Use Disk As Context Sink
use-disk-as-context-sink
Summary
Large pasted context is difficult to search, diff, summarize, and revisit. Put bulky or reusable context on disk so agents can pull the relevant slice while the thread stays focused on decisions and handoff.
Problem
Large pasted context is hard to search, summarize, diff, and revisit. It also competes with the current task for attention when most of the material is only occasionally relevant.
Preferred Move
Put large or reusable context on disk in structured files. Let agents use search, parsers, scripts, and summaries to pull the slice they need. Keep the thread focused on decisions and handoff rather than raw bulk context.
Tradeoff
Small, ephemeral context can stay in the prompt when writing a file would add ceremony. Move context to disk when it is large, reusable, structured, or likely to be consulted by more than one session.
Agent Instruction
When context is large or reusable, ask for or create a disk-backed source of truth. Use tools to
query it instead of trying to keep all details in the active conversation.Examples
Bad: the thread carries a giant table that the agent must reason over from memory.
Here are 5,000 rows of dependency metadata pasted into chat...
Good: the data is on disk and queried as needed.
dependencies.csv is checked in under references/. Use DuckDB or a small script to inspect the rows
needed for this decision.