Question 1

Why does Private Key Management matter?

Accepted Answer

**What are the four tiers of private key storage? What is the security strength and applicable scenario of each?**

From most basic to most secure, Agent private key storage has four tiers:

**Tier 1: Plaintext environment variables (most dangerous — prohibited in production)**
Storing the private key directly in a `.env` file or OS environment variables. Common mistakes: `.env` file accidentally committed to GitHub, or `printenv` logs leak the key to a logging system. Acceptable for development/testing, absolutely not for production.

**Tier 2: Encrypted Secret services (baseline — minimum requirement for production)**
Railway Secrets, AWS Secrets Manager, HashiCorp Vault — these services store keys encrypted; the application decrypts at runtime. The key doesn't exist in plaintext in application code or version control. Railway's Secrets feature is sufficient for this tier and developer-friendly. Main risk: access management of the service itself — if your Railway account is hijacked, attackers can read all Secrets. Enable Railway 2FA and IP access restrictions.

**Tier 3: HSM / KMS (Hardware Security Module)**
AWS KMS, Google Cloud KMS, Azure Key Vault — the private key never leaves the hardware security module in plaintext; signing operations complete inside the HSM; the application can only send 'data to be signed' and receive 'signed result,' never seeing the key itself. The biggest security upgrade for Agents — even if application code is completely compromised, attackers cannot export the private key. Suitable for high-value Agents. Cost: AWS KMS ~$1-5/month + per-API-call fees.

**Tier 4: MPC (Multi-Party Computation) Wallet**
The private key never exists in complete form anywhere — it is split into multiple shards; a threshold number of shards must cooperate to complete a signature (e.g., 2-of-3). Privy, Turnkey, Web3Auth offer commercial MPC solutions. Currently the highest security strength for private key management, with the highest cost and engineering complexity.

Question 2

How does Private Key Management work?

Accepted Answer

**What unique challenges do Agent private keys have that human-held keys don't?**

Agent private key management differs from human-held keys in several fundamental ways that ordinary wallet security advice cannot cover:

**Challenge 1: Private key is accessible to code 24/7**
For manual wallets, the private key resides in a hardware wallet or memory — used only 'when you're sitting at the computer operating,' perhaps only minutes of exposure daily. Agent private keys are available on servers 24/7 — attackers don't need to wait for a moment when you're 'online'; breaking the server or injecting malicious instructions can trigger signing at any time. This means the Agent private key attack surface is 144× that of a manual wallet (24 hours vs. ~10 minutes).

**Challenge 2: Prompt Injection can cause the Agent to 'proactively' use the private key for harmful actions**
With manual key holding, attackers must physically steal the device or trick you into operating it yourself. With Agents, attackers can use Prompt Injection to make the LLM believe 'executing this transfer is the correct strategy,' then the Agent proactively calls the private key to complete a malicious transfer — throughout the process the key is never 'stolen,' but it's 'authorized for use' to complete a malicious operation.

**Challenge 3: Continuous operation means no 'cooling-off period'**
In manual operations, if your computer is attacked, you typically notice quickly (computer slowing down, anomalous popups). An Agent's service can be silently manipulated in the background while appearing to run normally; attackers can use 'continuous operation' to accumulate small, steady losses that are hard to detect.

**Challenge 4: DevOps processes create new attack surfaces**
Private keys need to be injected into CI/CD pipelines, deployment scripts, and logging systems — each touchpoint is a potential leak. Manual wallets don't need to be integrated into DevOps processes, but Agent private keys almost certainly do. This is why 'encrypted Secret services' are the minimum requirement, not `.env` files.

Question 3

How is Private Key Management applied in practice?

Accepted Answer

**What is key rotation? How often should Onchain Agents rotate private keys?**

Key rotation refers to periodically replacing old private keys with new ones and migrating the assets in the Agent's operation wallet from the old address to a new one. The purpose of rotation: even if the old key was silently leaked at some point (but not yet exploited by the attacker), rotation invalidates the old key, confining the impact of the leak to within the rotation period.

**Why silent leakage is a real risk**: Your server, CI/CD system, and logging system may at some point be silently accessed by an attacker without your knowledge. Attackers may observe your system for a period before choosing the optimal moment to exploit a leaked key. Key rotation ensures that even if this silent leak occurs, the attacker's exploitable window is limited.

**Rotation frequency recommendations**:
- Low-value Agent (< $5,000 operations): rotate every 3 months
- Mid-value Agent ($5,000-$50,000 operations): rotate monthly
- High-value Agent (> $50,000 operations): rotate weekly
- Events triggering mandatory immediate rotation: any suspected leak event, server migration, team member departure (if team has key access)

**Rotation process**:
1. Generate new EOA address (or new MPC shards)
2. Migrate all assets and ERC-20 approvals from old operations wallet to new address
3. Update all system configurations (Railway Secrets, address whitelists in tool functions)
4. Revoke old address's ERC-20 approvals from all protocols
5. Verify new address functionality (run through testnet first)
6. Securely destroy old private key (delete from Secrets Manager — not just 'stop using it')

**Feasibility of automated rotation**: For EOA + Secrets Manager solutions, steps 1-4 can be automated (periodically triggered); steps 5-6 are recommended for human confirmation. MPC solution rotation is provided by service providers with tools, making it easier to automate.