Domain: Runtime / Execution
Status: Canonical
The Drift Resolution Cycle is the practical engine of Context-Driven Engineering (CDE). It resolves the gap between human intent and repository reality through a deterministic loop of analysis, normalization, and processing.
- Analyze: Audit the repository context to identify the Drift from the original intent.
- Normalize: Transform ambiguous requests into a Bounded Workflow (
workflow.md). - Process: Execute the discrete steps using validated CLI primitives and scripts.
- Validate: Verify the final state against the repository truth (
dev.kit doctor). - Capture: Distill successful logic back into the repository as a reusable Skill.
To ensure high-fidelity results, dev.kit enforces a strict Normalization Boundary. Chaotic intent is never executed directly; it must be filtered into a structured workflow.md.
- Intent-to-Plan: Ambiguity is eliminated before execution begins.
- State Persistence: The current status (
planned | in_progress | done) is tracked at the repository level. - Fail-Open Resilience: Every workflow step includes a fallback mechanism for continuity during tool failures.
| Artifact | Role | Location |
|---|---|---|
plan.md |
The raw, normalized task objective. | .udx/dev.kit/tasks/<id>/ |
workflow.md |
The deterministic execution sequence. | .udx/dev.kit/tasks/<id>/ |
feedback.md |
The iterative engineering log. | .udx/dev.kit/tasks/<id>/ |
To maintain high-fidelity momentum across multi-turn interactions:
- Proactive Catch-Up: At the start of every session, agents identify unfinished tasks (
dev.kit task active). - Nudge Mechanism: The system proactively reminds users to resolve stale state or pending syncs.
- Clean Handoff: Completed tasks are pruned from the workspace (
dev.kit task cleanup) to prevent context noise.
- Primitive-Only: Agents are auto-authorized to use
dev.kitcommands. Non-standardized commands require user confirmation. - Grounding First: Every session begins with environment hydration (
dev.kit ai sync). - No Shadow Logic: Every action must be discoverable and reproducible via repository source code.
The resolution cycle is operationalized through deterministic UDX engines:
| Requirement | Grounding Resource | Role |
|---|---|---|
| Normalization | udx/dev.kit |
Dynamic discovery and task management. |
| Execution | udx/worker |
Deterministic runtime for logic iteration. |
The Drift Resolution Cycle is built on mathematical and operational principles of delivery flow:
- Predictable Delivery Flow: Managing throughput and cycle time through bounded WIP.
- Observation-Driven Management: Normalizing task assignment and efficiency through AI-identified patterns.
UDX DevSecOps Team