The Agent Skills Directory

[COMMAND_EXECUTION] (MEDIUM): The sub-skills code-reviewer, code-archaeologist, and test-results-analyzer are provided with Bash tool access. The instructions specifically guide the agent to run project-specific commands like npm test, pytest, and go test. This creates a risk where malicious code in a project's build or test configuration (e.g., package.json scripts) could be executed by the agent.
[REMOTE_CODE_EXECUTION] (MEDIUM): The skill's primary function involves executing code from the external repositories it analyzes. In the context of an untrusted repository, this constitutes a vector for remote code execution within the agent's environment.
[Indirect Prompt Injection] (LOW): The skill lacks security boundaries when processing external code and logs. An attacker could embed instructions in code comments or metadata to manipulate the agent's analysis or output.
Ingestion points: Codebase files accessed via Read, Grep, and LS tools (e.g., in core_code-archaeologist.md).
Boundary markers: Absent; no delimiters or warnings to ignore instructions within analyzed files are present in the prompt templates.
Capability inventory: Bash, Write, Edit, MultiEdit, NotebookEdit, and TodoWrite tools across various sub-agents.
Sanitization: None; the agent processes and interprets raw file content and log data directly.

code-quality-specialist