ai-time-saved

See how long your coding task would take a human — after AI just did it.

中文文档 • Install • FAQ

---

You just finished a feature in 3 minutes with Claude Code. But how long would it have taken a human? ai-time-saved tells you — automatically, after every task.

Just finished: Add work time tracking to monitoring dashboard

🧑‍💻 Junior engineer: ~4-6 hours
   — Understanding the plugin architecture alone takes 1-2 hours
🧑‍💼 Senior engineer: ~1.5 hours
   — Knows the pattern, just writes it
🤖 AI changed 4 files, ~120 lines of code
   — New sampler script + check logic + HTMX template + page layout

Fair warning: The estimates are LLM vibes, not stopwatch science. Think of it as a fun speedometer, not a billing system.

How It Works

 You complete a coding task with Claude Code
          │
          ▼
 Claude detects the task is done (commit, summary, etc.)
          │
          ▼
 The skill auto-triggers and reviews what was changed
          │
          ▼
 Outputs a casual Junior / Senior / AI time comparison

One file, zero dependencies:

Component	What it does
SKILL.md	A prompt-based Claude Code skill — no hooks, no scripts, no background processes. Claude reads it, follows the instructions, done.

Install

git clone https://github.com/zchdu/ai-time-saved.git
cd ai-time-saved
bash install.sh

This copies the skill to ~/.claude/skills/time-estimate/. That's it.

Or manually:

mkdir -p ~/.claude/skills/time-estimate
cp skills/time-estimate/SKILL.md ~/.claude/skills/time-estimate/

Usage

Auto mode (default): Claude triggers the estimate automatically after completing a coding task — new features, bug fixes, refactors, config changes.

Manual mode: Type /time-estimate anytime to get an estimate for the work just done.

Estimation Logic

The skill uses rough baselines, adjusted for actual task complexity:

Task type	Junior (0-2yr)	Senior (5+yr)
Config / text tweak	15-30 min	5-15 min
Single-file bug fix	1-3 hr	20-60 min
New endpoint / function	2-5 hr	30-90 min
Multi-file feature (frontend + backend)	4-12 hr	1.5-4 hr
Complex refactor (many files)	1-3 days	4-8 hr
New system from scratch	2-5 days	1-2 days

Adjustment factors:

• Unfamiliar codebase → junior +100%, senior +30%
• Complex business logic → +30-50%
• External API integration → +20-40%
• Well-documented codebase → -20%

File Structure

~/.claude/
└── skills/
    └── time-estimate/
        └── SKILL.md        # The entire skill — one file

FAQ

How accurate are the estimates?

Ballpark. Expect 2-3x variance. A "2 hour" estimate might mean 1-4 hours in reality. It's directionally useful, not precise.

Does it cost extra tokens?

~1-2k tokens per invocation. Negligible compared to the task itself.

Can I stop it from auto-triggering?

Add disable-model-invocation: true to the frontmatter in SKILL.md. Then it only runs when you type /time-estimate.

Does it work in languages other than English?

Yes. It adapts to the language you're using in the conversation. The examples show Chinese output because the author speaks Chinese.

How do I uninstall?

bash uninstall.sh

Or just rm -rf ~/.claude/skills/time-estimate.

Requirements

• Claude Code with skills support

Contributing

Issues and PRs welcome. Keep it simple — it's a toy.

License

MIT