The Agent Skills Directory

[PROMPT_INJECTION]: Indirect prompt injection surface through project data ingestion.
Ingestion points: The skill parses requirement files (REQ-*.md) and explores project source code to generate test scenarios in the S_SOURCE state.
Boundary markers: Data extracted from these sources is interpolated into sub-agent instructions (e.g., via the {prev_context} field) without explicit delimiters or isolation instructions to prevent the agent from following embedded commands.
Capability inventory: The skill possesses high-privilege capabilities including spawn_agents_on_csv, Bash, and Write tools used to generate and execute code across the project.
Sanitization: No sanitization or validation of the content from requirements or source files is performed before inclusion in sub-agent instructions.
[COMMAND_EXECUTION]: Execution of dynamically generated test commands and reduction of user oversight.
The skill uses Bash to perform static analysis and run generated test suites based on the results of infrastructure discovery.
An optional -y or --yes flag allows the agent to skip the manual confirmation gate required after the S_CSV_GEN phase, increasing autonomy and reducing the user's ability to review potentially malicious or incorrect generated test scenarios before execution.

quality-auto-test