from antigravity-skills824
Use the filesystem as an overflow and persistence layer for agent context — store large tool outputs, plans, logs, and sub-agent artifacts to keep active prompt
This skill teaches agents how to use the filesystem as a durable, searchable overflow layer for context. Instead of stuffing large outputs or long-running plans into the prompt, agents write them to files, return compact summaries or references, and selectively read back only the required sections. The skill covers scratch pads, plan persistence, sub-agent handoffs, terminal/log capture, and dynamic skill loading.
Activate when tool outputs exceed token budgets (e.g., >2k tokens), tasks span multiple turns, multiple agents must share state, or logs/terminals need selective retrieval. Avoid for single-turn interactions or when latency is critical.
Best suited for agents that support filesystem tools (Cursor, Claude Code, Copilot-like integrations). Works with any agent that can read/write files and run shell-style searches.
This skill has not been reviewed by our automated audit pipeline yet.
