Claude Cowork Agentops — Communication skill for Claude Code

claude-cowork-agentops

A [Claude Cowork](https://www.anthropic.com/) automation kit that runs a daily **housekeeping** pass over [claude-mem](https://github.com/thedotmack/claude-mem) and distills it into **conventions (CLAUDE.md), reusable logic (skills), and long-term memory (Obsidian)** — with AgentOps (monitoring, alerting, evaluation, governance) built in.

claude-mem is a recency-window local memory: it carries recent context into the next session, and older knowledge gets pushed out and lost. This kit is the daily batch that **rescues what should persist into the long-term layer before it ages out**.

Set up entirely from Cowork chat: paste [`setup.md`](setup.md) and Claude will ask you a few questions (output language, project folder, vault, which tasks) and configure everything for you.


The daily monitoring dashboard in Obsidian (Dataview + Charts): per-machine activity/token table and trend lines, generated automatically each night.

Memory hierarchy

Layer	Role	Location
CLAUDE.md	Conventions / structure (per project)	each project (proposed uncommitted in git)
skill	Reusable logic	project-specific → /.claude/skills/ (auto, uncommitted); cross-project → ~/.claude/skills/ (proposal)
claude-mem	Short/mid-term memory (local, transient)	~/.claude-mem
Obsidian vault	Long-term memory (shared across devices)	local path (synced via Obsidian Sync, etc.)

Flow

Claude Code work -> claude-mem (local, short/mid-term)
        | daily batch (Cowork scheduled task / 00:00)
        |-> CLAUDE.md   : conventions/structure (proposed uncommitted in the most relevant dir)
        |-> skills      : project-specific -> /.claude/skills (auto); cross-project -> ~/.claude (proposal)
        \-> Obsidian vault (long-term)
              |- reports/YYYY-MM/DD/