The Agent Skills Directory

[Command Execution] (HIGH): The skill executes arbitrary shell commands retrieved from project manifest files (e.g., npm run build, pytest). If the agent is used to analyze or 'verify' a malicious repository, an attacker can embed destructive commands within these configuration files which the skill will then execute on the host system.
[Indirect Prompt Injection] (HIGH):
Ingestion points: The skill reads package.json, pyproject.toml, Cargo.toml, and various linter/type-check configurations.
Boundary markers: Absent. There are no mechanisms to prevent the execution of embedded instructions found within the project data.
Capability inventory: Full subprocess execution for npm, uv, python, cargo, and several security scanners across all supported languages.
Sanitization: None. The skill implicitly trusts the 'scripts' and 'commands' defined in the project structure.
[Dynamic Execution] (MEDIUM): The logic dynamically constructs command strings based on detected project types and configuration presence, increasing the risk of command injection if file names or script names are manipulated.

verification-loop