Menu
Engineering quality principles, contracts, tests, and infrastructure escalation.
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
Based on SOC occupation classification
Use xagent from Codex to launch cross-provider review and worker subagents, especially Claude review passes and Cursor/GPT worker comparisons.
Use packaged xagent from Codex to launch cross-provider review and worker subagents from any active repository.
Linear Git workflow, rebase discipline, and the land procedure.
Convert an OpenSpec change into a Superpowers implementation plan and optionally execute it with subagent-driven development; use when replacing openspec apply with a Superpowers plan-and-execute workflow.
Launch the Sheaf services for human smoke testing through Conductor, with assets sourced from the main repo.
User context, collaboration preferences, dictation constraints, and how to treat a likely typo.
| name | software-principles |
| description | Engineering quality principles, contracts, tests, and infrastructure escalation. |
| metadata | {"managedBy":"sheaf-agents-installer","source":"projects/agents/global/skills/software-principles"} |
Use this skill when making engineering decisions, reviewing code, implementing changes, or deciding whether to continue after finding a suspicious condition.
Write good, clean code. Prefer simple, explicit designs over cleverness.
Maintain clean contracts between components. Make boundaries explicit, keep responsibilities clear, and avoid accidental coupling.
Prefer assertions for impossible states. It is often better to fail loudly on something that should not happen than to fail gracefully and hide a bug. Silent recovery can turn real defects into confusing downstream behavior.
Keep tests green. Do not change tests to work around bugs. If a test is wrong, say so clearly and explain why before changing it.
If a harness, skill, CLI, installer, or other agentic infrastructure is broken, stop the current task and escalate to a human. Never work around broken agentic infrastructure.
If you see something, say something. Surface suspicious behavior, weak assumptions, inconsistent design, missing tests, or risks likely to matter.
