xz1220/kempt/kempt-checkup

Kempt Checkup — diagnose agent debt (read-only)

You are running a read-only diagnosis of the user's local agent state with the Kempt CLI. Nothing is modified, nothing leaves the machine.

Steps

1. Locate the CLI. Run command -v kempt. If missing, try node <kempt-repo>/bin/kempt.mjs

(ask the user where the Kempt repo is, or suggest npm install && npm link inside it). Use whichever works as $KEMPT below.

1. Scan. Run:

``bash kempt scan --json ``

This performs the read-only scan and prints the full report as JSON.

1. Interpret the report for the user. From the JSON, extract:

• healthScore (0–100) and verdict
• the five dimensions (memory / tools / conflict / runtime / reclaimable),

each with severity (crit/warn/good), metricValue and summary

• summary.totalFindings and summary.reclaimableBytes
• worstDimension — lead with this one.

Present a short, plain-language readout in the user's language: what is hurting the agent most, why it matters (stale memory gets re-read every session; unused skills eat context budget; conflicting instructions make behavior random), and the 2–3 highest-impact items from findings (use title + reason; findings are already path-redacted).

1. Offer the next step, don't take it. If anything is cleanable, tell the user a reversible

cleanup is available via the kempt-clean skill or kempt clean (dry-run by default). Do not clean anything in this skill.

Rules

• This skill is strictly read-only: only kempt scan, kempt report, kempt plan and

kempt history may be run. Never run kempt clean --yes or kempt empty-recycle here.

• Never paste raw secrets or file contents into the conversation; the report is already

redacted — stick to what it provides.

• If the scan fails, show the error and suggest kempt help; do not retry more than once.
• Health score bands for your summary: ≥80 tidy ("kempt"), 60–79 needs a tidy-up,

40–59 noticeable debt, <40 the agent is being actively degraded.