Mistake L2 · Context engineering informational

One Project Per Outcome, Not One Giant Catch-All

What it is

An organising rule: create a separate Project per distinct outcome or client, rather than dumping everything into one catch-all Project whose mixed context dilutes every answer.

One mega-Projectmuddied context· Mixed files & topics· Rules that half-apply· Answers pulled off-targetOne per outcomeclean context→ Only relevant files→ Rules that all apply→ Sharper, on-topic answers
Focused Projects keep context clean; a catch-all muddies every answer.

Why it works

A Project's power is its shared context — but that only helps if the context is coherent. Pile unrelated work into one Project and the files, instructions, and history pull answers in conflicting directions. Splitting by outcome keeps each Project's context tight, so the model draws only on what's relevant to the task in front of it.

When to use it

As soon as you notice a Project spanning genuinely different bodies of work, or answers in it getting vaguer as unrelated material accumulates.

When not to use it

Closely related workstreams that share the same files and rules — over-splitting those just fragments context that belonged together.

Prompt

I currently keep <describe what's in the Project>. Should this be one Project or several? Split by distinct outcome or client, and tell me which files and instructions go with each.

Example

Breaking a bloated 'Work' Project into 'Client A', 'Client B', and 'Internal Docs' makes each thread pull only its own guidelines instead of a blur of all three.

Common mistakes

  • Running every project through one 'everything' Project.
  • Splitting so finely that related work loses its shared context.
  • Leaving stale files from an old effort in an active Project.

Related