This skill is documentation-only guidance for the third-party `nla` CLI, scoped by `allowed-tools: Bash(nla:*)`. All 26 static findings are false positives: the flagged backtick/code-fence patterns are markdown formatting in a how-to guide, the reconnaissance flags are documented `nla escrow:*` usage examples, and the env/private-key mentions describe how a user supplies their own wallet key and LLM API keys to the CLI. No finding shows the skill itself reading, transmitting, or exfiltrating secrets, and no prompt-injection or data-exfiltration intent was found. Residual risk is inherent to the underlying tool (on-chain transactions cost real funds and require a private key), not to this skill's instructions.
The skill is not overtly malicious, but it operates a high-impact escrow arbitration CLI. Several Markdown backtick and reconnaissance alerts are false positives, while arbitration commands and secret handling are confirmed risks. Auto mode adds prompt-injection and irreversible on-chain decision exposure.
Auto mode uses an LLM to judge escrow demand and fulfillment text. Untrusted escrow text could steer the model and affect on-chain decisions.
The skill says the command displays demand and fulfillment text, then documents LLM auto mode. That creates a clear prompt-injection exposure even without a malicious sample in the file.
The skill allows API keys and private keys through command-line flags. Shell history or process listings can expose these secrets.
The file explicitly mentions API-key and private-key flags and includes an API-key flag example. The exposure path depends on the local shell and operating system.
Irreversible On-Chain Decisions in Broad Auto Mode
The skill documents automatic arbitration for all pending requests while noting that each decision becomes a permanent on-chain attestation.
The impact is clear from permanent attestations and all-pending auto examples. The exact harm depends on escrow value and user confirmation behavior in the CLI.
Capability review items (8)
These are real local capabilities that may be expected for this skill, so they require review but are not counted as confirmed malicious behavior.
This command can submit arbitration decisions for a specific escrow through an external CLI. Auto mode increases impact because decisions may be made without manual review.
In **auto mode** (`--auto`), the command uses the LLM provider/model specified in the escrow's deman
The line documents auto mode for the external arbitration command and says it can use provider credentials. This is a legitimate feature but carries execution and decision risk.
The prerequisite section references installing and configuring the external nla CLI. This is expected for the skill, but it is still an external-command dependency.