private-l1-metadata-leakage

Use when the agent must understand the privacy limits of a "private"/permissioned Avalanche L1 — what still leaks even with restricted RPC (insiders, validators, node operators, logs) — and how to def

SKILL.md

GitHub

Private L1 — Metadata & Data Leakage

Overview

A permissioned/private L1 provides access privacy (a walled garden), not data privacy. Even with restricted RPC ([[restrict-rpc-access]]), plenty is visible inside the wall. Know these limits before calling a chain "private."

What still leaks on a private L1

  • Plaintext to all participants. Anyone with RPC access reads all balances, txs, and contract state. There is no encryption.
  • Validators & node operators see everything — transactions, state, and (at the node) mempool and peer info.
  • Logs & backups. Node logs, snapshots, and explorer/indexer instances persist plaintext data.
  • The transaction graph among participants is fully visible to insiders.
  • Operational metadata. Who runs which validator, deployment patterns, and timing.

Threat model by adversary

Adversary Sees
Outsider (no RPC) nothing (if RPC truly restricted)
Authorized participant all on-chain data (plaintext)
Validator / node operator all data + mempool/peer metadata
Compromised RPC gateway everything flowing through it

Defense in depth

  1. Add data privacy. Deploy eERC on the L1 so even insiders can't see amounts/balances ([[combining-eerc-with-private-l1]]).
  2. Least-privilege RPC. Per-member credentialed gateways, not a shared endpoint ([[restrict-rpc-access]]).
  3. Harden node infra. Restrict log retention, secure snapshots/backups, lock down explorer/indexer access.
  4. Compartmentalize. Separate chains per trust domain if participants shouldn't see each other's data.
  5. Govern operators. Vet validator/node operators; they are trusted parties.

Decision: is a private L1 enough?

  • Outsiders are the only threat → private L1 alone may suffice.
  • Participants must not see each other's data → private L1 + eERC (or FHE) is required.

Common Pitfalls

  • Marketing "private" = "encrypted." It isn't; insiders see plaintext.
  • Trusting node operators implicitly. They see everything.
  • Forgetting logs/backups/explorers as leak vectors.

AI Agent Prompt

"Act as a privacy auditor for a permissioned L1. Enumerate what still leaks (insiders, validators, logs, backups, graph) despite restricted RPC, and design defense in depth including eERC for data confidentiality."

References

See references/README.md (academy permissioning 'access vs data privacy', node chain-configs).