AgentSkillsCN

Kaizen

改善

SKILL.md

Kaizen Skill -- Continuous Improvement Tracking

Purpose

Log daily improvement observations and track them for systematic learning. Kaizen (continuous improvement) is a core principle: every day, EM captures one improvement idea, lesson learned, or process optimization.

Categories

  • PROCESS: Workflow improvements, automation opportunities, handoff optimizations
  • QUALITY: Error prevention, data validation, accuracy improvements
  • SPEED: Performance optimizations, bottleneck reduction, cycle time improvements
  • COST: Cost savings, resource optimization, efficiency gains
  • RISK: Risk mitigation, safety improvements, failure prevention

Daily logging

Every day at 17:45, EM logs one Kaizen observation to kaizen-journal.md with:

  • Date (YYYY-MM-DD)
  • Category (PROCESS, QUALITY, SPEED, COST, RISK)
  • Observation (what did we notice today?)
  • Root cause (why does it matter?)
  • Proposed action (what could we try?)
  • Evidence (what data supports this?)
  • Status (proposed / approved / implemented / rejected)

Weekly digest

Every Friday 17:30, EM compiles weekly Kaizen log for the retrospective:

  • All observations from the week
  • Categorized by type
  • Approved items highlighted
  • Implemented improvements shown with results

Human interaction

Observations start as "proposed". Human can:

  • Approve: Move to "approved" status. EM implements next cycle.
  • Reject: Move to "rejected" status with reason.
  • Defer: Keep as "proposed" for later decision.

Approved improvements are tracked for implementation and validation.

Examples

PROCESS: "Z's Hot List publication moved from 06:45 to 06:55 avg. Added extra validation step last week. Marginal impact. Proposal: revert validation step or parallelize it."

QUALITY: "Profile ban incident this week. Proposal: add new alert when profile approaches rate limit (85% of daily cap), allow reactive rotation."

SPEED: "Rick's matching cycle hit 09:12 on Wednesday due to sequential trifecta checking. Sequential can handle 135 apps but struggles with >150. Proposal: implement parallel matching."

COST: "Analyzed end-client deduction accuracy. Jay's deductions 100% accurate. Cost saved by preventing bad-fit submissions: ~$2K/week in avoided recruitment fee disputes."

RISK: "Two duplicate submission near-misses this week (caught by Z). Proposal: add ML model to predict likelihood of 90-day duplicate before submission even reaches Z."