Skip to content

SKILL.md

Latest commit

 

History

History
251 lines (185 loc) · 11.1 KB

SKILL.md

File metadata and controls

251 lines (185 loc) · 11.1 KB
Raw
name zerion-agent-management
description Manage Zerion agent tokens and security policies — the primitives for autonomous trading and signing. Create / list / use / revoke agent tokens; create / list / show / delete policies (chain locks, allowlists, transfer/approval gates, expiry). Use whenever the user asks to set up an agent token, configure a policy, or enable autonomous trading. Required by `zerion-trading` and `zerion-sign`.
license MIT
allowed-tools Bash

Zerion — Agent Token & Policy Management

Agent tokens authorize the CLI to sign transactions on behalf of a wallet without a passphrase prompt. Policies attached to a token scope what the token can do (chains, allowlist, deny rules, expiry). This is the foundation for safe autonomous trading and off-chain signing.

Setup

If a zerion command fails with command not found, install once:

npm install -g zerion-cli

Requires Node.js ≥ 20. For auth see the zerion umbrella skill. To execute trades after setup → zerion-trading. To sign messages/typed-data → zerion-sign.

When to use

  • "Create an agent token"
  • "Set up autonomous trading"
  • "Restrict the bot to chain X / address Y"
  • "Block transfers" / "block approvals" / "expire the token in N days"
  • "Switch which agent token is active"
  • "Revoke a token"

Agent vs manual operations

Operation Type Notes
agent list-tokens, agent list-policies, agent show-policy, agent use-token Agent Read-only or config-only. Safe autonomously.
agent create-token (default) Manual Interactive passphrase prompt. Humans run it.
agent create-token --passphrase-file <path> Agent-capable Reads passphrase from a chmod 600 file. For CI / headless / scripted setup once the file is provisioned by a human or secrets manager.
agent revoke-token, agent create-policy, agent delete-policy Manual Require passphrase or confirmation. Humans must run these directly.

Read-only — agents may invoke freely

zerion agent list-tokens              # Tokens, attached policies, active flag, wallet binding
zerion agent list-policies            # All policies with rules summary
zerion agent show-policy <id>         # Full policy details
zerion agent use-token --wallet <wallet>   # Switch the active token (config edit, no passphrase)

Token + policy management

The default flow is human-interactive. agent create-token also has a non-interactive variant (--passphrase-file) for automation once the passphrase file is provisioned. Everything else in this section requires a human.

Create an agent token

# Interactive policy picker (recommended)
zerion agent create-token --name <bot> --wallet <wallet>

# Attach an existing policy by ID
zerion agent create-token --name <bot> --wallet <wallet> --policy <id>

# Multiple policies (AND semantics — all must pass)
zerion agent create-token --name <bot> --wallet <wallet> --policy <id1>,<id2>

The token is auto-saved to ~/.zerion/config.json under agentTokens and (if no token was active before) becomes the default. Trading commands (zerion-trading) and signing commands (zerion-sign) read it from config.

Non-interactive token creation (--passphrase-file)

--passphrase-file <path> takes a path to a plain-text file whose entire contents are the wallet passphrase (the same string a human would type at the TTY prompt). The CLI reads the file once and uses its contents to unlock the keystore. Nothing else lives in the file — no JSON, no key=value, no header.

For CI, headless servers, or scripted agent setup, write the passphrase to such a file instead of typing it:

# Write the passphrase to a file readable only by you
umask 077                                   # ensure new files default to 0600
printf '%s' "$ZERION_PASSPHRASE" > ~/.zerion-pass
chmod 600 ~/.zerion-pass                    # required — refused if looser

# Create the token non-interactively
zerion agent create-token \
  --name bot --wallet my-agent \
  --policy <id> \
  --passphrase-file ~/.zerion-pass

# Clean up immediately
shred -u ~/.zerion-pass 2>/dev/null || rm -f ~/.zerion-pass

For ephemeral storage (recommended on Linux), use a tmpfs mount instead of $HOME:

# Linux only — /run is tmpfs (RAM-only, gone on reboot)
sudo install -d -m 0700 -o "$USER" /run/zerion
printf '%s' "$ZERION_PASSPHRASE" > /run/zerion/pass && chmod 600 /run/zerion/pass
zerion agent create-token ... --passphrase-file /run/zerion/pass

On macOS, use /private/tmp/<your-user>/... or a per-app dir under ~/Library/Caches/. macOS does not ship a user-writable /run.

File format

Plain UTF-8 text. The entire file (minus exactly one optional trailing \n or \r\n) is treated as the passphrase.

Rule Reason
Content = raw passphrase bytes only No JSON, no key=value, no comments, no quotes. Anything in the file is part of the passphrase.
One trailing \n or \r\n stripped So echo "pass" > file works as expected. Additional newlines are kept.
Leading / trailing spaces inside the passphrase are preserved Passphrases may legitimately contain them. CLI does not .trim().
UTF-8 encoded Read via readFileSync(path, "utf8"). Non-UTF-8 bytes become replacement chars.
Non-empty Empty file or newline-only file → rejected.
Mode 0600 (POSIX) Refused otherwise — see Rules below.
Regular file, owned by current uid Symlinks-to-files are followed; the target must still pass perm + ownership checks.

Examples (file bytes → passphrase used by CLI):

Command File bytes Passphrase result
printf '%s' 'hunter2' > f hunter2 hunter2 ✅
printf '%s\n' 'hunter2' > f hunter2\n hunter2 ✅
echo 'hunter2' > f hunter2\n hunter2 ✅
echo '"hunter2"' > f "hunter2"\n "hunter2" (quotes included!) ❌
echo ' spaces ' > f spaces \n spaces (spaces kept) ✅
printf '' > f (empty) rejected ❌
printf '\n' > f \n rejected ❌

Canonical form — no trailing newline, no surprises:

umask 077
printf '%s' 'YOUR-PASSPHRASE' > ~/.zerion-pass
chmod 600 ~/.zerion-pass

Verify before using:

wc -c ~/.zerion-pass     # byte count == passphrase length
xxd ~/.zerion-pass | head # eyeball raw bytes — no BOM, no quotes, no CRLF

Rules

  • File must be mode 0600 (owner read/write only). Any group/other bits → CLI refuses with passphrase_file_error.
  • File must be owned by the current uid (POSIX). Reading another user's file (e.g. via symlink) is refused. Matches SSH IdentityFile behavior.
  • Trailing newline (\n or \r\n) is stripped; leading/trailing spaces inside the passphrase are preserved.
  • Empty file is rejected.
  • Permission and ownership checks are skipped on Windows (POSIX mode bits and uid not meaningful there) — use NTFS ACLs to restrict access instead.

Recommended patterns:

Environment Source Notes
Local dev ~/.zerion-pass (chmod 600) Delete after use.
Docker Bind-mount a tmpfs file --tmpfs /run/zerion:mode=0700 then write passphrase inside.
Kubernetes Mount a Secret as a file Default mount mode is 0644 — set defaultMode: 0600 in the volume spec.
GitHub Actions Write ${{ secrets.X }} to a temp file, chmod 600, then run Cleanup is automatic at job end.
HashiCorp Vault vault agent template renders to a 0600 file Renew & rotate centrally.

Threat model: same as an SSH private key on disk. Anyone with active-session read access to the file can unlock the keystore. Keep the file path off shared filesystems, out of git, out of process listings. Argv-passed passphrases (--passphrase <value>) are not supported because they leak to ps aux, shell history, and CI logs.

--passphrase-file only affects agent create-token for now. wallet create, wallet import, and other passphrase-gated commands still require an interactive TTY.

Revoke a token

zerion agent revoke-token --name <bot>
zerion agent revoke-token --id <id>

Create a policy

A policy is a set of rules a token must pass before signing. At least one rule required.

# Chain lock — restrict to specific chains
zerion agent create-policy --name safe-base --chains base,arbitrum
zerion agent create-policy --name monad-only --chains monad

# Expiry — token deactivates after duration (e.g. 24h, 7d)
zerion agent create-policy --name short-lived --expires 7d

# Block raw native transfers (ETH/SOL send command)
zerion agent create-policy --name no-transfers --deny-transfers

# Block ERC-20 approvals (prevents allowance grants)
zerion agent create-policy --name no-approvals --deny-approvals

# Allowlist — only allow interaction with specific contract addresses
zerion agent create-policy --name dex-only --allowlist 0xUniRouter,0xCowSwap

# Combined rules (AND semantics)
zerion agent create-policy --name strict \
  --chains base \
  --expires 7d \
  --deny-transfers \
  --deny-approvals

Delete a policy

zerion agent delete-policy <id>

Tokens that referenced the deleted policy will no longer pass that rule check.

Policy reference

Flag Effect
--chains <list> Restrict to listed chains (comma-separated)
--expires <duration> Token deactivates after duration. Format: <n>h, <n>d (e.g. 24h, 7d)
--deny-transfers Block raw native transfers (zerion send of native asset)
--deny-approvals Block ERC-20 approve calls
--allowlist <addrs> Only allow interaction with listed contract addresses (comma-separated)

Policies execute as locally-spawned scripts (policies/*.mjs in the CLI repo). They run on every signing attempt before the transaction is built.

Recommended setup pattern

# 1. Create wallet (manual — see zerion-wallet skill)
zerion wallet create --name agent-bot

# 2. Create a tight policy
zerion agent create-policy --name swap-only \
  --chains base,arbitrum \
  --expires 30d \
  --deny-transfers \
  --allowlist 0xUniswapRouter,0x1inchRouter

# 3. Create the agent token bound to that policy
zerion agent create-token --name agent-bot \
  --wallet agent-bot \
  --policy <swap-only-id>

# Now zerion-trading and zerion-sign work autonomously

Common errors

Code Cause Fix
no_agent_token No active token for wallet agent create-token --wallet <wallet>
agent_token_expired Policy --expires lapsed Create a fresh token
policy_denied Action blocked by policy agent show-policy <id> to see rules; revise or use a different token
policy_not_found Policy ID doesn't exist agent list-policies to find valid IDs
policy_no_rules create-policy with no flags Add at least one rule (--chains, --expires, --deny-*, --allowlist)
token_name_exists Duplicate --name Choose another name or agent revoke-token --name <bot> first
passphrase_file_error --passphrase-file path missing, wrong perms, or empty chmod 600 <path>; ensure file exists and is non-empty