Skip to main content

context-awareness

AI agents often fall into the trap of "Vibe Coding"—writing code that looks correct but ignores the existing patterns, standards, and lexicon of your specific project. This is a form of Context Drift.

Rigour's Universal Context Awareness Engine prevents this by dynamically mining your codebase to build a "Golden Record" of established patterns.

The Problem: "Guess and Hope" (Vibe Coding)​

The most common frustration with AI agents is the narrative vs. reality gap. An agent will confidently state: "I have fixed the type errors and reformatted the code. You can push now."

But in reality:

  • MyPy/Lint is still fails with 13 errors.
  • Context Drift: The agent used a variable name that doesn't exist or doesn't follow project standards.
  • CI Failures: You lose 10 minutes waiting for a CI pipeline that was doomed from the start.

Rigour acts as the firewall between the agent's creative guesses and your project's technical reality.

  1. Discovery: Rigour scans your environment files, CI configs, and existing source code.
  2. Anchoring: It identifies BASE_URL as a project-standard "Anchor".
  3. Verification: When the agent introduces a variation like BASE_URL_PRODUCTION or PROD_BASE_URL, Rigour flags it as Context Drift.

How it Works​

Rigour doesn't rely on hardcoded rules. Instead, it uses Statistical Normalcy:

  • Anchor Discovery: Scans for variables, class naming patterns, and service structures that appear frequently or are defined in "Truth Sources" (like .env or Dockerfile).
  • Drift Detection: Flags new code that diverges significantly from discovered anchors.
  • Pattern Collision: Warns when new names are "too similar" to existing ones but incorrectly formatted.

Configuration​

Context awareness is enabled by default. You can tune it in rigour.yml:

gates:
  context:
    enabled: true
    sensitivity: 0.8  # Threshold for drift detection
    mining_depth: 100 # Number of files to sample for patterns

Why it Matters​

Dynamic context awareness turns Rigour from a simple "linter on steroids" into a Strategic Quality Gate. It ensures that as your codebase grows, your AI agents stay anchored to the reality of your project, not the "vibe" of their training data.