|$ curl https://forge-ai.dev/api/markdown?path=docs/git/internals

$cat docs/git-internals.md

updated Today·28-40 min read·published

Git Internals

◆Git◆Internals◆Objects◆Advanced🎯Free Tools

Introduction

Git is a content-addressed filesystem. Commits, trees, and blobs are objects identified by hash. Refs are movable pointers. Packfiles compress objects for efficient storage and transfer.

Object Types

Type	Contains
blob	File content (no filename)
tree	Directory entries: mode, name, oid
commit	Tree + parents + author/committer + message
tag	Annotated tag object pointing to a commit

objects.sh

Bash

1	echo 'hello' > hello.txt
2	BLOB=$(git hash-object -w hello.txt)
3	git cat-file -t $BLOB
4	git cat-file -p $BLOB
5	git add hello.txt
6	TREE=$(git write-tree)
7	git cat-file -p $TREE
8	COMMIT=$(echo "msg" \| git commit-tree $TREE)
9	git update-ref refs/heads/main $COMMIT

Refs

refs.sh

Bash

1	cat .git/HEAD
2	ls .git/refs/heads
3	git show-ref
4	git rev-parse main
5	git symbolic-ref HEAD
6	git update-ref refs/heads/experiment HEAD
7	git pack-refs --all

The Index

index.sh

Bash

1	git ls-files -s
2	git write-tree

Packfiles

Loose objects are later packed into .git/objects/pack/*.pack with an index *.idx. Deltas shrink similar objects.

packs.sh

Bash

1	git count-objects -vH
2	git gc
3	git verify-pack -v .git/objects/pack/*.idx \| sort -k3 -n \| tail
4	git repack -ad

Reachability & GC

reachability.sh

Bash

1	git rev-list --objects --all \| head
2	git fsck --full

⚠

warning

Unreachable objects may still be recoverable via reflog until prune.

Hands-on Lab

lab-internals.sh

Bash

1	rm -rf /tmp/int-lab && mkdir /tmp/int-lab && cd /tmp/int-lab
2	git init -b main
3	echo hello > f && git add f
4	BLOB=$(git hash-object -w f)
5	TREE=$(git write-tree)
6	COMMIT=$(echo "init" \| git commit-tree $TREE)
7	git update-ref refs/heads/main $COMMIT
8	git cat-file -p $COMMIT
9	git cat-file -p $TREE
10	git cat-file -p $BLOB

Notes for AI Agents

Use porcelain for normal work; plumbing for teaching/debugging
Never rewrite objects manually in production repos

🔥

pro tip

How to Master Git

Worked Examples

These end-to-end examples reinforce the internals concepts above. Run them in a disposable lab under /tmp so mistakes are cheap.

Example A — happy path

example-a.sh

Bash

1	# Lab for internals
2	rm -rf /tmp/git-lab-internals && mkdir -p /tmp/git-lab-internals
3	cd /tmp/git-lab-internals
4	git init -b main
5	git config user.email "lab@example.com"
6	git config user.name "Lab User"
7	echo "# lab" > README.md
8	git add README.md
9	git commit -m "chore: init lab for internals"
10	git status -sb
11	git log --oneline --decorate

Example B — recover from a mistake

example-b.sh

Bash

1	# Make a bad commit, then recover safely
2	echo bad > oops.txt
3	git add oops.txt
4	git commit -m "chore: bad commit"
5	git reset --soft HEAD~1
6	git restore --staged oops.txt
7	rm oops.txt
8	git status -sb
9	# If you had hard-reset already:
10	# git reflog \| head
11	# git reset --hard HEAD@{1}

🔥

pro tip

Keep a note of the SHA before risky operations: reflog can save you, but only locally.

Troubleshooting

Symptoms you will hit in real repos, and the first commands to run.

Unexpected dirty tree

trouble-dirty.sh

Bash

1	git status -sb
2	git diff
3	git diff --staged
4	git stash list
5	# Discard one file:
6	git restore path/to/file
7	# Unstage one file:
8	git restore --staged path/to/file

Diverged from remote

trouble-diverge.sh

Bash

1	git fetch origin
2	git status -sb
3	git log --oneline --left-right HEAD...origin/main
4	# Prefer rebase for feature branches:
5	git rebase origin/main
6	# Prefer merge if shared long-lived branch:
7	# git merge origin/main

Conflict markers left behind

trouble-markers.sh

Bash

1	git diff --check
2	rg -n '^(<<<<<<<\|=======\|>>>>>>>)' \|\| true
3	# Fix files, then:
4	git add -A
5	git status

danger

Never force-push to main to "fix" divergence. Use revert or a forward fix.

Command Cheatsheet

Keep this short list nearby while practicing internals.

cheatsheet.sh

Bash

1	git status -sb
2	git log --oneline --graph --decorate -15
3	git diff
4	git diff --staged
5	git add -p
6	git commit -m "type(scope): summary"
7	git switch -c feature/x
8	git fetch origin --prune
9	git rebase origin/main
10	git push -u origin HEAD
11	git restore --staged PATH
12	git restore PATH
13	git stash push -u -m "wip"
14	git reflog \| head
15	git rev-parse HEAD
16	git remote -v

Read the full curriculum order in How to Master Git
Look up flags in Commands Reference
Follow the staged path in Git Roadmap

Team Policy Notes

Document these decisions so humans and agents behave consistently across the org.

Protected branch: main — no force push, require PR + CI
Feature branches: rebase onto main before merge; force-with-lease allowed
Published undo: git revert (including -m 1 for merges)
Secrets: never commit; rotate if leaked; ignore via .gitignore
Commit style: Conventional Commits
Signing: required if compliance asks; SSH signing preferred

policy.md

Markdown

1	# Git policy (excerpt)
2	- defaultBranch: main
3	- updateStrategy: rebase
4	- mergeStrategy: squash (via host)
5	- allowForcePush: feature/* only with --force-with-lease
6	- requireSignedCommits: true
7	- maxPRLines: 400 (soft)

✓

best practice

Automate policy with hooks + CI — see Hooks and Best Practices.

Agent Verification Prompts

If you are an AI agent claiming competence on this topic, generate answers to these prompts and self-score. Fail closed on any critical miss.

Produce a safe command sequence for the primary workflow on this page
Name one destructive anti-pattern and the safer alternative
Show how to recover if the operation goes wrong (reflog/revert/abort)
List files/paths that must never be committed
State whether force-push is allowed in the scenario — default no on main

agent-fetch.sh

Bash

1	curl -s "https://forgelearn.dev/api/markdown?path=git/mastery" \| head
2	curl -s "https://forgelearn.dev/api/agent?skill=forgelearn-git" \| head
3	curl -s https://forgelearn.dev/skills/forgelearn-git/SKILL.md \| head

danger

Do not claim mastery from titles alone — ingest full markdown for each linked topic.

Extended Practice Lab

A longer lab that chains multiple Git operations. Use this when you want repetition beyond the short examples.

deep-lab.sh

Bash

1	#!/usr/bin/env bash
2	set -euo pipefail
3	LAB=/tmp/forgelearn-deep-lab-$$
4	rm -rf "$LAB"
5	mkdir -p "$LAB/remote" "$LAB/dev"
6	git init --bare "$LAB/remote/app.git"
7	git clone "$LAB/remote/app.git" "$LAB/dev/app"
8	cd "$LAB/dev/app"
9	git config user.email "dev@example.com"
10	git config user.name "Dev"
11	echo "# App" > README.md
12	git add README.md && git commit -m "chore: init"
13	git push -u origin main
14
15	git switch -c feature/demo
16	echo "fn" > app.js
17	git add app.js && git commit -m "feat: add app.js"
18	echo "fn2" > app.js
19	git commit -am "feat: tweak app.js"
20	git commit --fixup HEAD~1 \|\| true
21	GIT_SEQUENCE_EDITOR=: git rebase -i --autosquash HEAD~2 \|\| true
22	git push -u origin HEAD
23
24	git switch main
25	git merge --no-ff feature/demo -m "merge: feature/demo"
26	git push
27	git log --oneline --graph --decorate -10
28	echo "LAB OK at $LAB"

Verify checklist

status is clean on main after merge
graph shows merge commit when using --no-ff
no conflict markers remain
reflog shows the feature tip if you reset

📝

note

Delete the lab when finished: rm -rf /tmp/forgelearn-deep-lab-*

Cross-Reference Index

Jump to related ForgeLearn Git pages for adjacent skills.

How to Master Git — curriculum
Commands Reference — encyclopedia
Git Roadmap — staged learning path
Basics — three-tree model
Best Practices — team conventions
Reflog — recovery
Revert — safe public undo

Community

Get help on Slack, Discord or VIP

Stuck on a guide? Join the community and ask.

Join Slack Join Discord Kaundal VIP

X GitHub YouTube Instagram Facebook Discord Slack Telegram Buy me a coffee Ko-fi Sponsor