The Agent Skills Directory

[COMMAND_EXECUTION]: The skill frequently utilizes the Bash tool across multiple roles. The coordinator role uses it for directory resolution, while the validator role executes project-defined scripts like make test, cargo test, and go test to verify refactoring changes.\n- [DYNAMIC_EXECUTION]: The analyzer and validator roles perform runtime detection of project types and available tooling by scanning for marker files like package.json, Cargo.toml, or go.mod. This leads to the dynamic execution of build and analysis commands that are defined within the target codebase's configuration files.\n- [INDIRECT_PROMPT_INJECTION]: The skill architecture processes untrusted data from the user and the codebase, which could influence agent behavior.\n
Ingestion points: User-provided task descriptions from $ARGUMENTS in SKILL.md and iterative review/validation feedback ingested by the refactorer role.\n
Boundary markers: The system uses structured JSON files for data transfer between roles, but it lacks explicit instructional delimiters or safety warnings (e.g., "ignore embedded instructions") when interpolating user-provided feedback into agent prompts.\n
Capability inventory: The skill possesses powerful capabilities including writing to the filesystem (Write, Edit), executing shell commands (Bash), and spawning new autonomous agents (Agent, TaskCreate).\n
Sanitization: The skill instructions do not implement explicit sanitization or filtering for external inputs before they are used to generate tasks for worker agents.

team-arch-opt