kaizen

Plugin kaizen from marketplace kaizen-plugin in shm11C3/kaizen-plugin. Commands read from the repo's manifest — 13 days ago.

Skills (15)

• experiment

  Run one experiment end to end on the open hypothesis: change the code on its branch, commit, deploy, benchmark, evaluate with the fixed rule, record the decision, and leave the servers on stable or on the promoted candidate. Use after hypothesize, or to finish an experiment that was interrupted.

• hypothesize

  Turn an observation into one testable hypothesis and open an experiment for it. Use after observe, or when the human proposes a change: writes the Candidate Manifest (hypothesis, cited evidence, allowed change, expected effect) that deploy.sh requires before any candidate reaches a server.

• improve-workflow

  Improve recurring ISUCON operations. Refine reusable skills and supporting tools during practice and competition; preserve practice problem/environment-specific procedures as knowledge. During competition also build target-specific skills, test them, integrate into the private plugin main and roll o

• knowledge

  Use the measured pattern corpus. Before a hypothesis, match an observation dump against the pattern signatures and read the matching pattern files; after an experiment, confirm it was ingested and improve the pattern file (checklist, signature) from what happened. Also the place to add a new pattern

• observe

  Collect the evidence a hypothesis must cite. Use before every hypothesis and after every promotion: resets the counters, runs or records one benchmark, dumps MySQL digests, request mix, and indexes into a named observation file, and summarizes the top three costs.

• plan

  Choose experiments by current scoring impact, evidence, uncertainty and total time. Use after setup, after each decision, when predictions fail or strategy signals fire. Maintains .kaizen/plan.md and reconsiders mechanisms before repeating small tweaks.

• prepare-candidate

  Prepare a claimed ISUCON team candidate from an evidence-backed hypothesis through implementation, related validation, commit and READY artifacts. Use for local worker preparation while the integration owner measures another candidate; does not grant target reservations or adoption authority.

• prepare-tests

  Select or add focused, specification-backed checks for a concrete ISUCON change or observed quality failure. Use when existing checks cannot answer that risk or a held candidate needs a reproducer; broad preparation is optional rather than a gate before every benchmark.

• profile

  Explain a measured ISUCON bottleneck with existing profilers, access/SQL analysis or bounded OpenTelemetry HTTP-to-DB tracing. Use when an observation needs causal detail before choosing an optimization; produce evidence and a testable hypothesis.

• recover

  Put every server back on the stable revision and prove it. Use when a candidate failed, the human says stop, a deploy was interrupted, or status shows deployed != stable with no open experiment.

• refactor

  Create a bounded ISUCON preparation phase when concrete code coupling repeatedly obstructs an optimization or quality repair. Preserve behavior with existing contract tests, separate the structural change from the optimization, and measure its downstream payoff.

• retro

  Capture ISUCON failures and recurring friction. Refine reusable skills during practice; preserve practice problem-specific procedures as knowledge. During competition also turn target-specific recurring work into tested skills through improve-workflow and private plugin-main integration.

• setup

  Prepare a repository and its servers for the kaizen loop. Use at the start of a contest or practice session, or whenever .kaizen/config.sh is missing, hosts changed, or the stable revision must be re-pinned. Writes .kaizen/config.sh from a preset or the day's manual, pins STABLE_REV, deploys it, run

• status

  Show where the improvement loop stands: stable and deployed revisions, open experiment and its samples, baseline, budget, and the log of closed experiments. Use when resuming, before deciding what to do next, or when the human asks how it is going.

• team

  Join or resume an autonomous ISUCON team after cloning the competition repository. Use for Claude or Codex team participation, task discovery and atomic claims, candidate submission, serialized integration and benchmarks, human decisions, or control-host operations. Keeps all shared operating rules

Hooks (3)

• SessionStart

  python3 "${CLAUDE_PLUGIN_ROOT}/hooks/guard.py"

• PreToolUse

  python3 "${CLAUDE_PLUGIN_ROOT}/hooks/guard.py"

• Stop

  python3 "${CLAUDE_PLUGIN_ROOT}/hooks/guard.py"

Similar plugins