Compiling a Dungeon: A Real-World ISL Case Study

Compiling a Dungeon: A Real-World ISL Case Study

The Intent Specification Language (ISL) was used to generate a full-featured boardgame-style dungeon crawler as a stress test for spec-driven development. The project resulted in 56 generated source files totaling ~330 KB, all produced from 51 top-level ISL specifications by a single part-time developer.

Why This Matters

Traditional prompt-driven AI development often fails as projects scale due to context fragmentation and destructive re-generation of code. This case study demonstrates that replacing prompts with a compilable specification makes LLM-driven development predictable and maintainable, even as complexity grows beyond thirty components. By treating code as a read-only build artifact and the specification as the source of truth, integration drift is collapsed to near zero via generated TypeScript interface signatures (.sign.ts).

Key Insights

• The project generated 56 source files totaling ~330 KB from 51 top-level ISL specifications in 2026.
• A 7% bug-to-feature ratio was maintained over 73 commits, as fixes were pushed into the specification rather than the code (Marconi, 2026).
• Productivity tripled after the logic-test generation tooling was implemented on 2026-03-29, proving the immediate ROI of verification loops.
• Effect Lifecycles using Apply, Active While, Expires When, and Cleanup blocks in ISL 1.6.2 ensure deterministic state management for complex game logic.
• Context fragmentation is mitigated by generating signatures for dependencies, allowing a component to compile against real interfaces instead of the entire project context.

Practical Applications

• Use Case: High-complexity React application state management using ISL Role qualifiers. Pitfall: Failing to specify ‘Cleanup’ fields in effect lifecycles, which results in subtle state leaks.
• Use Case: Maintaining legacy data quirks by specifying leniency in the ISL runtime instead of forcing a rewrite. Pitfall: Deeply nesting dependency graphs which invalidates the entire build on minor root changes.
• Use Case: Behavioral testing via ISL logic-test specs rather than framework-specific unit tests. Pitfall: Treating regression specs reactively after bugs occur rather than writing them as initial acceptance criteria.

References:

• https://dev.to/francesco_marconi_282058b/compiling-a-dungeon-a-real-world-isl-case-study-1109
• https://dungeon-demo-isl.netlify.app