The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes project plans and mission details through natural language prompts, creating a surface for indirect prompt injection.\n- [PROMPT_INJECTION]: Ingestion points: The plan_project(prompt) and create_mission(prompt) tools in SKILL.md accept untrusted natural language text.\n- [PROMPT_INJECTION]: Boundary markers: The instructions lack delimiters or explicit instructions for spawned agents to ignore embedded commands within mission prompts.\n- [PROMPT_INJECTION]: Capability inventory: Dispatched agents utilize the dispatch_mission tool to operate in isolated git worktrees with full tool access and automated merge capabilities.\n- [PROMPT_INJECTION]: Sanitization: No evidence of input validation, prompt sanitization, or safety filtering is provided in the skill definitions.

claude-devfleet