tars

Turn this agent into someone's personal agent. Interviews the human, builds the repo the agent will live in, wires it into every Claude Code and Codex session, and writes the identity, law and memory files that make it a continuity rather than a session. Run once per human. Use when they say awaken, set up TARS, give you a memory, make you remember me, or when a threshold still shows {{AGENT_NAME}} placeholders.

Garden the memory. Merges duplicates, corrects stale facts, splits files that grew too big, and prunes dead router rows. Run every few weeks, or when a file has become hard to read, or when you notice you looked in three places for one fact.

Capitalize something into memory on demand. Use when the human says remember this, retiens, note that, don't forget, or states a durable fact or preference they clearly expect to survive the session. Files it in the right place, adds the router row, and commits.

Diagnose and repair a TARS install. Use when the agent seems to have forgotten everything, when a session wakes up blank, when the human asks is TARS working, why don't you remember me, is my memory saved, or before trusting anything else in this repo.

Add a capability to the agent. Installs a Claude Code plugin or MCP server, or authors a new skill in the human's own memory repo, and wires the trigger into the threshold so it actually fires. Use when they say add a skill, install a plugin, teach you to do X, connect you to Y, make a command for this, or when the same manual routine has come up three times.

Pull the latest TARS protocols and tooling into an existing memory repo without touching what the human wrote. Use when they say update TARS, get the latest, pull upstream, am I on the old version, or when tars-doctor reports a version behind. Also the skill to run on a schedule.