Back to skill

Security audit

Gasless Crosschain Executor

Security checks for vulnerabilities and agentic risk

Overview

The skill is a coherent crypto-swap workflow, but its install path depends on unreviewed external code for wallet and trading operations, so it should be reviewed before use.

Install only from a pinned, reviewed release or commit that includes the actual scripts, dependency manifests, lockfiles, and tests. Before using with real funds, verify the helper code locally, use a low-balance wallet or hardware-backed signer, never paste secrets into chat, and require the exact plan hash approval before any signature or broadcast.

Vulnerability Patterns
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Findings (1)

T03 · Remote Payload Retrieval and Execution

Error
Location
README.md:109
Finding

Mutable External Code Is Retrieved and Executed Without Verifiable Pinning

Content
View full analysis

Vulnerability Details

File Location: README.md:109-111, README.md:159, and README.md:181-184
Vulnerability Type: Remote payload retrieval and supply-chain exposure
Risk Level: High

Vulnerable Code

bash
git clone https://github.com/galpha-ai/gasless-crosschain-executor.git
cd gasless-crosschain-executor
./scripts/install.sh --target <agent>

The setup documentation further states:

bash
pip install -r requirements.txt

cd examples && npm install && cd ..

Technical Analysis

The installation procedure clones a mutable Git repository and subsequently executes its installation and setup scripts. It does not pin the clone to an audited commit, verify a cryptographic checksum or signature, or otherwise establish that the downloaded content matches the reviewed artifact.

The audited package contains only Markdown documentation and a sample JSON plan. It does not contain the referenced scripts/, examples/, dependency manifests, lockfiles, or test files. Consequently, the behavior of install.sh, setup.sh, Python dependencies, Node dependencies, wallet generators, signing helpers, and network wrappers cannot be verified from this artifact.

Although the documentation claims that SDK versions are pinned, that claim cannot be confirmed because the relevant manifests and lockfiles are absent. Running pip install and npm install against externally retrieved manifests also expands the trust boundary to package registries and transitive dependencies.

Attack Path

  1. An attacker compromises the upstream repository, a maintainer account, a mutable branch, or a referenced dependency.
  2. The attacker modifies scripts/install.sh, scripts/setup.sh, a helper script, or a dependency manifest.
  3. A user follows the documented installation procedure and clones the current mutable repository state.
  4. The user executes install.sh or setup.sh, or insta ...[truncated 1154 chars]
Remediation
View remediation

Remediation Suggestions

  1. Include every referenced script, example, manifest, lockfile, and test in the distributed Skill artifact so the effective payload can be audited.
  2. Pin installation to a reviewed Git commit hash or signed release tag rather than the repository's current default branch.
  3. Publish cryptographic checksums for release archives and verify them before extracting or executing any file.
  4. Sign releases and commits using a verifiable maintainer identity, and fail closed when signature verification fails.
  5. Pin all direct and transitive dependencies to exact versions using committed lockfiles.
  6. Use hash verification for Python dependencies, such as a fully pinned requirements file installed with pip --require-hashes.
  7. Use deterministic Node installation such as npm ci with a committed lockfile instead of unconstrained npm install.
  8. Do not automatically execute setup scripts during Skill installation. Present the exact commands and requested permissions for explicit user review.
  9. Run dependency installation and tests in a sandbox without wallet files, signing credentials, provider tokens, or access to production Agent configuration.
  10. Add CI checks that confirm the packaged artifact contains the exact reviewed files and that release checksums correspond to the signed source revision.
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (7)

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
75% confidence
Finding

Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.

Content

Scanner excerpt · README.md (reported line 126)May include surrounding context.

md
- `--link` — symlink the source instead of copying. Edits in this repo show up immediately in the agent. Recommended for development.
- `--no-setup` — skip `./scripts/setup.sh` after install (e.g. when the venv already exists at the target).
- `--force` — overwrite an existing install without asking.

For ClawHub / OpenClaw publishing:

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
85% confidence
Finding

Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.

Content

Scanner excerpt · README.md (reported line 161)May include surrounding context.

md
This creates `./.venv`, installs all Python deps from `requirements.txt`, runs `npm install` for the Fusion+ helpers when Node is available, and verifies the install by running the test suite. Re-run any time to upgrade deps in place.

Flags: `--no-tests` (skip verification), `--no-node` (skip npm even if Node is present), `VENV=/some/path ./scripts/setup.sh` (override venv location).

After setup, either activate the venv or run scripts directly:

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
80% confidence
Finding

Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.

Content

Scanner excerpt · SKILL.md (reported line 24)May include surrounding context.

md
## Non-negotiable safety model

- Never ask the user to paste a private key, seed phrase, mnemonic, raw keystore password, wallet backup, or one-time code into chat.
- Never print, log, store, or echo any private key material. Use a local environment variable, local key vault, hardware wallet, or wallet connector managed outside the conversation.
- Do not execute a real trade unless the user has reviewed an exact execution plan and explicitly authorized that exact plan.
- Do not sign arbitrary calldata or typed data unless the provider, chain ids, token addresses, spender or settler addresses, amounts, recipient, deadline, nonce, slippage, and cancellation or refund path have been checked.

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
80% confidence
Finding

The instruction to choose provider mode 'from observable state' and 'do not ask the user' gives the agent authority to make execution-path decisions that affect fees, gas requirements, relayer trust, approval needs, and failure modes. Even though the skill includes strong guardrails elsewhere, reducing user involvement in provider selection can cause the agent to route through a riskier or less desired execution path without explicit confirmation.

Content

Scanner excerpt · SKILL.md (reported line 52)May include surrounding context.

md
- Monitor or recover: use provider status APIs and cancellation or refund instructions.
   - Bootstrap a wallet: the user has no wallet or does not know how to configure one. Use the wallet bootstrap workflow before any other workflow that needs a signer.

2. Choose provider mode. The choice follows from observable state; do not ask the user. Validator enum: `{1inch-fusion-plus, 1inch-fusion, 1inch-aggregator, custom}`.

   Cross-chain (`source_chain_id != destination_chain_id`) → `1inch-fusion-plus`. Pipeline: `build_order_fusion_plus.mjs` → `local_signer.py` (typed-data) → `submit_fusion_plus.mjs` → `status_fusion_plus.mjs` → `submit_secret.mjs`.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

Line L062 explicitly says the agent 'should not ask which provider to use' and must decide automatically. This is a natural-language policy concern because it hard-codes behavioral choice without user opt-in, despite different providers having materially different execution characteristics.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

Line L083 states the agent 'should not ask the user which provider to use' and should derive the choice from preflight. This prescriptive natural-language instruction suppresses user choice over execution mode and provider selection.

Content

No source excerpt is available for this finding.

Vague Triggers

Low
Category
Not specified by scanner
Confidence
86% confidence
Finding

This JSON manifest-like file defines a sample plan but provides no explicit invocation conditions, trigger phrases, or exclusion rules. For manifest-scoped quality review, the absence of any trigger scope can make activation behavior unclear if this file is used as part of a skill definition.

Content

No source excerpt is available for this finding.