ADR 26007: Formalization of Phase 0: Intent Synthesis (Requirements Engineering)

Title¶

Standardizing a “Human-Lead + Reasoning LLM” Requirements Engineering phase (Phase 0) to eliminate architectural drift in the aidx pipeline.

Status¶

Proposed

Date¶

2026-01-16

Context¶

The current aidx industrial framework (ADR 26005) initiates with Phase 1: Research. However, this assumes that the initial problem statement is sufficiently specified and free of business-logic ambiguities. In practice, starting the automated “Research-Apply” pipeline with an underspecified prompt leads to:

• Token Waste: Local 14B editors and cloud architects re-rolling logic due to unclear goals.

• Cognitive Bottlenecks: Local models failing to resolve multi-agent role definitions or complex business constraints.

• Requirements Volatility: Significant changes to artifacts/plan.md during implementation because the initial intent was not synthesized.

Decision¶

We will implement a mandatory Phase 0: Intent Synthesis (Requirements Engineering) gateway before any aidx session.

1. Lead-Driven Discovery: A Human Lead utilizes a high-reasoning web-based model (e.g., Gemini Thinking/Pro) to stress-test the problem statement, clarify business goals, and define the high-level system architecture.

2. Artifact Generation: The output of Phase 0 must be a formal artifacts/MISSION_ID.md (or artifacts/agents.md) file.

3. The Mission Contract: This artifact serves as the “source of truth” for Phase 2 (Architect), ensuring that technical planning is grounded in a verified business intent.

4. Traceability: All subsequent code changes must be traceable back to the specific goals defined in the Phase 0 Mission artifact.

Consequences¶

Positive¶

• Zero-Drift Execution: High-fidelity intent ensures the aidx pipeline operates on a “Contract of Intent”.

• Reduced Cognitive Load: Offloads abstract ambiguity resolution to high-reasoning models before local hardware is engaged.

• ISO 29148 Compliance: Establishes a formal Requirements Elicitation process as per international standards.

Negative¶

• Increased Human Involvement: Requires the Lead to participate in an initial discovery chat before automation starts.

• Manual Handoff: The Mission artifact must be manually or semi-automatically moved from the web chat environment to the local repository. Mitigation: Use standardized MyST-Markdown templates to ensure copy-paste compatibility.

Alternatives¶

• Automated Intent Prediction: Rejected. Human-in-the-Loop is mandatory on the Requirements Engineering step.

References¶

• ADR 26005: Formalization of Aider as Primary Agentic Orchestrator

• ISO 29148: Requirements Engineering

Participants¶

1. Vadim Rudakov

2. Senior AI Systems Architect