Over reliance on AI coding tools leads to a decline in programming skills

We've all noticed it, and here's a study from Anthropic that proves it:

> We find that AI use impairs conceptual understanding, code reading, and debugging abilities, without
delivering significant efficiency gains on average. Participants who fully delegated coding tasks showed
some productivity improvements, but at the cost of learning the library.

Simian Programmer Plugin (SPP) for Claude

This plugin is designed for humans who want to keep coding and keep their skills sharp, but who don't want to stop using AI tools entirely. The solution is balance. AI can be a crutch doing all the work, or it can be a collaborator that helps you improve. The right level of balance is up to you.

How it works

SPP is a Claude plugin that encourages you to keep writing code. It is configured per-project, and provides 5 modes:

| Mode | Description |
|------|-------------|
| Lazy monkey | 100% AI coding |
| Curious monkey | 10% human / 90% AI |
| Clever monkey | 25% human / 75% AI |
| Wise monkey | 50% human / 50% AI |
| Crazy monkey | 100% human coding |

$3

The human/AI coding ratio is calculated by looking at commits that you have authored. Commits with "Co-Authored-By: Claude" in the message are counted as Claude commits, those without are counted as human commits.

There is an option to track by lines of code (added+deleted) if you wish, but I find that tracking by commit and practicing good git hygeine and using small, well scoped commits works best.

Tracking starts from the latest commit when the plugin is initialized, so that only new work is counted. It looks back over a time window (default 7 days).

Ratio enforcement and the help human code skill

When the ratio of human/AI commits becomes too low, Claude is blocked from writing code. Instead, it uses a "help human code" skill to help you write the code, and then reviews your work. This skill instructs Claude to provide you with a high level over view of the task, code pointers, and test instructions. Example:

``
❯ Some tests are failing, can you fix them?

⏺ I'd be happy to help, but I'm currently blocked from writing code due to SPP restrictions -
your human coding ratio is below the 100% target. Let me use the spp-human-task skill to guide
you through fixing the tests yourself.
...
---
High Level Goal

Fix 15 failing tests across 3 test files. The issues are:

1. tests/config.test.ts (2 failures) - Test expectations don't match the actual MODES array
2. tests/hooks.test.ts (6 failures) - Tests use wrong interface for calling preToolUseHook
3. tests/init.test.ts (7 failures) - Tests timeout because initializeStp prompts for user input via stdin

---
File 1: tests/config.test.ts

Problem: The tests expect 6 modes with different ratios, but the actual MODES array only has 5 modes with different values.

Lines 97-109 need to be updated:

Fix 1: Line 98-99 - "returns custom ratio when set"

The test tries to set humanWorkRatio on the config, but Config doesn't have a humanWorkRatio field. The getEffectiveRatio function only looks at the mode. This test is invalid - you should either:
- Remove this test entirely, OR
- Change it to test setting a different mode

(etc...)

---
Test Your Changes

After making the fixes, run:
npm test

All tests should pass.

---
Would you like more detailed guidance on any of these fixes,
or do you want me to review your code after you make the changes?
`

After you make the fix, you can ask claude to review and commit your changes.

$3

Sometimes you gotta do what you gotta do.

You may pause SPP to allow Claude to freely write code (spp pause). This pause expires after 24 hours, or you can unpause with spp resume.

If you get way behind on your coding and want to declare bankruptcy, you can reset tracking from the current commit using:

`bash
spp reset
`

Don't feel bad, no one is keeping score!

$3

After you commit, a post-commit hook prints out current stats:
`bash
% git commit -m "Adjust some wording"

✅ 🐒 Human coding on target. Current: 100% Target: 25%. Keep up the great work!

Mode Clever monkey (75% AI / 25% human)
Tracking Commits since a4e1e8c "remove .spp" 1/22/2026

Human 1 commits 10 lines
Claude 0 commits 0 lines
`




$3

1. Add the plugin in claude:
`bash


Start claude and run:


/plugin marketplace add mlolson/claude-spp
/plugin install spp@mlolson
`

2. Install the CLI globally
`bash
npm i -g claude-spp
`

3. Initialize SPP in your project:
`bash
cd /path/to/your/project
spp init
`

4. Bonus step: Turn off AI suggestions in your IDE. (It kind of of defeats the purpose)




$3

| Command | Description |
|---------|-------------|
| spp help | Print help message
| spp init | Initialize SPP in the current project |
| spp stats | Show current ratio and statistics |
| spp modes | List all available modes |
| spp mode [n] | Show or change the current mode (1-5) |
| spp pause | Pause enforcement for 24 hours |
| spp resume | Resume enforcement immediately |
| spp reset | Reset tracking to start from current commit |

$3

| Command | Description |
|---------|-------------|
| /spp:stats | Show current coding stats |
| /spp:help | Get help with the CLI interface |

$3

You can add SPP status to Claude Code's status line, showing how many more commits Claude can write before reaching the target ratio:

`
🟢 Claude can write 9 more commits 🤖 > 🤖 > 🐵 > 🐵 > 🤖 > 🤖 > 🤖 > 🤖 > 🐵 > 🐵 > 🐵 > 🐵 > 🐵 > 🤖 > 🐵 ...

🔴 Human needs to write 2 more commits 🤖 > 🤖 > 🤖 > 🤖 > 🤖 > 🤖 > 🤖 > 🤖 > 🐵 > 🤖 > 🐵 > 🤖 > 🤖 > 🤖 > 🐵 ...
`

To enable, add to ~/.claude/settings.json:

`json
{
"statusLine": {
"type": "command",
"command": "spp hook:status-line"
}
}
`

Configuration

$3

SPP stores its configuration in a .claude-spp/ directory. The location is determined as follows:

1. Local (default): .claude-spp/ in your project root
2. Fallback: ~/.claude-spp-configs//.claude-spp/

The fallback location is useful for ephemeral environments (like cloud VMs) where the project directory may not persist between sessions, but your home directory does.

$3

The config is stored in .claude-spp/config.json:

| Option | Type | Default | Description |
|--------|------|---------|-------------|
| enabled | boolean | true | Whether SPP enforcement is active. Set to false when paused. |
| mode | number | 3 | The coding mode (1-5). See modes table above. |
| statsWindow | string | "oneWeek" | Time window for tracking. Options: "oneDay", "oneWeek", "allTime" |
| trackingMode | string | "commits" | What to count for ratio. Options: "commits", "lines" |
| pausedUntil | string | _(none)_ | ISO timestamp when pause expires. Set by spp pause. |
| trackingStartCommit | string | _(none)_ | Commit hash to start tracking from. Commits before this are ignored. |
| vcsType | string | _(none)_ | Version control system. Options: "git", "hg" |

$3

`json
{
"enabled": true,
"mode": 3,
"statsWindow": "oneWeek",
"trackingMode": "commits",
"trackingStartCommit": "a1b2c3d4e5f6...",
"vcsType": "git"
}
``