Ardi Skill

AdvisoryAudited by Static analysis on May 7, 2026.

Overview

No suspicious patterns detected.

Findings (0)

Artifact-based informational review of SKILL.md, metadata, install specs, static scan signals, and capability signals. ClawScan does not execute the skill or run runtime probes.

What this means

A user may believe the wallet can only interact with two Ardi contracts, while the code can also support broader financial contract interactions.

Why it was flagged

This presents a narrow contract-scope safety claim. Provided source also shows buy-and-stake flows through DEX routers, ERC20 approvals, veAWP, and AWPAllocator, so the claimed boundary under-discloses actual signing targets.

Skill content
contract_allowlist: true     # Only ArdiNFT + ArdiEpochDraw on Base mainnet (chain id 8453); addresses are compiled into the binary
Recommendation

Update the security description to list all contract targets and transaction types, and require clear user confirmation for each non-minting financial action.

What this means

Installing the skill can run remote code that was not necessarily the exact reviewed version.

Why it was flagged

The bootstrap script fetches and executes an installer from the live main branch, so the installed code can differ from the artifact under review and is not pinned by version or checksum.

Skill content
URL="https://raw.githubusercontent.com/awp-worknet/ardi-skill/main/install.sh" ... INSTALL_DIR="$INSTALL_DIR" sh "$TMP"
Recommendation

Pin installer and binary versions, verify checksums or signatures, and prefer executing the reviewed bundled installer instead of fetching from a moving branch.

ConcernHigh Confidence
ASI10: Rogue Agents
What this means

Mining activity may continue after setup and may consume gas or submit transactions without the user watching each cycle.

Why it was flagged

The skill can install a persistent unattended mining process, which is high-impact because the broader skill performs wallet-backed on-chain actions.

Skill content
auto-mine — install (or refresh) the unattended mining daemon ... On systemd hosts a user timer is now running.
Recommendation

Only enable auto-mine on a dedicated wallet/host, document exact cadence and budgets, and prominently surface the status and stop commands before starting the timer.

What this means

A mistaken or over-eager invocation could spend funds, approve tokens, or lock assets.

Why it was flagged

The command can perform real swaps, approvals, locks, and stake allocations. This is purpose-aligned for onboarding, and the code describes quote/confirmation flow, but it is financially sensitive.

Skill content
1) (optional) ETH → USDC → AWP swap ... 2) approve AWP, deposit into veAWP ... 3) allocate locked stake to the agent
Recommendation

Use quote-only mode first, confirm amounts and lock duration with the user, and avoid `--yes` unless the user explicitly approved the final transaction plan.

What this means

If awp-wallet is available and unlocked, the skill can request signatures for financially meaningful actions.

Why it was flagged

The skill does not claim direct private-key access, but it relies on an unlocked wallet bridge for transaction and typed-data signing.

Skill content
no_direct_key_access: true ... all signing flows through awp-wallet send-tx + sign-typed-data ... session_token_only: false    # awp-wallet is unlocked-by-default; no session-token model
Recommendation

Use a separate low-balance agent wallet, review each wallet prompt, and avoid keeping high-value funds in the signing wallet.