The Agent Skills Directory

[COMMAND_EXECUTION]: The skill makes extensive use of the Bash tool to perform framework discovery, execute test suites (e.g., vitest, jest, pytest), and manage containerized services via docker compose or testcontainers.
[COMMAND_EXECUTION]: Uses the CronCreate tool to establish a persistent "Coverage Drift Monitor" that periodically executes shell commands to track coverage regressions, which persists the agent's activity across sessions.
[REMOTE_CODE_EXECUTION]: Implements a PreToolUse hook that executes an external script (run-hook.mjs) through the shell, allowing for arbitrary code execution whenever the Bash tool is invoked.
[PROMPT_INJECTION]: The "Heal Loop" workflow (Phase 5) creates a surface for indirect prompt injection by reading potentially untrusted source code and test logs to generate fixes.
Ingestion points: Phase 5 reads project source code and test execution results from the file system.
Boundary markers: There are no defined delimiters or instructions provided to the generator agents to distinguish between the code being tested and the agent's instructions.
Capability inventory: The agent possesses powerful capabilities including Bash, Write, Edit, and CronCreate that could be exploited if malicious instructions are processed.
Sanitization: No evidence of input validation or sanitization exists for the content ingested from the project environment before it is passed to sub-agents.

cover