X Layer Execution Guard#

Use this skill to turn an agent swap intent into a guarded execution decision and proof artifact.

The installed skill includes its own Python runtime in runtime/ and two scripts in scripts/:

• scripts/run_execution_guard.py runs the guard CLI.
• scripts/check_agentic_wallet.py checks whether onchainos and Agentic Wallet are reachable.

Operating Rules#

1. Default to --no-execute or --execution-mode proof while evaluating a request.
2. Only use --live or --execution-mode agentic-wallet after explicit user approval for a real wallet action.
3. Treat proof mode as simulated execution evidence. Only agentic-wallet mode can return a real transaction hash.
4. If OnchainOS API credentials are missing, the runtime returns a mock install-smoke result. Do not present mock output as live proof.
5. For live X Layer runs, use chain 196 unless the user explicitly chooses another chain.

OpenClaw Quick Start#

Set the skill directory first. In OpenClaw it is usually:

bash

SKILL_DIR="$HOME/.openclaw/skills/xlayer-execution-guard"

If installed into a local workdir with ClawHub, use:

bash

SKILL_DIR="$PWD/skills/xlayer-execution-guard"

Install the only runtime dependency if needed:

bash

python3 -m pip install --user -r "$SKILL_DIR/requirements.txt"

Check the Agentic Wallet environment:

bash

python3 "$SKILL_DIR/scripts/check_agentic_wallet.py"

Run a safe pre-execution judgment:

bash

PYTHONPATH="$SKILL_DIR/runtime" python3 "$SKILL_DIR/scripts/run_execution_guard.py" \
  --agent strategy-office \
  --intent-id strategy-office-round-001 \
  --from USDC \
  --to USDT \
  --amount 10 \
  --amount-mode readable \
  --slippage 0.5 \
  --max-impact 1.20 \
  --no-execute \
  --output guard-proof.json

Run proof-mode closed-loop evidence without touching the wallet:

bash

PYTHONPATH="$SKILL_DIR/runtime" python3 "$SKILL_DIR/scripts/run_execution_guard.py" \
  --agent strategy-office \
  --intent-id strategy-office-proof-001 \
  --from USDC \
  --to USDT \
  --amount 10 \
  --amount-mode readable \
  --execution-mode proof \
  --output guard-proof.json

Run a live Agentic Wallet execution only after user approval:

bash

PYTHONPATH="$SKILL_DIR/runtime" python3 "$SKILL_DIR/scripts/run_execution_guard.py" \
  --agent strategy-office \
  --intent-id strategy-office-live-001 \
  --from USDC \
  --to USDT \
  --amount 10 \
  --amount-mode readable \
  --slippage 0.5 \
  --execution-mode agentic-wallet \
  --wallet default \
  --chain 196 \
  --output guard-live-proof.json

--live is an alias for --execution-mode agentic-wallet.

Required Environment#

For real OnchainOS route judgment, export API credentials in the OpenClaw environment or ~/.config/onchainos.env:

bash

export ONCHAINOS_API_KEY="..."
export ONCHAINOS_API_SECRET="..."
export ONCHAINOS_API_PASSPHRASE="..."
export ONCHAINOS_CHAIN_INDEX="196"

For live wallet execution, onchainos must be installed and logged in:

bash

onchainos wallet login
onchainos wallet status

What The Guard Checks#

The runtime uses OnchainOS DEX Aggregator APIs for:

• token discovery through /api/v6/dex/aggregator/all-tokens
• liquidity source discovery through /api/v6/dex/aggregator/get-liquidity
• aggregated and per-DEX quote comparison through /api/v6/dex/aggregator/quote

The output includes:

• pre_execution.verdict: execute, resize, retry, or block
• pre_execution.checks: quote availability, price impact, fallback coverage, token tax, honeypot flags, gas and fee fields
• execution.status: simulated_success, success, broadcasted, failed, or not_executed
• post_execution.proof_id and moltbook_summary
• closed_loop_validation: whether the pre-execution verdict matched the post-execution outcome

Failure Handling#

• Missing API credentials: output is mock mode and suitable only for install verification.
• wallet_ready=false: run onchainos wallet login in the same OpenClaw environment.
• verdict=block: do not execute; show the block reason.
• verdict=resize: execute only if the user accepts the resized risk outcome.
• Live execution returns no tx hash: preserve the full JSON output and inspect execution.error.