The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is primarily designed to execute shell commands for code validation. It explicitly instructs the agent to find and run scripts located in the project's package.json file (e.g., pnpm lint, pnpm test). While these are standard developer tools, the execution of shell scripts from the local environment is a high-privilege action.
[INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it reads and interprets data from potentially untrusted sources in the local environment.
Ingestion points: The skill reads the content of package.json and the standard output/error logs generated by linting and testing tools.
Boundary markers: There are no boundary markers or instructions telling the agent to treat the tool output as non-executable data, which could allow maliciously crafted tool logs to influence agent behavior.
Capability inventory: The skill possesses the ability to read arbitrary files and execute shell commands defined in the local project configuration.
Sanitization: No sanitization or filtering is applied to the tool outputs or package.json content before the agent analyzes them to decide whether to proceed with a commit or release.

quality-guardian