Browse / Second brain / Software development / agent-memory

agent-memory

isaacbitran

Rules, a reference lint, and an example for agent memory that stays correct for years.

Open isaacbitran/agent-memory → Checked against the four rules.

agent-skillsai-agentsclaude-codeknowledge-managementlinterllmmarkdownmemorypython

Does not say

no agent config and no instruction file at the root.

In its own words

README.mdisaacbitran/agent-memory

Memory for AI agents that stays correct for years: small Markdown files, supersession instead of deletion, and a lint gate on every write.

Then just say "remember that we picked Postgres over Mongo because of joins". The agent files it, lints it, and finds it again next month.

Optional: lint every memory write automatically with a Claude Code hook in .claude/settings.json:

I run it in my own knowledge base and in Edel AI, an analyst a college football staff uses every week.

How it works

Every fact is distilled once, when it arrives, into small Markdown files. Agents answer from the distilled current truth, not from raw transcripts. A lint gate checks every write, and anything ambiguous waits for a human.

Quoting 26% of the README. Read the rest on GitHub ↗

Read from the repository's git tree on 2026-08-17.