ordercli-hardened
Warn
Audited by Gen Agent Trust Hub on Apr 21, 2026
Risk Level: MEDIUMEXTERNAL_DOWNLOADSCREDENTIALS_UNSAFEDATA_EXFILTRATIONCOMMAND_EXECUTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill installs the
orderclibinary from a third-party GitHub repository (steipete/ordercli) using the Go package manager or Homebrew. This introduces a dependency on external code that is not from a pre-approved trusted organization. - [CREDENTIALS_UNSAFE]: The tool is designed to manage and ingest highly sensitive authentication materials, including user passwords (via
--password-stdin), browser cookies imported directly from Chrome profiles, and bearer tokens for Deliveroo and Foodora. This creates a significant risk of credential exposure if the agent's environment is compromised. - [DATA_EXFILTRATION]: The skill accesses personally identifiable information (PII) including order history, delivery addresses, and status updates. Although the instructions include explicit guardrails forbidding the agent from piping this data to network endpoints, the underlying capability to read this data and perform network operations exists.
- [COMMAND_EXECUTION]: The skill executes shell commands to interact with the system, including reading from browser application support directories (
$HOME/Library/Application Support/ordercli/) and managing local configuration files. - [PROMPT_INJECTION]: The
SAFETY.mdfile and the skill metadata contain multiple self-authoritative claims of being 'hardened' or 'armored.' These instructions are designed to influence the agent's behavior and the evaluation of the skill's safety profile. - [DATA_EXFILTRATION]: The skill processes potentially untrusted data from delivery service APIs that could contain indirect prompt injections.
- Ingestion points: Output from
orderclicommands such as history and order lists (SKILL.md). - Boundary markers: Absent. The skill does not use delimiters to wrap the output of the CLI tool.
- Capability inventory: Shell execution capabilities for binary installation and service interaction (SKILL.md).
- Sanitization: Absent. There is no evidence of sanitization for the data retrieved from external APIs before it is processed by the agent.
Audit Metadata