Coders In Hoods

Courses
Answers
Setup
Bookmarks

Coders In Hoods Academy

Every course is free. Sign in to start, and your progress, quiz scores and bookmarks follow you.

Academy

Courses
Paths
Exercise answers
Setup guides
Bookmarks

Coders In Hoods

The agency
Coaching for teams

Social

LinkedIn
X (Twitter)

Contact

codersinhoods@gmail.com

© Coders In Hoods · since 2019

Free · no card

Coders In Hoods Academy

Sign in to start learning

Free, no card. We use your account to save progress, quiz scores, bookmarks and occasionally send updates from our team.

Course overview

Agent Memory

Free · sign in to start and track progress

1 · How agents remember

1Memory types for agents
2The context window as working memory
Exercise answers
- Answer: Measure fact drift across repeated compaction
- Answer: Re-inject matter state after compaction in Claude Code
3Files, skills and memory tools
4Memory frameworks and benchmarks

2 · Memory in production

5Scoping, retention and erasure
6Write, update and forget
7Shared state and memory poisoning
8Evaluating memory

CapstoneA memory layer for a 12-week case Reference implementation

Answers Lesson 2: The context window as working memory

Step-by-step answer

Stretch 30–45 min6 steps

Re-inject matter state after compaction in Claude Code

A SessionStart hook registered for the compact source that re-reads a synthetic matter.json after every compaction and adds it to Claude Code's context as a short, authoritative Markdown block. The facts live in a file the hook reads, never in the conversation, so a summary can't paraphrase them.

Back to the exercise

Before you start

Before you start:Install Claude Code Python with uv

The hook is one standard-library Python script. It reads matter.json, renders it as about 1,000 characters of Markdown, and prints it as additionalContext. Claude Code runs it again every time the session compacts. The decision that matters most is the split between the two files Claude sees after /compact. CLAUDE.md holds the rule ("matter facts come from the MATTER STATE block, and the block wins"). The hook holds the state, which changes during the matter and so can't be pasted into an instruction file.

Previous answerMeasure fact drift across repeated compaction Next answer Build and attack a per-matter memory-tool handler

On this page

Before you start
The exercise
1. Read the hook contract
2. Write the matter record
3. Write the hook script
4. Register it for compaction
5. Run it
6. Write up the deliverable
Common mistakes
All files

The hook is one standard-library Python script. It reads matter.json, renders it as about 1,000 characters of Markdown, and prints it as additionalContext. Claude Code runs it again every time the session compacts. The decision that matters most is the split between the two files Claude sees after /compact. CLAUDE.md holds the rule ("matter facts come from the MATTER STATE block, and the block wins"). The hook holds the state, which changes during the matter and so can't be pasted into an instruction file.

Free · sign in to start

Sign in to keep reading the answer.

Free, no card. We use your account to save progress, quiz scores, bookmarks and occasionally send updates from our team.

6 more sections
Progress, quiz scores and bookmarks