eyal-gross

Airport Investment Agent — Data skill for Claude Code

Data community

AI agent that helps an investor pick which US airports to invest in.

How to install Airport Investment Agent

This entry records only its repository, not the path inside it, so there is no exact command to give. Open eyal-gross/airport-investment-agent and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Airport Investment Agent does

AI agent that helps an investor pick which US airports to invest in. Public BTS/FAA data pipeline, deterministic capacity-need score, 5 read-only tools in a Claude tool loop, and a number check that blocks any figure no tool returned. 26-question eval, 0 untraceable numbers.

Alternatives in Data

Financial Services — Reference agents, skills, and data connectors for the financial-services workflows we see most — investment ba 34.5k ★
Agentlas Storm — Run a force-robust Stormbreaker loop — route to real agents, execute a verified pipeline to completion 1.1k ★
Ralph The Marketer - Autonomous Copywriter Loop — You are starting the Ralph the Marketer autonomous copywriting agent 698 ★

README

Airport Investment Agent

A chat agent that helps an investor decide which US airports are worth putting modernization money into, answering in plain English from public BTS and FAA data, where code computes every number and the LLM never does.

Ask "Which airports in New England are strong candidates for terminal expansion?" or "What is the unmet flight demand at SFO and why?" and you get a verdict, the reasons, the caveats, and a footer built by code with the data period, assumptions and a computed confidence label. It covers all 402 FAA primary airports (343 fully ranked), a deterministic capacity-need score, five read-only tools, and a number check that refuses to show an answer containing a figure no tool returned. On the 26-question eval it passes 25 to 26 of 26 per run, with 0 untraceable numbers across roughly 2,500 numbers in answers.

I designed the architecture, the score and the guardrails, and built it with Claude Code as my pair.

How it works

flowchart LR
    subgraph Build["Offline build (one command)"]
        S[BTS T-100, BTS Delay Causes,
BTS T-100 Segment, FAA TAF,
FAA boardings, OurAirports] --> F[fetch]
        F --> C[clean]
        C --> SC[score]
        SC --> K{37 data checks}
        K -- pass --> P[(data/serving
Parquet)]
        K -- fail --> X[build stops,
old tables kept]
    end
    subgraph Ask["Per question"]
        Q[Question + history] --> L[Claude tool loop
max 8 rounds]
        L <--> T[5 read-only tools]
        T --> P
        L --> N{Number check}
        N -- every number traced --> A[Answer + code-built footer]
        N -- untraced number --> R[one rewrite]
        R --> N2{still failing?}
        N2 -- no --> A
        N2 -- yes --> FB[fallback answer built by code]
    end

**1. Data pipeline.** `python -m airport_agent.pipeline.build` downloads six public sources, cleans each into one tidy table, computes metrics, scores, tiers and warnings, then runs 37 data checks. If any che