- Home
- Skills
- Trading & Investing
- Legion Stock Panel: AI Investment Committee
More screenshots
Works with the AI Agents you already use
Legion Stock Panel: AI Investment Committee
Asking one AI "should I buy this stock?" gets you one opinion with one set of blind spots, written by a model that has already made up its mind.
$3
Legion Stock Panel: AI Investment Committee
Example session with this skill installed
/legion-stock-panel ZYNX — run the panel like a hedge fund desk would and give me the final call and how confident I should be. I'm thinking of buying tomorrow.
- Read your context and instructions
- Compiled the legion stock panel
About this demo: a real run of the skill (lightly formatted) on
ZYNX, a fictional ticker with a fixed sample data file (strong uptrend, a guidance cut, euphoric retail sentiment, crowded positioning), so anyone can reproduce it. 8 analyst sub-agents ran: 4 hats × 2 rounds.
ZYNX panel: don't buy tomorrow. The panel's call is SELL, but with low confidence.
The panel converged in round 2 of 3. Three analysts (news, social, contrarian) say SELL and technical says BUY. Panel conviction is
0.29 on a 0–1 scale (low). You don't own ZYNX, so read this as "don't buy", not as a call to short it.
ZYNX — SELL (CONVERGED, round 2), conviction 0.29 (low) data as of 2026-10-30
Lean S=-0.58 disagreement V=0.66 agreeing share κ=0.79 panel 4/4
technical BUY 0.45 "Uptrend intact: above SMA20/50, MACD +, 0.4% off 52w high, but RSI 79 and 45% vol: pullback risk."
news SELL 0.6 "FY guidance cut on softer enterprise demand outweighs the 12% EPS beat; CFO exit, 2 downgrades vs 1 upgrade."
social SELL 0.55 "88% bullish on 5.2x volume with meme themes: late-stage froth, not an early crowd move."
contrarian SELL 0.6 "F&G 82, NAAIM 98, put/call 0.55: crowded greed; 3% short interest gives little squeeze fuel."
Risk brake: outsized move 9.1% (>= 3 sigma of 2.8% daily vol), conviction cap 0.4; not binding (call already 0.29).
What would change it: the technical uptrend vs. the guidance cut. Sentiment cooling while price holds its trend would undercut the social/contrarian case.
Gaps: SMA200, distance from 52-week low. Not financial advice; a structured second opinion only.
How the argument went
| Round | technical | news | social | contrarian | S | V | κ | Decision |
|---|---|---|---|---|---|---|---|---|
| 1 (blind) | BUY 0.6 | SELL 0.6 | SELL 0.5 | SELL 0.65 | -0.48 | 0.77 | 0.74 | ITERATE (too much disagreement) |
| 2 (saw peers) | BUY 0.45 | SELL 0.6 | SELL 0.55 | SELL 0.6 | -0.58 | 0.66 | 0.79 | CONVERGED → SELL |
Round 1: each analyst saw only its own data. Most of the weight leaned SELL, but disagreement (V = 0.77) was above the 0.75 limit, so the panel argued another round.
Round 2: technical kept BUY but cut its conviction from 0.6 to 0.45, accepting that RSI 79 and a price about 15% above the 50-day average make a sharp pullback likely. News held at 0.6: the uptrend may mean the market has shrugged off the guidance cut, but forward guidance outweighs a backward-looking beat.
Herding check passed: the SELL side already had 0.74 of the round-1 weight from 3 independent analysts, so this is real agreement, not a cave-in.
How confident to be
Not very. The panel agrees on direction, but 0.29 is weak. The strongest single signal is "don't chase": ZYNX jumped 9.1% on Friday, more than three times its normal daily move. Buying the day after that, against a guidance cut and euphoric retail sentiment, is the trade this panel argues against. The real counter-case is momentum, since the stock is at its 52-week high and still trending up.
Next steps
- Re-check the data. It is from the Friday close; confirm nothing material has happened since.
- The verdict is logged to the ledger. In 5 days run
/legion-stock-panel grade my past callsto see whether the panel beat SPY. - After 5+ graded calls per analyst, the panel starts weighting the analysts who have actually been right.
Connects securely to your tools. The creator never sees your data.
About this skill
Asking one AI "should I buy this stock?" gets you one opinion with one set of blind spots, written by a model that has already made up its mind. Legion Stock Panel turns your coding agent into a leaderless investment committee instead: four independent analysts, a math-based vote, and a memory of who has actually been right.
Extracted from Legion, a production multi-agent stock-signal engine.
What it does
1. Debates a ticker with four independent analysts
- Technical: trend, momentum, RSI, MACD, volatility
- News / catalyst: earnings, guidance, analyst actions, macro
- Social: retail sentiment and message volume
- Contrarian: fades crowded extremes using Fear & Greed, AAII, NAAIM, put/call and short interest
Each analyst runs in its own sub-agent and sees only its own slice of data, so you get four real opinions, not one model playing four roles. They vote blind, then see each other's arguments and may revise, for up to 3 rounds.
2. Lets math decide, not the AI A bundled zero-dependency script scores every round:
- S: which way the panel leans
- V: how much it disagrees
- κ: the share of the panel on the winning side
A call needs a 2/3 weighted supermajority and low disagreement. A late agreement where everyone caved to one loud analyst is rejected as herding. If the panel stays split, you get an honest NO_CONSENSUS instead of a forced BUY.
3. Applies a risk brake Deterministic rules cap conviction when VIX is 30 or higher or after a 3-sigma one-day move, and block new buys when VIX is 40 or higher. The brake never flips the direction.
4. Learns who to trust Every verdict is logged to a local ledger and graded against SPY (beating the market, not just going up). Each analyst earns three dials:
- Trust: earned slowly, lost twice as fast
- Calibration: whether its confidence means anything
- Information: whether it ever changes its mind
Before voting, each analyst sees its own graded record and a lesson it wrote from its misses.
5. Bonus for developers
The same leaderless voting protocol works for any multi-agent decision, such as PR review, moderation or grading. Import evaluatePanel() directly.
How to use
/legion-stock-panel NVDA
Run a panel on TSLA, I'm thinking of trimming my position.
/legion-stock-panel grade my past calls
The skill only activates when you ask for a panel, debate, committee or second opinion, or use the slash command. A quick "is NVDA a buy?" still gets your agent's normal fast answer.
Requirements
- Node.js 18+ (scripts have no dependencies)
- An agent that supports sub-agents is recommended (Claude Code). Others fall back to a clearly labelled single-context panel.
- Market data comes from whatever your agent can reach: a market-data MCP server (best), web search, or files you provide. Missing data is marked missing, never invented.
Disclaimer
For education and research only. Not financial advice and not a recommendation to buy or sell any security. Past panel performance does not predict future results.
Requires Node.js 18+ to run consensus scripts. Analysis quality depends on market-data MCP availability. Ledger requires consistent logging to improve hat weighting over time.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 14 days ago
Requires
Permissions
File Scopes
Terminal: runs the bundled Node scripts (consensus scoring, ledger). Read/write files: writes per-round vote JSON in the working folder and keeps the track-record ledger at ~/.stock-panel/ledger.json. No network access by the scripts themselves; market data comes from tools your agent already has.