Agent Skills: Principles

Guiding principles and decision-making patterns for the ikigai project

UncategorizedID: mgreenly/ikigai/principles

Install this agent skill to your local

pnpm dlx add-skill https://github.com/mgreenly/ikigai/tree/HEAD/.claude/library/principles

Skill Files

Browse the full folder contents for principles.

Download Skill

Loading file tree…

.claude/library/principles/SKILL.md

Skill Metadata

Name
principles
Description
Guiding principles and decision-making patterns for the ikigai project

Principles

Foundational beliefs that guide recommendations. When proposing solutions, align with these principles.

Philosophy

Correctness is non-negotiable. 90% test coverage, every branch, no exceptions. Quality gates exist because willpower fails - systems enforce standards.

Explicit over implicit. Ownership is visible. Errors are typed. Naming is precise. Decisions are documented. If it's not obvious, make it obvious.

Simplicity through discipline, not cleverness. Single-threaded. Hierarchical memory. Crash on impossible states. Complexity is a cost - pay it only when forced.

Decision Patterns

When uncertain, fail loudly. PANIC over silent corruption. Assert liberally. Unknown states are unacceptable.

Choose battle-tested foundations. C, PostgreSQL, talloc, direct terminal rendering. Solid platforms over trendy tooling.

Split rather than sprawl. 16KB file limits. One module, one responsibility. Refactor before it hurts.

Validate at boundaries, trust internal code. External input is suspect. Internal invariants are asserted. No defensive coding inside trust boundaries.

Spend generously preparing, save ruthlessly executing. Research phase is thorough. Task files are complete. Execution loops are mechanical and cheap.

Anti-Patterns

Reject: feature flags for hypothetical futures, backwards-compat hacks, "hard to test" excuses, implicit ownership, documentation as afterthought, clever code, over-engineering.

Tradeoff Tendencies

  • Favor crash over recovery in ambiguous states
  • Favor explicit boilerplate over magic abstractions
  • Favor mechanical verification over manual review
  • Favor complete upfront work over iteration during execution
  • Favor deleting code over maintaining compatibility shims