The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's core execution model relies on parsing user-provided requirement specifications to drive subsequent architectural planning and agent implementation tasks.
Ingestion points: In Phase 0 (SKILL.md), the orchestrator is instructed to read requirement specifications from .claude/sprint/[N]/specs.md and iteration status from status.md to prepare the execution context.
Boundary markers: The instructions lack specific boundary delimiters or safety prompts to ensure that the content within the specifications does not override the agent's core workflow instructions.
Capability inventory: The workflow involves high-privilege capabilities including the dynamic spawning of multiple autonomous sub-agents (python-dev, nextjs-dev, qa-test-agent), the generation of new source and configuration files (api-contract.md, backend-specs.md), and the cleanup of temporary files (Phase 5).
Sanitization: The skill does not define validation, sanitization, or safety filtering for the content within the specs.md or status.md files before they are utilized to generate planning documents and agent spawns.

sprint-workflow