aavegotchi · cinnabarhorse · Feb 18, 2026
diff --git a/README.md b/README.md
@@ -13,3 +13,12 @@ Skill contents:
 Install targets (kept in sync separately):
 - Local (Clawd): `/Users/coderdan/Documents/GitHub/clawd-workspace/skills/aavegotchi-baazaar/`
 - VPS (OpenClaw): `/home/agent/.openclaw/skills/aavegotchi-baazaar/`
+
+## Optional Secure Alternative: Coinbase CDP Server Wallet
+
+This repo still documents and supports the current private-key broadcast flow.
+
+For teams that want policy-controlled server signing and no raw private key in `.env`, see:
+- `aavegotchi-baazaar/references/cdp-integration.md`
+
+This is docs-only integration guidance, not a hard runtime migration.
diff --git a/aavegotchi-baazaar/SKILL.md b/aavegotchi-baazaar/SKILL.md
@@ -279,3 +279,12 @@ ERC1155 list (simulate):
 - `ERC1155Marketplace: Not approved` / approval errors: missing `setApprovalForAll` for listing, or missing ERC20 `approve` for buying.
 - Swap errors (e.g. `LibTokenSwap: swapAmount must be > 0`): bad swapAmount math or missing inputs.
 - Listing cancelled/sold or price changed: refetch from subgraph and re-simulate before broadcasting.
+
+## Optional: Integrate Coinbase CDP Server Wallet (No PRIVATE_KEY in .env)
+
+This skill currently supports the existing `cast send --private-key` flow documented above.
+
+If you want a hardened alternative with no raw private key in `.env`, follow:
+- `references/cdp-integration.md`
+
+That guide covers encrypted credential files (`CDP_CREDENTIALS_PATH`), import-vs-create wallet setup, policy allowlists, and a two-step prepare/execute intent pattern.
diff --git a/aavegotchi-baazaar/references/cdp-integration.md b/aavegotchi-baazaar/references/cdp-integration.md
@@ -0,0 +1,152 @@
+# Optional Integration: Coinbase CDP Server Wallet (No `PRIVATE_KEY` in `.env`)
+
+This guide is an optional security upgrade for signing and broadcasting transactions.
+
+It does **not** replace the existing skill flow today:
+- Existing `cast call` / `cast send --private-key` instructions in this repo remain supported.
+- Use this guide when you want policy-controlled server signing and no raw private key in environment variables.
+
+## Why Use CDP Here
+
+Threat model improvements:
+- No raw EOA private key in `.env`.
+- Transaction policy controls (chain, contract, method, spend constraints).
+- Deterministic prepare -> execute flow to reduce phishing/prompt-injection risk.
+
+References:
+- CDP Server Wallet v2 Quickstart: https://docs.cdp.coinbase.com/server-wallets/v2/quickstart
+- CDP Policy Engine API: https://docs.cdp.coinbase.com/api-reference/v2/rest-api/policy-engine/create-a-policy
+- OpenClaw tool approvals: https://docs.openclaw.ai/tools/exec-approvals
+
+## Prerequisites
+
+- Coinbase CDP project with Server Wallet access.
+- Base mainnet target (`chainId=8453`).
+- Existing Baazaar config from `SKILL.md` (`DIAMOND`, `GHST`, `USDC`, `SUBGRAPH_URL`).
+- `cast` installed for calldata/selector checks.
+
+## Secure Credentials Pattern (Encrypted File Path)
+
+Use a credentials file path, not plaintext env vars:
+
+```bash
+export CDP_CREDENTIALS_PATH="$HOME/.openclaw/secrets/cdp-baazaar.json.gpg"
+export CDP_DECRYPT_CMD='gpg --quiet --decrypt'
+```
+
+Example decrypted JSON:
+
+```json
+{
+  "apiKeyId": "cdp_api_key_id",
+  "apiKeySecret": "cdp_api_key_secret",
+  "walletSecret": "cdp_wallet_secret",
+  "networkId": "base-mainnet",
+  "accountAddress": "0xYourSignerAddress",
+  "policyId": "optional_policy_id"
+}
+```
+
+Recommended file mode before encryption: `chmod 600`.
+
+## Wallet Bootstrap: Import vs Create
+
+### Path A: Import existing EOA (recommended when preserving approvals/state)
+
+Use this when you want to preserve the current `FROM_ADDRESS` balances and token approvals.
+
+1. Export encrypted source key (one-time migration material).
+2. Import into CDP Server Wallet account using CDP tooling/API.
+3. Verify imported address matches current `FROM_ADDRESS`.
+
+### Path B: Create a new CDP wallet
+
+Use this when rotating infrastructure and starting from a new signer.
+
+1. Create wallet/account in CDP.
+2. Fund it and re-approve `GHST`/`USDC`/NFT approvals to `DIAMOND`.
+3. Update automation to use the new `accountAddress`.
+
+## Policy Setup Guidance (Baazaar)
+
+Restrict `evm.sendTransaction` to:
+- Chain: Base mainnet only.
+- Contract allowlist: `DIAMOND`, `GHST`, `USDC`, approved NFT contracts.
+- Method allowlist: only Baazaar-required methods.
+
+Selector derivation examples:
+
+```bash
+cast sig "approve(address,uint256)"
+cast sig "setApprovalForAll(address,bool)"
+cast sig "executeERC721ListingToRecipient(uint256,address,uint256,uint256,address)"
+cast sig "executeERC1155ListingToRecipient(uint256,address,uint256,uint256,uint256,address)"
+cast sig "swapAndBuyERC721(address,uint256,uint256,uint256,uint256,address,uint256,uint256,address)"
+cast sig "swapAndBuyERC1155(address,uint256,uint256,uint256,uint256,address,uint256,uint256,uint256,address)"
+cast sig "addERC721Listing(address,uint256,uint256,uint256)"
+cast sig "setERC1155Listing(address,uint256,uint256,uint256,uint256)"
+```
+
+If your policy system supports spend caps, set:
+- Per-tx max value.
+- Daily aggregate max value.
+
+## Prepare -> Execute Intent Pattern (Anti-Phishing)
+
+Use a two-step flow:
+
+1. `prepare`: build tx from validated onchain/subgraph state.
+2. `execute`: require explicit operator confirmation on the frozen intent.
+
+Recommended guardrails:
+- Intent TTL (example: 300s).
+- Intent hash must include chainId, `to`, calldata, value, and state snapshot IDs.
+- Reject execution if listing state drifted since prepare.
+
+Example intent record:
+
+```json
+{
+  "intentId": "sha256:...",
+  "action": "execute_erc721_listing",
+  "chainId": 8453,
+  "to": "0xA99c4B08201F2913Db8D28e71d020c4298F29dBF",
+  "data": "0x...",
+  "value": "0x0",
+  "createdAt": "2026-02-18T00:00:00.000Z",
+  "expiresAt": "2026-02-18T00:05:00.000Z",
+  "stateProof": {
+    "listingId": "123",
+    "priceInWei": "...",
+    "subgraphBlock": "..."
+  },
+  "status": "prepared"
+}
+```
+
+## OpenClaw Hardening Checklist
+
+- Keep wallet/broadcast tool disabled by default; allowlist only when needed.
+- Keep command execution approvals enabled for any broadcast path.
+- Never run dynamic user strings (`eval`, `bash -c`, `sh -c`).
+- Validate all addresses and uint inputs before encoding calldata.
+
+References:
+- https://docs.openclaw.ai/tools
+- https://docs.openclaw.ai/sandboxing
+- https://docs.openclaw.ai/tools/exec-approvals
+
+## Troubleshooting
+
+- Policy deny: verify chain, destination, method selector, and value caps.
+- Address mismatch: confirm imported/created CDP account address.
+- Replay/idempotency issues: bind idempotency key to `intentId`.
+- State drift: re-run `prepare` after listing changes.
+
+## Fallback (Current Legacy Flow)
+
+If CDP integration is unavailable, continue using existing documented flow in `SKILL.md`:
+- Simulate with `cast call`.
+- Broadcast with `cast send --private-key` only when explicitly instructed.
+
+This repo has **not** hard-migrated runtime behavior in this change.