Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Initialise a git repository with optional agent commit instructions and .gitignore. Use when users say "here be git", "init git", "initialise git", or otherwise indicate they want to set up version control in the current directory.
.claude/skills/mkurman-here-be-git/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-07 | ✗→✓ | ▲ Improved | 9% | 0% |
| case-09 | ✗→✓ | ▲ Improved | 79% | 0% |
| case-05 | ✓→✗ | ▼ Worse | 125% | 0% |
| case-08 | ✓→✗ | ▼ Worse | -10% | 0% |
| case-10 | ✓→✗ | ▼ Worse | -6% | 0% |
Initialise a git repository with optional configuration for agent workflows.
Run git init in the current working directory. Confirm to the user that the repository has been initialised.
Ask the user:
> Would you like me to add instructions for the agent to always commit when it's done with a task?
If the user confirms:
AGENTS.md exists in the current directoryThe commit instructions to add:
markdown## Git Workflow — CRITICAL, MANDATORY, NO EXCEPTIONS > **THIS IS NOT OPTIONAL.** Every agent operating in this repository MUST follow this workflow after every change, no matter how small. There are ZERO exceptions to this rule. Skipping a commit is never acceptable. If you have made any change to any file, you MUST commit before moving on. ### Rules 1. **ALWAYS commit your changes immediately after completing a task or reaching any logical stopping point.** Do not defer. Do not batch. Do not skip. 2. **ALWAYS commit before ending your session.** The working directory MUST be clean (all changes committed) when you finish. Leaving uncommitted changes is a failure state. 3. **Use clear, descriptive commit messages** that explain what was done and why. 4. **If in doubt, commit.** It is always better to have one commit too many than one too few. ### To be absolutely clear - "I'll commit later" — **No. Commit now.** - "It's just a small change" — **Commit it.** - "I'm about to make another change" — **Commit the current one first.** - "The user didn't ask me to commit" — **Irrelevant. Commit anyway.**
After creating or updating AGENTS.md, commit it with an appropriate message.
Ask the user:
> Would you like me to create a .gitignore? If so, what flavour or patterns should I include? (e.g., Node.js, Python, macOS, IDE files, or specific files/patterns)
If the user provides a flavour or patterns:
.gitignore based on their inputnode_modules/, dist/, .env, *.log, etc.__pycache__/, *.pyc, .venv/, venv/, .env, *.egg-info/, etc..DS_Store, .AppleDouble, .LSOverride, ._*.idea/, .vscode/, *.swp, *.swo, *.sublime-*.gitignore with an appropriate messageIf the user declines, skip this step.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 10,800 | 5,715 | -47% | 1 | 1 | 0% | 1,831 | 1,059 | -42% | 0 | 0 | — |
case-02 | fail→fail | 11,765 | 7,926 | -33% | 1 | 1 | 0% | 2,015 | 903 | -55% | 0 | 0 | — |
case-03 | fail→fail | 5,991 | 5,655 | -6% | 1 | 1 | 0% | 326 | 875 | +168% | 0 | 0 | — |
case-04 | pass→pass | 9,713 | 8,713 | -10% | 1 | 1 | 0% | 1,659 | 2,182 | +32% | 0 | 0 | — |
case-05 | pass→fail | 4,671 | 7,111 | +52% | 1 | 1 | 0% | 874 | 1,967 | +125% | 0 | 0 | — |
case-06 | pass→pass | 8,804 | 7,524 | -15% | 1 | 1 | 0% | 1,618 | 2,080 | +29% | 0 | 0 | — |
case-07 | fail→pass | 7,652 | 9,470 | +24% | 1 | 1 | 0% | 1,244 | 1,358 | +9% | 0 | 0 | — |
case-08 | pass→fail | 8,743 | 3,618 | -59% | 1 | 1 | 0% | 1,510 | 1,362 | -10% | 0 | 0 | — |
case-09 | fail→pass | 7,628 | 8,245 | +8% | 1 | 1 | 0% | 1,213 | 2,172 | +79% | 0 | 0 | — |
case-10 | pass→fail | 6,228 | 1,783 | -71% | 1 | 1 | 0% | 1,054 | 994 | -6% | 0 | 0 | — |
case-11 | pass→pass | 10,424 | 5,417 | -48% | 1 | 1 | 0% | 1,766 | 1,673 | -5% | 0 | 0 | — |
case-12 | pass→pass | 11,918 | 3,800 | -68% | 1 | 1 | 0% | 2,066 | 1,388 | -33% | 0 | 0 | — |
case-13 | pass→pass | 7,456 | 3,764 | -50% | 1 | 1 | 0% | 1,408 | 1,387 | -1% | 0 | 0 | — |
case-14 | pass→pass | 7,909 | 3,029 | -62% | 1 | 1 | 0% | 1,281 | 1,222 | -5% | 0 | 0 | — |
case-15 | fail→fail | 6,914 | 3,744 | -46% | 1 | 1 | 0% | 1,122 | 1,223 | +9% | 0 | 0 | — |
case-16 | pass→pass | 5,074 | 1,783 | -65% | 1 | 1 | 0% | 846 | 1,010 | +19% | 0 | 0 | — |
case-17 | pass→pass | 11,663 | 6,178 | -47% | 1 | 1 | 0% | 2,013 | 1,737 | -14% | 0 | 0 | — |
case-18 | fail→fail | 9,923 | 5,662 | -43% | 1 | 1 | 0% | 1,561 | 1,575 | +1% | 0 | 0 | — |
case-19 | fail→fail | 8,596 | 1,464 | -83% | 1 | 1 | 0% | 1,262 | 950 | -25% | 0 | 0 | — |
case-20 | pass→pass | 9,016 | 3,260 | -64% | 1 | 1 | 0% | 1,628 | 1,283 | -21% | 0 | 0 | — |
case-21 | pass→pass | 3,343 | 1,775 | -47% | 1 | 1 | 0% | 498 | 951 | +91% | 0 | 0 | — |
case-22 | pass→pass | 4,052 | 2,867 | -29% | 1 | 1 | 0% | 604 | 1,263 | +109% | 0 | 0 | — |
case-23 | pass→pass | 9,335 | 4,374 | -53% | 1 | 1 | 0% | 1,596 | 1,446 | -9% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 23 cases were attempted, and 21 counted toward the lift figure. The other 2 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of -20 percentage points is the difference between those two pass rates over the 21 comparable cases. 4 cases got worse with the skill loaded, and they are included in that figure.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.