whygit

Capture the why behind AI-assisted code changes, committed alongside the code itself.

Git records what changed. whygit records why — the reasoning, the alternatives considered, the prompts that shaped the direction. One command at the end of a Claude Code session. Everything committed as markdown, living alongside your code.

---

Install

One-liner (curl):

curl -fsSL https://raw.githubusercontent.com/warunacds/whygit/main/install-whygit.sh | bash

This installs whygit into the current directory (must be a git repo).

To install into a specific repo:

curl -fsSL https://raw.githubusercontent.com/warunacds/whygit/main/install-whygit.sh -o install-whygit.sh
chmod +x install-whygit.sh
./install-whygit.sh /path/to/your/repo

Manual:

git clone https://github.com/warunacds/whygit.git
cd whygit
./install-whygit.sh /path/to/your/repo

Then commit the installed files:

git add . && git commit -m "chore: add AI decision logging"

---

Usage

In any Claude Code session:

/commit              → log session reasoning + stage all + commit
/log                 → write log only, no commit
/rewind              → browse all sessions newest first
/rewind auth         → find sessions related to "auth"
/rewind 2026-02-26   → show all logs from that date
/learn               → mine ai-logs/ for reusable skills
/skills              → list active skills and unresolved conflicts

That's it.

---

How it works

whygit installs three Claude Code slash commands and a CLAUDE.md directive into your repo:

your-repo/
├── CLAUDE.md                        # Tells Claude to log reasoning and read skills
├── .claude/
│   ├── commands/
│   │   ├── commit.md                # /commit slash command
│   │   ├── log.md                   # /log slash command
│   │   ├── rewind.md                # /rewind slash command
│   │   ├── learn.md                 # /learn slash command
│   │   └── skills.md                # /skills slash command
│   └── skills/                      # Learned guardrails (grows via /learn)
│       ├── curl-pipe-bash-self-contained.md
│       └── ...
└── ai-logs/
    ├── 2026-02-26-auth-refactor.md
    ├── 2026-02-24-payment-webhooks.md
    └── ...

Each log captures:

• What was built or changed
• Key decisions and reasoning
• Alternatives considered
• Prompts that shaped direction
• Known limitations and follow-up

Logs are plain markdown. They live in your repo. They're committed with your code. Nothing leaves your machine.

The learning loop

Logs answer "why did we do X." Over time, you can turn them into guardrails:

• /learn reads unprocessed logs, finds lessons (wrong turns, shipped bugs, reversed decisions), and proposes concrete rules to save in .claude/skills/.
• Claude Code reads those skills at the start of every session, so old mistakes don't repeat.
• /skills lists what's in your rulebook today.

Nothing is automatic. You approve every skill before it's written, every lesson before it's committed.

---

Comparison

	whygit	entire.io
Capture method	On-demand via /commit	Automatic, continuous
Content	Curated summary	Full transcript + tool calls
Storage	ai-logs/ in your repo	Hidden git branch
Dependencies	None	CLI + account
Privacy	Fully local	Cloud dashboard
Cost	Free	TBD

Philosophy: entire.io captures everything automatically. whygit captures curated reasoning on demand. They're complementary — entire.io is the raw tape, whygit is the edited highlights.

---

Non-goals

• Not a replacement for good commit messages
• Not a full session transcript tool
• Not a web dashboard
• Not dependent on any external API at commit time

---

License

MIT — see LICENSE