This directory contains LLM prompt specifications that guide AI-assisted specification management for the uBPF project. Each workflow defines a structured, multi-phase process executed by an AI agent operating under a “Senior Systems Engineer” persona with built-in guardrails against hallucination, fabrication, and specification drift.
All three workflows share a common foundation:
bootstrap-specs.md — Spec ExtractionExtracts requirements, design, and validation specifications from an existing codebase that lacks formal documentation.
Phases: Repository scan → Draft extraction (requirements, design, validation) → Human clarification → Consistency audit → Human approval → Deliverable creation.
Use when: The project has working code but no formal specs, and you need to create the initial specification suite.
evolve-specs.md — Incremental DevelopmentEvolves specifications alongside new feature development, ensuring that specs, code, and tests stay synchronized as changes are made.
Phases: Requirements discovery → Specification changes → Spec audit → User review → Implementation changes → Implementation audit → User review → Deliverable creation.
Use when: Adding new features or modifying existing behavior and you need specs, implementation, and tests to advance together.
maintain-specs.md — Maintenance AuditAudits the full specification suite against the current codebase and tests to detect and correct drift.
Phases: Full audit (document-level, code-level, test-level) → Human classification of findings → Corrective patch generation → Patch audit → Human approval → Deliverable creation.
Use when: Specs, code, or tests may have diverged over time and you need to bring them back into alignment.
These workflows are designed to be provided as system prompts (or prompt context) to an LLM agent with tool access to the repository. Each workflow expects specific inputs (e.g., file paths, specification documents) documented in its Inputs section.