The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The installation instructions include a command that downloads and executes a script directly from the internet (curl -sSf https://gitlawb.com/install.sh | sh). This is a dangerous pattern as it allows for arbitrary code execution from a source outside of standard package management verification.
[CREDENTIALS_UNSAFE]: The documentation instructs users to pass highly sensitive credentials, specifically an Ethereum private key ($ETH_PRIVATE_KEY), as a command-line argument for several operations like gl name register. This practice is insecure because command-line arguments can be captured in shell history files, process lists, or system logs.
[EXTERNAL_DOWNLOADS]: The skill requires the installation of external software and packages (@gitlawb/gl, @gitlawb/opencode, and the gl binary via Homebrew) from sources that are not recognized as trusted technology providers. This increases the risk of supply chain attacks.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process external, potentially attacker-controlled content through pull requests, issues, and agent task payloads. A malicious actor could embed instructions in these fields to manipulate the agent's behavior.
Ingestion points: Pull request descriptions (gl pr create --body), issue bodies (gl issue view), and agent task payloads (gl task create --payload).
Boundary markers: None identified in the provided instructions; untrusted content is passed directly as strings.
Capability inventory: The skill has capabilities to execute shell commands via the gl CLI, perform network operations to node.gitlawb.com, and modify files via git operations.
Sanitization: There is no evidence of sanitization, escaping, or validation being applied to the external content before it is processed by the agent tools.

gitlawb