The Agent Skills Directory

[SAFE]: The skill implements a structured software development lifecycle. The tools utilized, such as Bash for running tests and file tools for code modification, are appropriate and necessary for its stated purpose as a development orchestrator.\n- [COMMAND_EXECUTION]: The workflow uses the Bash tool to perform repository research (git log) and optional task management via the 'werk' CLI. These actions are triggered by user-provided task descriptions and represent standard operations in a developer's environment.\n- [PROMPT_INJECTION]: The skill processes project files during the research phase, creating an inherent surface for indirect prompt injection. This is mitigated by its intended use-case as a development tool. Assessment of surface: 1. Ingestion points: Stage 1 Research reads local files using Grep, Glob, and Read tools; 2. Boundary markers: Not explicitly defined in the instructions; 3. Capability inventory: Includes Write, Edit, Bash, and Agent tools for execution stages; 4. Sanitization: Not performed on the content of the files read.\n- [PROMPT_INJECTION]: Reference documentation provided with the skill (agentic-prompts.md, system-prompts.md) uses keywords like 'CRITICAL' or 'IMPORTANT' solely within educational examples to illustrate prompt engineering anti-patterns. These are not malicious instructions and do not attempt to bypass agent safety filters.

dev