build-watchlist - SKILL.md Agent Skill

name: build-watchlist description: > Research sectors, pick leader stocks, and scaffold or maintain groups in ~/.lafmm/data/. Use this skill when the user wants to start tracking a new sector, asks "what sectors should I watch?", "add semis to my watchlist", "which stocks should lead my energy group?", "set up a new group for X", or any variation that implies they want to evaluate and add a sector to their Livermore system. Also use it when you notice the user trades a sector they are not tracking, or when they ask for a broad market scan to find opportunities. Use it proactively when the user's journal shows repeated trades in an un-tracked sector. Also use it when you notice signs that an existing group may need revisiting — a leader's volume has dropped well below its sector peers, a tracked stock consistently outperforms both leaders, or a group has produced no meaningful signals for an extended period. These are signs the group configuration is stale.

Build Watchlist

This skill covers both initial setup and ongoing maintenance of groups. Markets rotate. Leaders change. A group set up six months ago may need a leader swap, threshold re-tuning, or removal. The same workflow applies whether you are building from scratch or revisiting.

Interaction principle

Steps 1-3 (sector evaluation, leader selection, tracked stocks) involve judgment — present your analysis with reasoning, then pause for human confirmation before proceeding.

Steps 4-6 (check existing, scaffold, fetch/tune) are mechanical — execute them and report results. Only pause if an error occurs.

Steps 7-8 (starting trend, final confirm) are a single checkpoint — present everything together and ask once.

In short: research → pause → execute → report → confirm.

The workflow

1. Identify the sector

The user may name a sector directly ("add semis") or ask you to scan for opportunities. In either case, you need to determine whether the sector is worth tracking right now.

Use the news-context skill to check for active catalysts. A sector without catalysts can still be tracked — Livermore tracked sectors in quiet periods too, because he wanted to see the column transition when the trend started. But a sector with active catalysts is more immediately useful.

Read references/sector-criteria.md for the evaluation framework and the sector catalog with example leaders.

2. Pick two leaders

Leaders are the two most important stocks in the group. Their combined price forms the Key Price, which determines the group's trend. Picking the wrong leaders makes the entire group analysis unreliable.

references/sector-criteria.md has the full criteria. The short version: pick the two highest-volume, most-representative stocks that are correlated but not redundant (ideally different sub-sectors within the sector).

If you are unsure which stocks best lead a sector, say so — the human may have domain knowledge you lack. Present your candidates with reasoning and let the human decide.

3. Optionally pick tracked stocks

Any additional stocks placed in the group directory are auto-discovered as tracked stocks. They get their own engine but do not affect the Key Price. These are useful for spotting divergence — if a tracked stock enters Downward Trend while the Key Price is still in Upward Trend, that tension is worth surfacing.

Tracked stocks are optional. Start with leaders only if the user is unsure what else to include.

4. Check for existing group

Before scaffolding, check if the group already exists:

If data/{group}/group.toml exists and the user wants to add a tracked stock — just create the ticker directory and fetch its prices. No group.toml change needed. The engine auto-discovers it.
If the user wants to change leaders — warn them: changing leaders alters the Key Price retroactively. All existing Key Price pivots and signals will shift. The user must confirm before you proceed.
If the group does not exist, continue to scaffolding.

5. Scaffold the group

Create the directory structure and group.toml. This is the concrete output of the skill — files on disk that the engine can read.

# Create group directory
mkdir -p ~/.lafmm/data/{group}

# Create group.toml
cat > ~/.lafmm/data/{group}/group.toml << 'EOF'
name = "Semiconductors"
leaders = ["NVDA", "AVGO"]
swing_pct = 5.0
confirm_pct = 2.5
EOF

# Create ticker directories
mkdir -p ~/.lafmm/data/{group}/{TICKER}

Use swing_pct = 5.0 and confirm_pct = 2.5 as initial values. These are placeholders — they get calibrated in the next step.

6. Fetch prices and tune thresholds

After scaffolding, invoke the other skills in order:

fetch-prices — backfill 2 years of data for each ticker (SMA 200 needs ~250 bars for warmup):
```
./run .claude/skills/daily-update/scripts/fetch-prices.py NVDA AVGO --days 730
```
Before fetching, check if data/_adhoc/{TICKER}/ exists. If so, move it to the group directory instead of fetching, then delete the _adhoc/{TICKER}/ directory to avoid duplicates.
tune-thresholds — compute ATR-based thresholds from the fetched data:
```
./run .claude/skills/tune-thresholds/scripts/atr.py ~/.lafmm/data/{group}
```
Present the suggested values to the user. If they agree, update group.toml with the new swing_pct and confirm_pct.

7. Determine starting trend

After fetching prices, check the last 30 days of each leader to decide the initial column. If both leaders show higher highs and higher lows, use start_col = "UT". If both show lower highs and lower lows, use start_col = "DT". If ambiguous or leaders disagree, default to "UT" and note to the user that the engine will self-correct after the first full swing cycle — the first few pivots may not be meaningful.

Add start_col to group.toml if it differs from the default:

start_col = "DT"

8. Confirm with the user

Summarize what you set up: the group name, leaders, tracked stocks (if any), the data range fetched, and the calibrated thresholds. Ask if anything needs adjusting.

Connecting to profile and journal

If ~/.lafmm/profile.md specifies concentration limits (e.g., "max 5 sectors"), check how many groups already exist in data/ before adding a new one.

If ~/.lafmm/accounts/ has trade journals, scan for tickers the user trades frequently that are not in any existing group. These are natural candidates for a new watchlist group.

Examples

Example 1: Direct request

User: "I want to track semiconductors"

You would: read sector-criteria.md, confirm NVDA + AVGO as leaders (highest volume, different sub-sectors — GPU vs broadcom/networking), ask if they want to track AMD or INTC too, scaffold the directory, fetch 2 years of prices, run tune-thresholds, present the result.

Example 2: Opportunity scan

User: "What sectors should I be watching right now?"

You would: use news-context to scan tier-1 feeds across multiple sectors, identify which have active catalysts (earnings season in tech, OPEC decision in energy, FDA rulings in biotech), present a ranked list with reasoning, then scaffold whichever the user picks.

Example 3: Proactive suggestion

You notice the user's IBKR journal has 12 trades in biotech stocks over the last month, but no biotech group exists in data/. You suggest: "You've been active in biotech — want me to set up a group so the Livermore system can track it?"