The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes Product Requirement Documents (PRDs) from the repository to generate task drafts, which constitutes a surface for adversarial content in project documentation to influence agent output.
Ingestion points: Reads PRD content from files located at .ai-workflow/work/<feature-slug>/prd.md.
Boundary markers: The workflow explicitly requires that the PRD status is approved before task generation begins, ensuring a prior human review step.
Capability inventory: The skill writes new task files to the tasks/ directory and updates tasks-draft.md. It also invokes local Python validation scripts.
Sanitization: Uses a validation script (shared/scripts/gate_contract.py) to verify gate records and a shared queue controller to validate the dependency graph and task bindings.
[COMMAND_EXECUTION]: The skill uses shell commands to invoke utility scripts for validating task contracts and dependencies.
Evidence: The instructions direct the agent to execute shared/scripts/gate_contract.py <task.md> and shared/scripts/task_queue.py during the planning phase.
[DYNAMIC_EXECUTION]: Project-specific Python scripts are executed at runtime to perform validation tasks that ensure the integrity of the generated task queue.
Evidence: Runtime execution of shared/scripts/gate_contract.py is required to validate Slice Contracts and versioned gate records.

to-tasks