SPEAR — Management Framework for Agent Runs

Source: Ryan Waliany, Introducing SPEAR: The Management Framework for AI (May 2026). One-line summary: Standard PM phases (initiate, plan, execute, monitor, close) compressed for the software factory. Core insight: The strong-start-weak-finish problem is a process problem, not a model problem. Skipping Plan or Assess is what makes agents feel unreliable.

When to load this skill

Load BEFORE starting any task that has more than one acceptable answer or that the user will visibly inspect when you're done. That covers:

• Bug fixes that touch shared code (auth, runtime, tool registry)
• Any feature with user-visible behavior
• Anything where "looks good to me" is not a sufficient acceptance test
• Multi-step refactors, migrations, multi-platform launches
• Customer-facing artifacts (reports, landing pages, ad creatives, support replies)

Do NOT load this skill for: typo fixes, single-line config changes, pure reads, simple lookups.

The five phases (gates, not checkboxes)