Agent Skills
Production-grade engineering skills for AI coding agents
Stop agents repeating expensive calls and re-reading the same files
An MCP server that blocks redundant expensive tool calls and remembers what an agent already worked out, cutting token usage on tool output by a wide margin versus plain agent runs. It targets heavy users of Claude Code, Codex, Copilot, Gemini and a dozen other agent clients who want cross-session memory and quota pacing.
Installed as an MCP server, it interposes on tool calls: if the agent reaches for a file or resource it has already read, the server refuses and points it back at the cached result. It maintains its own memory across sessions and paces subscription quota usage. Note the caution from the video: installing it disables Claude Code's trust prompt, so you should understand the security implications before enabling it in a sensitive environment.
10 Repos conserve your token usage
Production-grade engineering skills for AI coding agents
Compress tool outputs, logs and RAG chunks before the LLM sees them
Self-improvement memory that helps coding agents learn from sessions
Persistent memory that carries agent context between sessions
Multi-phase security audits with verified, machine-readable findings
Persistent memory so coding agents stop forgetting context