compare-agent-workflows

Compare Agent Workflows

Purpose

Use this skill after research has produced candidates. The goal is to decide which ideas are useful for GroundLine and which are noise.

Routing Boundary

Use this after research has produced two or more candidates, or one candidate plus a clear GroundLine baseline. Use evaluate-agent-capability first when the user asks whether one artifact is any good. Use recommend-groundline-upgrades when the ranking already exists and the remaining task is a GroundLine change decision.

Comparison Axes

Score each candidate in plain text:

• runtime fit: Codex, Claude Code, Antigravity, or provider-neutral.
• job fit: current-state proof, research, planning, review, handoff, verification, safety, setup, or release.
• context cost: always loaded, skill-triggered, reference-only, script-only, or external tool.
• setup weight: none, optional CLI, optional MCP, plugin install, or hook.
• mutation risk: read-only, local write, provider-home write, remote write, production, access, billing, or secret exposure.
• overlap: replaces, complements, or duplicates an existing GroundLine skill.
• maintenance: stable primary source, active churn, unclear owner, or stale.

Comparison Rules

• Favor narrow, triggerable skills over global instructions.
• Favor source-backed references over copied prompt packs.
• Treat hooks as deterministic guardrails, not general workflow memory.
• Treat MCP as optional setup unless the user asks to enable it.
• Reject candidates that require broad provider scope outside GroundLine.

Output Contract

Use GroundLine Comparison from references/output-contracts.md.