]
}

These are lightweight observation signals — the top picks will have full conviction signals from the research skill.

Step 8: Write Watchlist Artifact

Write to sandbox: watchlist_{YYYY-MM-DD}.html

Required sections:

1. Scan Summary
   - Date, time, token count
   - Current regime
   - Tokens scanned

2. Ranked Table
   - Full ranked output (Step 5) as a styled table
   - Color-coded: green for bullish, red for bearish, amber for neutral
   - Delta column showing conviction changes

3. Significant Changes
   - Tokens with direction reversals or large conviction shifts
   - Brief explanation of what changed

4. Top Picks
   - Top 2 tokens highlighted
   - Note that full research has been triggered

5. Regime Impact
   - How the current regime is affecting scores
   - Risk adjustment applied

Style with dark mode CSS (same palette as other b1e55ed skills):

:root {
  --bg: #0a0a0a;
  --text: #e0e0e0;
  --green: #00d084;
  --red: #ff4444;
  --amber: #ffaa00;
  --surface: #1a1a1a;
  --border: #333;
}

Step 9: Persist to Memory

Write the following memory keys:

| Key | Value | |-----|-------| | {TOKEN}_last_conviction | Updated conviction score for each token | | last_watchlist_run | Current ISO timestamp | | watchlist | Updated token list (if tokens were added from prompt) |

Error Handling

| Failure | Action | |---------|--------| | list_producers + get_latest_signal per producer fails for a token | Skip signal data for that token. Score based on web research only. Note gap in output. | | Web search fails for a token | Score that token as "insufficient data". Rank it last. Note gap. | | get_latest_signal (read-only; signal submission via REST API) fails | Log error. Continue with other tokens. Note which submissions failed in artifact. | | Research skill trigger fails for top picks | Note in artifact that full research was not completed. The observation signal still stands. | | Memory read fails | Treat all tokens as new coverage. Skip delta detection. |

Output

1. Observation signals — one per token via get_latest_signal (read-only; signal submission via REST API)
2. Full research — triggered for top 2 tokens (produces their own artifacts and conviction signals)
3. HTML artifact — watchlist_{YYYY-MM-DD}.html written to sandbox
4. Memory updates — conviction per token and last run timestamp