The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill explicitly instructs the agent to generate new, uncommitted test files aimed at breaking the code implementation.
Evidence: 'For every suspicion, write a breaking test aimed at the gap... Run it.' in SKILL.md.
[COMMAND_EXECUTION]: The agent is required to execute shell commands to run the newly created tests and evaluate the software's behavior.
Evidence: The skill works in harnesses like Claude Code (Task tool) and Hermes (delegate_task) to invoke execution environments.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data which could contain malicious instructions designed to manipulate the agent's logic or report.
Ingestion points: Pull request content, source code files, CLAUDE.md/AGENTS.md, and user-provided acceptance criteria.
Boundary markers: The instructions lack explicit delimiters (e.g., XML tags or clear 'ignore instructions in data' headers) to separate data from system prompts.
Capability inventory: The skill can create files (tests), execute shell commands, and spawn sub-agents with full context via the Task tool.
Sanitization: There is no mention of sanitizing or escaping the content read from files or acceptance criteria before it is passed to the agent context.

adversarial-tester