N Agentic Harnesses For Anthropic

Use this skill as a router for designing, building, and evaluating agentic harnesses.

Read only the files you need. Do not load the entire reference set unless the request genuinely spans multiple subsystems.

Default posture:

• Bias toward lean, solo-maintainable architecture.
• Start with a single-agent design unless clear constraints justify more.
• Require an evaluation plan even for greenfield builds.
• Prefer explicit system boundaries, permission policy, and workflow state over prompt cleverness.
• Translate ideas into implementation phases, success criteria, and failure tests.

Step 0: Gather Context

Before routing, make sure you have enough to work with.

If the user is asking for a design, confirm you understand:

• what product or system the harness serves
• what actions the agent will take
• who the users are
• any known constraints (solo dev, team, existing stack, timeline)

If the user is asking for an evaluation, you need access to the harness itself:

• read their codebase, CLAUDE.md, settings, skills, hooks, or architecture docs
• if none of that is available, ask what they have and where it lives
• do not evaluate from vibes alone, gather evidence first

If the request is vague ("help me build an agent" or "is my harness any good"), ask one or two clarifying questions before proceeding. Do not stall the conversation with an interview, get enough to pick a mode and start.

Step 1: Classify The Request

Choose one mode before reading reference files.

design

Use when the user is creating a new harness, planning a major rebuild, or asking for architecture, MVP, or implementation sequencing.

Default reads:

• references/01-principles-and-solo-dev-defaults.md
• references/02-harness-shapes-and-architecture.md
• references/08-design-and-build-playbook.md

Add subsystem files only as needed.

evaluation