Skip to content

README.md

Latest commit

 

History

History
57 lines (42 loc) · 2.6 KB

README.md

File metadata and controls

57 lines (42 loc) · 2.6 KB

Kaizen hands-on tutorial

Learn Kaizen by using it on a repo where you already run a coding agent (Cursor, Claude Code, Codex, or another tailed agent). You will wire capture, read the local store, interpret metrics, run a retro, and see how experiments and optional sync fit the same loop.

Why this exists

Kaizen is a feedback loop, not a dashboard:

flowchart LR
  capture[Capture]
  summarise[Summarise]
  change[ProposeChange]
  measure[Measure]

  capture --> summarise
  summarise --> change
  change --> measure
  measure --> capture
Loading

Transcripts and hooks capture; summary, insights, guidance, and metrics summarise; retro, eval, and exp propose and measure change. The deep data story is in telemetry-journey.md.

After you have sessions in the store you can run kaizen eval run to call an LLM judge (requires [eval].enabled = true and ANTHROPIC_API_KEY). Low-scoring sessions appear as the H15 bet in kaizen retro. See evaluation usage and config.md#eval for setup.

Optional: enable [collect.outcomes] and [collect.system_sampler] in config to attach post-stop test rows and per-PID CPU/RSS samples (local-only); see outcomes.md and system-telemetry.md.

What you will know after

  • How init wires hooks and where data lands on disk.
  • When to use cache-first reads vs --refresh.
  • How --source local|provider|mixed changes observe-style commands when sync identity and [telemetry.query] are configured (optional).
  • How --all-workspaces aggregates across repos you have opened.
  • Which workflows are MCP vs shell-only.
  • Where to read next for proxy, sync, and experiments detail.

Paths through this tutorial

Time Do this
~5 min Part 1 → Part 2 first sections
~30 min Parts 1–4
Deep dive All parts 1–9 + linked reference docs

Parts

  1. Setup: install, init, doctor
  2. Observe: Web, sessions, summary, TUI
  3. Interpret: insights and guidance
  4. Repo intelligence: metrics and refresh
  5. Improve: retro
  6. Improve: experiments
  7. Optional: proxy, sync, telemetry
  8. Agents calling Kaizen: MCP
  9. Housekeeping: gc and completions

Full command flags: usage.md. Glossary: concepts.md.