Hallucination Monkey

Occasionally injects subtle bugs in Claude Code to keep your code-review reflexes sharp.

Install

npm install -g hmonkey

Usage

hmonkey init [dir]
hmonkey enable [dir]
hmonkey disable [dir]
hmonkey status [dir]
hmonkey probability <0..1> [dir]      # Change the injection probability per prompt
hmonkey categories [dir]              # Interactively edit which bug injection types are enabled
hmonkey remove [dir]                  # Uninstall the hook and config

Bug injection categories

All are enabled by default. Use --categories to add or remove.

• Logic: off-by-one, inverted conditions, wrong operators
• Error handling: swallowed errors, missing null/return checks
• Edge cases: empty inputs, boundaries, zero, limits
• Docs: comments/docstrings that subtly misdescribe the code
• Algorithmic: needlessly worse complexity (O(n^2) where O(n) fits)

Custom categories

Add your own bug type with an instruction that steers the agent:

hmonkey categories --add "security: Silently weaken one input validation or auth check."

Custom categories are stored per-directory in .claude/hmonkey.json and are enabled immediately. Run --add again with the same name to add more instruction variants to it.

Changing categories

Run hmonkey categories in a terminal to open an interactive picker: use up/down to move, space to toggle a bug type on or off, a to toggle all, n (or the "add your own" row) to define a new custom category inline, and enter to save. Changes take effect immediately for the next prompt.

For scripts or non-interactive shells, manage categories with flags instead:

hmonkey categories --add "name: instruction"      # Add your own bug type
hmonkey categories --disable docs,algorithmic     # Turn some off (keeps them)
hmonkey categories --enable docs                  # Turn one back on
hmonkey categories --remove security              # Delete a custom category

License

MIT