SuperAPI AI Directory

reflect-and-remember

Agents & models

View repository

Updated 10-15 16:44

About this skill

Reflective memory protocol

Reflection checklist (quickly go through it after completing a task)

  • Did this time involvepitfalls/detours? What was the root cause?
  • Was therecounterintuitive behavior(unexpected behavior of frameworks, APIs, or tools)?
  • Was therean architectural decisionora project conventionworth recording?
  • This informationwill it be used again next time? Will other developers benefit?

All no → skip, do not remember.

Storage location

Category Location Commit to git Content
Team shared Project .claude/memory/MEMORY.md Yes Deployment lessons, business rules, architectural decisions, environment key points, pitfall records
Sensitive private ~/.claude/projects/<project>/memory/MEMORY.md No Passwords, keys, tokens, personal paths

Writing process

  1. Read the target file, check whether relevant memories already exist
  2. Existing → Edit and update; none → append a new entry; contradiction → correct the old memory
  3. Group by semantics (## headings), not by time
  4. Keep the total number of lines within 150; when exceeded, merge and condense

Writing principles

  • Conclusion-oriented: write "Docker --env-file does not remove quotes", not a narrative of the process
  • Actionable: write "use sed for precise replacement, do not overwrite the entire .env"
  • Has context: write "Prisma P1012 → check whether DATABASE_URL has quotes"
  • No duplication: do not record duplicates of what is already in CLAUDE.md / skill
  • Sensitive separation: passwords and keys only go into private memory; in shared memory use "configured" instead of specific values