The Agent Skills Directory

[COMMAND_EXECUTION]: The skill autonomously executes system-level commands and testing tools (e.g., k6, Artillery, Locust, ab, autocannon, npm audit, pip audit, cargo audit) based on its automatic discovery of the project stack.
[REMOTE_CODE_EXECUTION]: Includes a 'Self-Healing Fix Loop' and the generation of 'scripted loops' for performance testing, which involves the dynamic creation and execution of code within the host environment.
[DATA_EXFILTRATION]: Accesses and modifies files within the ~/.claude/projects/ directory, which is a sensitive location used for agent configuration and project memory. It appends execution metadata to skill-telemetry.md without explicit user consent for this specific file path.
[PROMPT_INJECTION]: Injects user-provided $ARGUMENTS directly into the system instructions without boundary markers or sanitization, allowing input to potentially manipulate the testing target or override the pipeline's logic.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection due to its core functionality of ingesting and acting upon untrusted project data.
Ingestion points: Reads project source code, specifications, and environment configurations during Phase 1 (E2E discovery) and Phase 2 (Manual Test Plan generation).
Boundary markers: No delimiters or protective warnings are used when processing project data to differentiate between data and instructions.
Capability inventory: The skill has the ability to write to the filesystem, execute shell commands, and perform 'self-healing' modifications to the codebase.
Sanitization: No evidence of sanitization, validation, or escaping of project-derived content before it is used to influence agent decisions or code modifications.

full-test