Open a private security advisory rather than a public issue. Please do not include live credentials in the report.

This project reads three optional secrets from the environment:

They are read from .env or the process environment and are used nowhere else. Ignored by git:

• .env, .env.* (except .env.example), *.key, *.pem
• private/ — local notes, positions, cached tokens
• output/, _workspace*/ — run artefacts, which contain point-in-time market data

Only .env.example, which holds placeholders, is tracked.

tools/data.py never logs a key or echoes a full request URL. Agents are instructed never to write key values into reports, logs, or source. If you add a provider, keep both properties.

If you have committed a key: revoke it at the provider first, then rewrite history. Rotating the key is the step that actually matters — a key pushed to GitHub should be treated as public, since history rewriting does not reach forks, clones, or caches.

Toss issues one valid access token per client. Requesting a new one immediately invalidates the previous one, so parallel processes each fetching their own will invalidate each other mid-run. tools/data.py caches the token to disk (TOSS_TOKEN_CACHE, default /tmp/.toss_token.json) and re-issues only on expiry. On a shared machine, point that path somewhere only you can read.

The order and account endpoints are never called. This harness is read-only against market data and holds no trading authority. Keep it that way — a PR that adds order placement will be declined.

This is research tooling. It fetches public market data over HTTPS and writes files locally. It does not execute trades, hold funds, accept network input, or run a server. The realistic risk surface is credential leakage through committed files or pasted output, which is what the rules above are aimed at.