What problem does it solve?
The GNATprove skill keeps agents from spinning aimless proof runs by giving them a structured campaign and investigative instructions needed to prove Ada/SPARK units without manual environment assumptions.
Core Features & Use Cases
- Environment discovery: Detects whether to invoke
alr gnatprove or a bare gnatprove, advises when to consult the user, and avoids hardcoded paths.
- Proof campaign workflow: Guides agents through status files, quick-fix versus full campaign decisions, and subagents focused on individual subprograms.
- Deep reference library: Links to documents on GNATprove output, contracts, loops, ghost code, and refactoring so agents can learn the reasoning behind every recommendation.
- Use Case: When tackling SPARK code that fails at multiple points, the skill orchestrates subagent runs,
proof-status.md updates, and strategic review to safely close the campaign.
Quick Start
Ask the skill to run GNATprove on my SPARK project, focus on the declared subprogram, and summarize any outstanding proof obligations.