Temet CLI 0.3.6
Temet CLI 0.3.6: three new commands and a simpler onboarding
Temet CLI 0.3.6 adds temet rules, temet share, decision trace capture, and a simplified agent onboarding. One command to audit, one to export rules, one to publish your operator proof.
What is new in 0.3.6
Three new commands. `temet rules` exports your correction patterns as reusable rules for your agents. `temet share` publishes your operational loops to your public card. And `temet audit` now shows your corrections, emerging rules, and loops alongside the existing skill output. The onboarding is also simpler: one agent-first session flow, a cleaner publish pipeline, and llms.txt updated so agents can read the product without help.
temet audit now captures decision traces
When you correct your AI agent, Temet captures a structured decision trace: what the agent proposed, what you chose instead, on which file, and why. Run `temet audit --track` and your corrections are persisted locally in ~/.temet/traces/. After a few sessions, patterns emerge automatically. No config, no API key, no setup beyond the one command you already run.
temet rules exports what you learned
Once your correction patterns stabilize (3+ occurrences across 2+ sessions), `temet rules` exports them. Three formats: `--format markdown` for documentation, `--format json` for tooling, `--format claude` for direct injection into your CLAUDE.md. The result: your next session starts with your rules already loaded. You stop correcting the same things.
temet share publishes your operator proof
Your public card at /a/{address} now shows operational loops instead of just skill badges. Each loop shows: the problem you identified, the corrections you captured, the rules you stabilized, and the improvement measured. `temet share` publishes after explicit confirmation. Nothing goes public without your approval.
Simpler onboarding for agents and users
The agent onboarding is now a single live session. No manual token copy, no relay configuration. The publish flow is aligned between CLI and web. And llms.txt describes the product in a format any agent can read: what Temet does, how to start an audit, where to find the documentation, and how to read a public profile.
Try it now
Install nothing. Run `npx @temet/cli audit` on any project with Claude Code sessions. After a few sessions with --track, run `temet rules --format claude` to see your first exported rules. If you want to publish your operator proof, run `temet share`. Three commands, zero config, local by default.
Audits local sessions and surfaces repeated skills.
FAQ
Do I need an API key?
No. Audit, rules, and share work locally without any API key. The optional narration feature uses ANTHROPIC_API_KEY if available.
Will this break my existing audits?
No. New fields default to zero when missing. Existing snapshots and tracking continue to work.
How many sessions before useful rules appear?
A rule needs 3+ corrections across 2+ sessions. In practice, 5-10 normal sessions produce 3-5 actionable rules.
Next step
Start with npx @temet/cli audit, then add tracking, the macOS companion, and background hooks only after the base result already feels useful.
Install 0.3.2Published March 16, 2026