scroll

Your codebase already knows. Scroll extracts it.

What It Does

Scroll reads git history, GitHub PRs, and issues, then uses LLM-powered extraction to produce structured knowledge entries: decisions, learnings, mistakes, observations, goals. These feed into engram and agent instruction files (CLAUDE.md).

# Initialize in your repo
scroll init

# Extract knowledge from last 20 commits
scroll ingest -n 20 -p my-project

# Deposit into engram knowledge base
scroll deposit -p my-project

# Sync into CLAUDE.md for agent context
scroll sync -t claude

The Pipeline

git log / GitHub PRs → LLM extraction → structured entries → engram deposit
                                                          → CLAUDE.md sync
                                                          → search/query

Why Not Just Read Git Log?

Commit messages say WHAT changed. Scroll extracts WHY — the decisions, tradeoffs, and learnings buried in diffs and PR discussions.

From a commit diff:

-name = "caliber"
+name = "caliber-trust"

Scroll extracts:

DEC-001: Changed PyPI distribution name from 'caliber' to 'caliber-trust'. The original name was already taken by an existing ML library.

Commands

Command	What it does
scroll init	Initialize scroll in a repo
scroll ingest	Extract knowledge from git/GitHub
scroll deposit	Push entries into engram
scroll sync	Inject knowledge into CLAUDE.md
scroll list	List extracted entries
scroll search	Search entries by text
scroll stats	Show summary statistics
scroll health	Check knowledge base health

Requirements

• Python 3.10+
• An Anthropic API key (for LLM extraction)
• Git repository

License

MIT