The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill acts as a synthesis tool that converts user-led technical discussions into an orchestration prompt for a separate, more capable AI session. This creates a risk where malicious or contradictory instructions provided by the user during the planning phase could be encapsulated into the high-privilege instructions for the implementation phase.\n- Ingestion points: Technical requirements, architecture decisions, and scope definitions provided by the user in SKILL.md during the iterative discussion phases (Explore and Refine sections).\n- Boundary markers: The skill uses instructional headers and requires a specific prompt frame defined in references/execution-prompt-contract.md, but lacks strict technical delimiters or "ignore embedded instructions" warnings to isolate user-provided content from the structural instructions in the final output.\n- Capability inventory: The generated Session 2 prompt is designed for an environment with broader capabilities, including repository modification, multi-agent spawning (using Sol and Terra models), and verification execution as specified in the orchestration and model strategy sections.\n- Sanitization: No explicit sanitization, validation, or escaping of user-provided strings is performed before they are interpolated into the implementation prompt contract.

feature-orchestrator