Context Saver

What it does

Context Saver acts as an execution middleware that prevents large API responses and raw logs from flooding an agent's context window. Instead of direct output, it runs commands in sandboxed subprocesses, indexes the full output in an FTS5 SQLite database, and returns only a compact, intent-filtered summary to the agent.

When to use it

• When executing data-heavy skill commands (APIs, analytics, large searches).
• In multi-step pipelines where raw output from one step would exceed token limits for the next.
• During session resumption after context compaction to restore state via snapshots.
• When the agent's context window is running low and token efficiency is critical.

What's included

• Scripts: ctx_run.py (summarized execution), ctx_batch.py (multi-skill pipelines), ctx_session.py (snapshotting and event logging), ctx_search.py (querying indexed data), and ctx_stats.py (efficiency metrics).
• Instructions: Procedural guidance on implementing intent-based filtering and managing session continuity through snapshots.

Compatible agents

Designed for agents with access to a Python runtime and local filesystem, compatible with Claude Code, Codex, and other MCP-capable environments.