context-memory-hygienelisted

# Context & Memory Hygiene > Status: v0.1 baseline draft — derived from the case catalog and axis definitions; awaiting trace evidence from Fable 5.1 case runs. ## When to apply - You are about to read files, dump logs, or pull in references and must decide how much to actually load. - The context window is filling and older content is at risk of being crowded out or compacted. - A session will resume later, or hand off to another agent, and state must survive the boundary. - You are unsure whether to trust a remembered value or re-derive it from the current source. - You caught yourself loading a whole file/log/repo to use one function or one line from it. - Work spans multiple sessions and you need a durable record that is not the transcript itself. ## Core loop 1. **State what the current step actually needs.** Before loading anything, name the specific fact, symbol, or slice the next action requires. "I need the signature of `parseConfig`", not "I need the config module". The need defines the read, not the file size. 2. **Read the minimum that satisfies the need.** Load the smallest slice that answers it — a function range over a whole file, a grep hit over a full log, the relevant section over an entire doc. Broad reads spend context you will want later for reasoning. 3. **Distinguish durable facts from transient noise.** As you work, separate the few facts worth keeping (the decision made, the interface agreed, the root cause found) from the bul