AI Skill StudioAI Skill Studio
Skill Library
Presets:

MEMORY.md generator: decisions and lessons that outlast a session

Record architecture decisions, known gotchas and key paths so every new agent session starts from them. Runs in your browser; nothing is uploaded.

Target Standard & File Format

AI Agent Rules & Skills
Context Shields & Privacy (Ignore Files)
Integrations & Machine Documentation
AI Governance & Multi-Agent Planning Suite

1. Persistent Technology Context Matrix

Application framework, language invariants, persistence layer, and engineering philosophy:

2. Architectural Decision Records (ADRs)

Formal architectural decisions across context resets (Status, Context, Decision, Consequences):

3. Operational Gotchas & Pitfalls

Critical traps, edge cases, and hard-earned learnings that all agents must heed:

4. 5-Step Agent Execution Loop Protocol

Standard agent execution loop (Ingest → Plan → Execute → Verify → Update):

5. Domain Invariants & Key File Anchors

System boundaries, core package directories, and project invariant constraints:

6. Session History

Chronological record of memory additions and changes over time:

MEMORY.md: how it works

Agents start each session without memory of the last one. MEMORY.md is a Markdown file where decisions (as numbered architecture decision records), known gotchas and important paths are written down, so the next session starts from them instead of rediscovering them. It is a convention: tell the agent in CLAUDE.md or AGENTS.md to read it at the start and add lessons at the end.

Where the file goes

  • ./MEMORY.mdRepo rootRead at session start; agents append lessons and decisions, when your rules file says so.

What goes in it

Architecture decision recordA numbered decision with its reason, so it is not reopened by the next session.
### ADR-001: Process files in the browser
- Decision: All parsing runs client-side; no upload endpoints.
- Reason: Users paste secrets and private code.
- Consequence: Large files are limited by browser memory.
GotchasTraps an agent would otherwise hit again.
## Gotchas
- Monaco must be imported with next/dynamic and ssr: false.
- `npm run build` uses webpack, not Turbopack; do not change it.

Tips

  • Number decisions (ADR-001, ADR-002) and never rewrite old ones; add a new one that supersedes it.
  • Write down gotchas the moment an agent hits them.
  • Keep entries short and factual so the file stays cheap to load.
  • Remove entries that are no longer true.

Not sure MEMORY.md is the file you need? Each AI coding tool reads its own file from its own place.

Which file do I need?

Frequently asked questions

How does MEMORY.md help AI agents?

It gives each new session the decisions and lessons from earlier ones: what was decided and why, and which traps to avoid. Without it, agents repeat past mistakes or reopen settled questions.

Do agents read MEMORY.md automatically?

No. Add 'Read MEMORY.md at the start of each session and record new lessons in it' to your CLAUDE.md or AGENTS.md.