Skip to main content

Reliability: Context Quality

1. Context Quality: Give the Agent the Right Reality​

Use this lever when the agent is detached from the actual system. It may invent a cache client instead of using the existing abstraction, follow a generic framework pattern instead of the codebase pattern, or miss a constraint held in tests, docs, or nearby code.

The operator move is selective grounding: load the facts that constrain this task, not every fact available. For a rate-limiting change, useful context includes:

  • the existing middleware pattern
  • how anonymous and authenticated users are identified
  • the current Redis client or cache abstraction
  • the API error shape callers expect
  • the routes, roles, and failure modes that require protection

Grounding finds those facts. Context engineering keeps them usable by controlling placement, size, lifetime, and isolation.

The diagram shows context quality as a filter rather than a larger container: relevant facts enter the work chain while stale or irrelevant history remains outside it.

Reliability levers

Better context raises baseline reliability

Context quality raises baseline Rrelevant context• middleware pattern• Redis abstraction• error contractirrelevant historystale notespatternmiddlewareclientRedislimitAPI shapeless noise → higher R per step
Context quality is selective loading, not maximal loading. Relevant facts feed the task chain; stale or irrelevant history stays outside the working context.

This lever raises the starting quality of each step. It does not fix a task that is too broad, a vague target, or a bad assumption that has already propagated.

Next: Reliability: Orchestration