
from cass_memory_system374
A Bun/TypeScript CLI that builds procedural memory for AI coding agents: extracts rules from session history, manages playbooks, and enforces safety via a traum
CM (CASS Memory System) is a command-line tool that converts episodic agent sessions into a reusable procedural playbook. It provides context commands to surface relevant rules, anti-patterns, and historical snippets to assistants before they act. The system tracks confidence decay, validates proposed rules against history, and offers onboarding starters and safety guards.
Use CM when you want agents to retain and reuse practical lessons across sessions and agent types — for example, surfacing troubleshooting rules before a debugging task, or blocking dangerous commands detected by the trauma guard. It’s intended for teams building robust agent workflows and automated reflection pipelines.
Designed for developer-focused assistants and agent integrations that can call local CLIs or MCP servers (Claude Code, Codex/Copilot, Cursor). It’s particularly valuable in environments where multiple agent types operate and cross-agent memory is desired.
CASS Memory System (cm) is a Bun/TypeScript CLI that provides procedural memory for AI coding agents — a three-layer architecture (episodic, working, procedural) with confidence decay and trauma guard safety system. SKILL.md is exceptionally comprehensive with clear docs. Three scripts included: coverage runner, e2e smoke test, and a postinstall patcher. All failed due to missing `bun` runtime, not code issues.
bunThe curl|bash install pattern is the main security concern (-35). Otherwise the skill is well-architected with a thoughtful safety system (Trauma Guard). Scripts are well-written but Bun-dependent. The patch script modifies @xenova/transformers internals for standalone binary compatibility — unusual but well-documented with clear intent.