Aiprox Swarm

40-persona prediction engine for binary market questions. Returns YES/NO/SKIP with confidence score and reasoning from a simulated trader swarm.

Security Scan

The skill's behavior matches its description (it forwards questions to a 40-persona swarm), but it sends user-provided context over plain HTTP to a raw IP address (185.56.20.122) — this is a strong privacy and provenance risk and should be verified before use.

Purpose & Capability

Name/description match the runtime instructions: the skill simply submits binary questions to an external swarm endpoint and returns the response. There are no unexpected local binaries, environment variables, or install steps. The only oddity is that the SKILL.md hard-codes a numeric IP and an HTTP endpoint rather than a documented HTTPS domain; that reduces provenance and increases risk even though it is functionally consistent with the stated purpose.

Instruction Scope

The instructions direct the agent to POST user-provided questions and any optional context to http://185.56.20.122:3030/v1/task. Context can include arbitrary text (news, price levels, etc.), and there is no instruction to redact sensitive or private data. Critically, the endpoint uses unencrypted HTTP (not HTTPS) and a raw IP address, so any transmitted content can be intercepted or tampered with in transit and the server's identity is not verifiable. This makes accidental exfiltration of secrets or sensitive customer data likely if such information is included in the 'context'.

Install Mechanism

No install spec and no code files — instruction-only skill. This minimizes on-disk execution risk because nothing is written or executed locally by the skill package itself.

Credentials

The skill requests no environment variables or credentials, which is proportionate to its described network-forwarding role. However, the lack of required credentials plus the use of an open HTTP endpoint means the skill will accept and transmit any user-supplied text; that amplifies data-exposure risk even though no explicit credential requests are made.

Persistence & Privilege

The skill does not request persistent or elevated platform privileges. always is false and there is no mechanism that would cause it to modify other skills or system-wide settings. Autonomous invocation is allowed (normal default) but does not combine with other elevated privileges here.

What to consider before installing

This skill sends the question and any 'context' you provide to an external server at http://185.56.20.122:3030 — over unencrypted HTTP and to a raw IP address. Before installing or using it: (1) Do not send any secrets, credentials, personal data, or proprietary documents as 'context'. (2) Prefer skills hosted on a verifiable HTTPS domain or with published server details; verify aiprox.dev and the operator before trusting results. (3) Test with only synthetic or public data first. (4) If you require confidentiality, block outbound network access for this skill or avoid it entirely. (5) If you need stronger assurances, ask the publisher for documentation about storage policies, TLS support, and the reason for using a raw IP.

SWARM

Submit a binary YES/NO market question to a 40-persona AI trader swarm. Each persona reasons independently — bulls argue for YES, bears argue for NO, neutral analysts vote freely. Confidence-weighted votes are aggregated into a final verdict with reasoning and a groupthink warning when consensus is suspiciously one-sided.

Personas: 18 bulls · 18 bears · 4 neutral analysts
Domains: crypto, macro, political, sports, general
Endpoint: http://185.56.20.122:3030/v1/task

When to Use

• Polymarket-style binary questions: "Will X happen before Y?"
• Bitcoin and crypto price targets
• Macro events: Fed decisions, CPI outcomes, rate cuts
• Election and political outcomes
• Any question where you want a structured bull/bear debate rather than a single model's opinion

Usage Flow

1. Provide task — the binary question (minimum 10 words)
2. Optionally provide context — recent news, price levels, or market data appended as recent news: ...
3. SWARM fans the question out to 40 personas in parallel (~30–60s)
4. Returns prediction (YES/NO/SKIP), confidence (1–85), reasoning, voter_summary, domain, and groupthink_warning

Security Manifest

Permission	Scope	Reason
Network	185.56.20.122:3030	Direct swarm endpoint

Make Request — Basic

curl -X POST http://185.56.20.122:3030/v1/task \
  -H "Content-Type: application/json" \
  -d '{
    "task": "Will Bitcoin exceed $150,000 before the end of 2025?"
  }'

Response — Basic

{
  "prediction": "YES",
  "confidence": 38,
  "reasoning": "ETF Flow Tracker: Spot ETF inflows averaging $800M/week create structural demand that historically precedes new highs. | Global Liquidity Bull: M2 expansion in China and Europe is feeding into risk assets with a 6-month lag. | Leverage Flush Analyst: Open interest is elevated and funding rates are positive — a flush before the move up is likely but doesn't change the directional call.",
  "voter_summary": "Bull avg: 61 (18 active, 0 SKIP) | Bear avg: 23 (18 active, 0 SKIP) | Neutral: 3 YES / 1 NO / 0 SKIP | Margin: 38pts",
  "domain": "crypto",
  "persona_count": 40,
  "groupthink_warning": false
}

Make Request — With Context

curl -X POST http://185.56.20.122:3030/v1/task \
  -H "Content-Type: application/json" \
  -d '{
    "task": "Will the Fed cut rates at its June meeting?\n\nRecent news: CPI printed 2.4% in March, below expectations. Jobs report added 175k jobs, unemployment ticked up to 4.1%. Fed chair Powell said the committee needs more confidence inflation is moving sustainably to 2% before cutting."
  }'

Response — With Context

{
  "prediction": "NO",
  "confidence": 29,
  "reasoning": "Higher-For-Longer Bear: Powell's explicit signal of needing more confidence, combined with still-solid employment, gives the committee cover to hold. | Macro Bear: The jobs market has not deteriorated enough to force the Fed's hand — June remains a stretch. | Economist: Base rates favor holding: the Fed has cut at only 3 of the last 12 June meetings when unemployment was below 4.2%.",
  "voter_summary": "Bull avg: 31 (17 active, 1 SKIP) | Bear avg: 60 (18 active, 0 SKIP) | Neutral: 1 YES / 3 NO / 0 SKIP | Margin: 29pts",
  "domain": "macro",
  "persona_count": 40,
  "groupthink_warning": false
}

Example Prompts

Will Bitcoin exceed $150,000 before the end of 2025?

Will Ethereum outperform Bitcoin over the next 90 days?
Context: ETH/BTC ratio is at 0.038, a 4-year low. Pectra upgrade ships next month.

Will the S&P 500 be higher than today in 6 months?
Context: Current level 5200. Fed on hold. Q1 earnings beat expectations by 7%.

Will there be a US recession before the end of 2026?

Will Solana flip Ethereum by market cap before 2027?

Output Fields

Field	Type	Description
prediction	string	YES, NO, or SKIP
confidence	number	1–85. Margin between bull and bear weighted averages.
reasoning	string	Top 3 reasons from the winning camp, highest-confidence personas first.
voter_summary	string	Bull avg, bear avg, neutral split, and margin in one line.
domain	string	Detected domain: crypto, macro, political, sports, or general.
persona_count	number	Always 40.
groupthink_warning	boolean	true when >80% of non-SKIP votes point the same direction.

Notes

• SKIP is a valid result. It means the bull and bear camps are within 15 confidence points — the swarm is genuinely split.
• Groupthink warning fires when consensus is very high. This can mean the question is already settled, or that the swarm lacks the context to form a contrarian view. Add news context and re-run.
• Minimum 10 words. Very short questions without context return a 400 error — the swarm needs enough signal to reason against.
• ~30–60 seconds per call. 40 parallel LLM calls run concurrently via DeepSeek.
• Domain weighting. Crypto questions weight crypto-domain personas at 1.5×. Macro questions weight macro experts. General questions use equal weights.

Trust Statement

SWARM processes your question transiently. Questions and responses are not stored beyond the active request. No authentication or API key required for the default public endpoint.