README.md

Sufficient Reason

An agent skill for shaping proportionate software changes before code structure and scope harden into decisions.

It asks one question at the moment a mechanism is chosen: what pays for this? A layer, file, abstraction, control, default or process must serve a requirement, consumer, constraint, obligation, reachable risk or deliberate decision. The test also catches under-engineering: when a real domain distinction needs a type or boundary, refusing to model it fails for the same reason.

What it changes

• Keeps a direct task from becoming an unrequested framework or operating system.
• Finds the existing owner before adding another implementation of the same concern.
• Prevents plausible guesses from becoming unverified defaults.
• Models recurring domain distinctions without abstracting predicted future cases.
• Keeps security and verification controls proportionate to reachable risks.

Install

npx skills add CodingCossack/sufficient-reason -g -y

For selected agents:

npx skills add CodingCossack/sufficient-reason -g -y \
  -a codex -a claude-code -a gemini-cli -a opencode

Codex native installer:

$skill-installer install https://github.com/CodingCossack/sufficient-reason/tree/main/skills/sufficient-reason

Use

The skill can route automatically while a software change still has open decisions about scope, ownership, defaults, structure or controls. Explicit invocation also works:

Use $sufficient-reason to shape this change and keep every mechanism tied to a real need.

Responsibility boundary

• sufficient-reason shapes the change while its design is still open.
• change-with-proof proves changed behaviour and completion claims.
• anti-machinery decides what supporting apparatus may remain after the task.

The skills complement each other but remain independently installable.

Structure

skills/sufficient-reason/
├── SKILL.md
└── agents/
    └── openai.yaml

Licence

MIT