[mvanhorn]

Summary

• Private keys in ~/.config/polymarket/config.json were stored as plaintext - any process running as the same user could read them
• wallet create and wallet import now encrypt keys using the standard Ethereum keystore format (AES-128-CTR + scrypt), the same format used by MetaMask, Geth, and Foundry
• All commands that need the key prompt for a password (or read POLYMARKET_PASSWORD env var for CI/scripts)
• --no-password flag available for plaintext storage (not recommended)
• New wallet export subcommand to decrypt and display the key for backup
• Backward compatible: existing plaintext configs continue to work

New dependencies

• rpassword - secure terminal password input (no echo)
• rand 0.8 - RNG for keystore encryption
• alloy signer-keystore feature - standard keystore encrypt/decrypt

Fixes #18
Supersedes #19 (pre-refactor, needs full rewrite after #13)

Test plan

• cargo fmt --check passes
• cargo clippy -- -D warnings passes
• cargo test passes (131 tests)
• polymarket wallet create prompts for password and stores encrypted keystore
• polymarket wallet create --no-password stores plaintext (with warning)
• polymarket wallet import <key> prompts for password
• polymarket wallet export decrypts and shows key
• All commands prompt for password when key is encrypted
• POLYMARKET_PASSWORD=xxx polymarket wallet address works without interactive prompt

This contribution was developed with AI assistance (Claude Code).

---

Note

High Risk
Changes how private keys are stored, decrypted, exported, and held in memory—core wallet security behavior that reviewers should scrutinize carefully.

Overview
Wallet create/import now default to password-protected storage using the standard Ethereum keystore format (wallet.json under the config keystore dir) via Alloy’s signer-keystore, instead of writing the private key into config.json. --no-password keeps the old plaintext-on-disk behavior (with warnings), and config gains an encrypted flag so existing plaintext setups still load unchanged.

Unlocking prompts interactively (or uses POLYMARKET_PASSWORD) when commands need the key from an encrypted wallet; the decrypted key is cached for the rest of a single command and cleared after interactive shell lines. wallet export decrypts and prints the key for backup.

Keystore writes use a temp directory and replace a fixed wallet.json filename so overwrites do not leave stale files that could be picked up by mistake. Dependencies add rpassword, rand, and the keystore crypto stack in the lockfile.

^{Reviewed by Cursor Bugbot for commit efcfbf3. Bugbot is set up for automated code reviews on this repo. Configure here.}

[cursor]

Stale Bugbot comment from a previous run.

[mvanhorn]

Addressed both cursor findings in d491d80:

1. Atomic keystore write - save_encrypted now writes to a .tmp subdirectory first. Old keystore files are only deleted after the new one is successfully written, then moved into place. If encrypt_keystore fails, the old keystore remains intact.

2. Stale keystore cleanup in plaintext path - Added cleanup_keystore_files() helper called in both cmd_create and cmd_import plaintext branches, so old encrypted keystore files don't linger on disk when overwriting with --force --no-password.

[mvanhorn]

Merged current main into this branch in ae96342. The conflicts were in Cargo.toml and Cargo.lock from the move to polymarket_client_sdk_v2: the keystore code now builds against the v2 SDK, with signer-keystore, rpassword and rand added next to main's alloy features. I also collapsed one nested if let in cleanup_keystore_files that the current clippy flags.

cargo fmt --check, cargo clippy --all-targets -- -D warnings and cargo test (82 unit, 49 integration) all pass locally.

[cursor]

Cursor Bugbot has reviewed your changes using default effort and found 2 potential issues.

^{Reviewed by Cursor Bugbot for commit ae96342. Configure here.}

[mvanhorn]

Both Bugbot findings were valid and are fixed in efcfbf3.

Password prompted on every key resolve: the decrypted key is now kept for the rest of the command, so approve set asks once and can't stop partway through at a later prompt. The shell drops it after each command, and any config write drops it, so a replaced wallet is never served from the old key.

Keystore decrypt picks an arbitrary file: the keystore is now always written as wallet.json and that is the only file read, so a leftover keystore or .DS_Store can't be selected. The rename replaces the old file directly, then anything else in the directory is removed, so a failed move no longer leaves the directory empty.

I added a unit test for the single unlock and an integration test that overwrites a wallet, adds stray files and reads the address back. cargo fmt --check, cargo clippy --all-targets -- -D warnings and cargo test (83 unit, 50 integration) pass locally. I did not run approve set against a live RPC.