agentsop-context-scope-discipline

What problem does it solve?

This Skill prevents coding-agent context dilution by enforcing disciplined separation between “read” context and the smaller “edit working set,” so the model stops editing wrong files and stays focused during multi-file changes.

Core Features & Use Cases

• Working-file budget enforcement: Keeps the editable working set (files you add for writing) under about ~25k tokens, using token-budget monitoring to reduce mid-task quality collapse.
• Read vs edit separation: Delegates breadth to a read-only repo-map (signatures) while reserving full-text, write-permitted files for only what you will actually change this turn.
• Operational SOP for multi-file edits: Uses classify → locate-then-add → add/read → watch → drop → split-task patterns, including concrete decision rules like “understand 10, edit 2.”

Quick Start

Ask the coding agent to follow the context-scope discipline rule: only load into the editable working set the few files it will actually edit, keep that working set under about 25k tokens, and treat all other files as read-only via the repo-map or explicit reads.