Skip to content

Security: agisoham/ARIA

Security

SECURITY.md

Security policy

ARIA is a personal research project, but it touches API keys and (eventually) brokerage accounts, so secret hygiene and execution safety are treated as engineering requirements rather than good intentions.

Reporting a vulnerability

Found a leaked credential, an unsafe default, or a way to make the system place an order it shouldn't? Please open a GitHub issue describing the problem — or, if it involves an exposed secret, avoid posting the value itself and say only where it appears so it can be rotated first.

There's no bounty; this is a solo learning project. Reports are still very welcome.

How secrets are handled

  • Nothing real is committed. .env files are gitignored (.env, *.env, with only .env.example allowed through), and private strategy topics are excluded by default — llm-council/topics/* is ignored except the neutral template.
  • A pre-commit hook runs gitleaks (entropy + heuristics) so a key can't be committed by accident. See .pre-commit-config.yaml.
  • A second scanner runs before publishing. The publish/sync tooling refuses to proceed if anything key-shaped appears in the staged files.
  • Keys are per-provider, free-tier, and rotatable. Any credential that has ever appeared outside a local .env — including in a chat window or a screenshot — is treated as compromised and rotated.

If you fork this repo: copy llm-council/.env.example to .env and fill in your own keys. Never commit that file.

Execution safety (why this repo can't lose your money)

These are enforced constraints, not aspirations — the full list lives in AGENTS.md:

  • No live trading by default, ever. Live execution is opt-in, human-gated, and only reachable after the paper-trading gates pass.
  • LLMs never place trades. The AI council debates design; deterministic, tested code produces signals. No agent has an order-placement path.
  • No agent may write to a broker API. An AI coding agent working on this repo is blocked from money-moving commands by hook-level guardrails, not by prompt instructions alone.
  • Kill-switch first. The Risk Guardian (drawdown limits, VaR/CVaR, stop-loss, kill-switch) is a Phase-2 prerequisite — it is built before any capital is deployed, not after.

Scope

This project trades only its author's own capital. It does not manage third-party money, and nothing in this repository is financial advice.

There aren't any published security advisories