policy-author

name	policy-author
description	Use when designing a new ChainShield Policy for a wallet, treasury, or protocol multisig. Walks through risk profile, rule selection, default values, and remediation hookup so the resulting policy enforces real protection without blocking routine ops.

policy-author

A Policy is the contract between the user's intent and the risk gate. A bad policy either blocks legitimate work (loud, annoying, gets disabled) or allows the attack it was supposed to stop (silent, dangerous). This skill produces a policy that is right-sized for its owner.

Inputs to gather first

1. Owner address (the wallet under protection).
2. Class of owner: hot wallet, treasury multisig, agent-operated wallet, contract treasury.
3. Typical outflow per day in ETH-equivalent terms.
4. Largest legitimate single transfer the owner ever needs to make.
5. List of known-good destinations (cold vault, payroll splitter, audited DEX router, bridge, custody address).
6. List of known-good interaction protocols (Aave, Lido, Uniswap routers, Compound, etc.) with their main entry-point addresses.
7. Whether the owner ever needs to sign infinite approvals (rarely yes; usually no).
8. Available remediation playbooks in KeeperHub (revoke-approvals, safe-vault-evac, pause-automation).
9. Active notification channels (Discord webhook, Telegram bot, email).

Default values by owner class

Hot operations wallet (low balance, frequent activity)

{
  "rules": {
    "maxTransferEth": 0.5,
    "maxDailyOutflowEth": 2,
    "allowedDestinations": ["<known-good-1>", "<known-good-2>"],
    "forbiddenSelectors": ["0xa22cb465", "0xf2fde38b", "0x715018a6"]
  },
  "remediation": {
    "onBlock": ["revoke-all-approvals"],
    "notifyChannels": ["discord-ops"]
  }
}

Rationale: caps are tight relative to use; allowlist keeps daily ops smooth; admin-takeover selectors are blanket-forbidden.

Treasury multisig (high balance, infrequent activity)

{
  "rules": {
    "maxTransferEth": 50,
    "maxDailyOutflowEth": 100,
    "allowedDestinations": ["<cold-vault>", "<payroll-splitter>"],
    "forbiddenSelectors": ["0xa22cb465", "0xf2fde38b", "0x715018a6", "0x095ea7b3"],
    "approvalCapByToken": {
      "<usdc-address>": "1000000000000"
    }
  },
  "remediation": {
    "onBlock": ["safe-vault-evac", "revoke-all-approvals"],
    "notifyChannels": ["telegram-security", "discord-ops"]
  }
}

Rationale: per-tx cap reflects the largest legit operation; daily cap is 2x that to allow paired moves; raw approve is forbidden in favor of explicit per-token caps; remediation moves funds and revokes approvals on any block.

Agent-operated wallet (autonomous, narrow purpose)

{
  "rules": {
    "maxTransferEth": 0.1,
    "maxDailyOutflowEth": 1,
    "allowedDestinations": ["<single-target-protocol>"],
    "forbiddenSelectors": ["0xa22cb465", "0xf2fde38b", "0x715018a6", "0x095ea7b3"]
  },
  "remediation": {
    "onBlock": ["pause-automation", "safe-vault-evac"],
    "notifyChannels": ["discord-ops"]
  }
}

Rationale: the agent has one job; deviation from that job is suspicious and should pause the agent itself.

Authoring procedure

1. Start from the matching default above.
2. Replace maxTransferEth with the answer to "largest legit single transfer". Round up by ~20% to avoid false positives.
3. Replace maxDailyOutflowEth with 2 * maxTransferEth for treasuries, or actual measured throughput for hot wallets.
4. Fill allowedDestinations with the answers from inputs 5 and 6. Address-checksum is not required; the engine compares case-insensitively.
5. Always include the three admin-takeover selectors in forbiddenSelectors. Add 0x095ea7b3 (approve) only if you ALSO add an approvalCapByToken entry for any token the owner needs to interact with; otherwise the owner cannot grant any allowance at all.
6. Fill remediation.onBlock with playbook ids that already exist in KeeperHub. The engine attempts each in order until one succeeds.
7. Fill remediation.notifyChannels with channel keys registered in the running engine (engineOptions.notificationChannels). Unregistered keys are silently dropped.
8. Submit the policy via POST /policies. The response includes a UUID; reference it from /evaluate calls.

Validation

Before deploying, run these check intents through POST /evaluate:

Intent	Expected verdict
Safe transfer to allowlisted destination, value <= cap	ALLOW
Same call, value just above cap	BLOCK
Calldata starting with 0x095ea7b3 (approve) to a non-allowlisted token	BLOCK if approve is forbidden, else CONFIRM
Transfer to an unknown destination, value within cap	REQUIRE_HUMAN_CONFIRMATION

If any check disagrees, fix the policy before going live.

Common mistakes

• Setting maxTransferEth = maxDailyOutflowEth. The first legitimate tx exhausts the daily budget. Aim for at least 2x.
• Empty allowedDestinations "to be permissive". Empty disables the rule entirely; it does not allow everything subject to other rules. Be explicit.
• Forgetting forbiddenSelectors. The policy is only as strong as the rules that fire; admin-takeover selectors should never be on the path for routine ops.
• Putting playbook ids in onBlock that do not exist in KeeperHub. The engine logs the failure but the block still stands.
• Hardcoding addresses in lowercase from a checksummed source. The engine normalizes for you, but be consistent in the policy itself for diff readability.