wisci

WISCI gives you four commands — Write, Isolate, Select, Compress — that keep your sessions sharp. Save what matters, load only what's relevant, research without polluting your context, and hand off cleanly between sessions. Version 2 makes the loop close itself: a deterministic staleness engine, per-stream handoffs that never clobber each other, and session hooks that surface your stored context without being asked.

Installation

WISCI skills use the Agent Skills open standard (SKILL.md format), supported across all major AI coding agents. The staleness engine requires git and python3 (both preinstalled on macOS/Linux dev machines).

[!NOTE] /isolate uses subagent spawning, which works best on Claude Code and Codex CLI. Session hooks and dynamic state injection are Claude Code features — on other platforms the skills degrade gracefully to plain instructions.

When to Use What

The Problem

LLM context windows degrade in four predictable ways — each addressed by a specific WISCI command:

• Context Poisoning — Wrong facts persist and compound across a session. /write preserves exact facts with source tracking. /select auto-strips sections whose referenced files have changed.

• Context Distraction — The model fixates on accumulated history instead of using its training. /isolate keeps research noise in separate windows. /compress deliberately compacts to stay in the effective range.

• Context Confusion — Irrelevant information leads to wrong tool selection and document usage. /select loads only relevant, validated context. /isolate gives each agent a clean, focused window.