agentsclimarketplace

N agentic harnesses

Skill mattdweigand-sketch/agent-skills/skills/n-agentic-harnesses

Portable agent skill library for Codex, Claude Code, and other AGENTS-aware tools. Covers eval loops, research, writing, and project hygiene.

Install
npx -y skills add mattdweigand-sketch/agent-skills --skill n-agentic-harnesses

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Design, evaluate, and improve agentic harnesses — the orchestration layer around LLM-powered tools, agents, assistants, copilots, workflow runtimes, and AI-driven product features. Use this skill whenever the user mentions building an agentic system, structuring tool use, adding permissions or approval gates, designing multi-step AI workflows, managing context windows or memory, making agents durable or resumable, evaluating or pressure-testing an existing harness, planning phased implementation for an AI product, reviewing agent architecture, improving agent UX or observability, or asking how to know if their harness is actually good. Also use when the user describes problems that imply harness gaps — like agents doing unexpected things, context getting stale, sessions not surviving crashes, tools running without permission, or costs spiraling — even if they do not use the word "harness."

SKILL.md

4.9 KB, as published. Nobody here has run it

N Agentic Harnesses

Router for designing, building, and evaluating agentic harnesses. Read only the files you need.

Default posture: Lean, solo-maintainable architecture. Single-agent design unless constraints justify more. Require an evaluation plan even for greenfield builds. Explicit system boundaries and permission policy over prompt cleverness.

Step 0: Gather Context

Request TypeConfirm
DesignProduct/system served, agent actions, users, constraints (solo dev / team / stack / timeline)
EvaluationAccess to codebase, AGENTS.md, settings, skills, hooks, or architecture docs
VagueAsk 1-2 clarifying questions, then pick a mode

Step 1: Classify the Request

ModeUse WhenDefault Reads
designNew harness, major rebuild, architecture, MVP, or implementation sequencing01, 02, 08
evaluationExisting harness — gaps, risks, missing primitives, UX/ops upgrades01, 09
design + evaluationTarget architecture + comparison to current state, acceptance criteria01, 02, 08, 09

Step 2: Classify the Product Shape

  • code agent / chat assistant / workflow orchestrator / internal copilot / embedded AI product feature / hybrid system

Pick the closest shape and state the assumption if ambiguous.

Step 3: Read the Smallest Useful Reference Set

FileRead When
references/01-principles-and-solo-dev-defaults.mdAlmost every request — defines default decision posture
references/02-harness-shapes-and-architecture.mdSystem shape, boundaries, lifecycle, transports, deployment
references/03-tools-execution-and-permissions.mdTool registries, tool calling, approval gates, sandboxes, trust tiers
references/04-state-sessions-and-durability.mdSessions, resumability, retries, idempotency, approval waits, long-running work
references/05-context-memory-and-evaluation.mdContext windows, retrieval, memory, provenance, evals, replay tests
references/06-agents-and-extensibility.mdMulti-agent design, plugins, hooks, skills, extension surfaces
references/07-ux-observability-and-operations.mdStreaming UX, health checks, logs, analytics, budgets, supportability
references/08-design-and-build-playbook.mdBuild-ready plan from idea to implementation
references/09-evaluation-and-improvement-playbook.mdFindings, missing primitives, upgrade priorities, acceptance tests
references/10-example-requests-and-output-patterns.mdPrompt examples or response structure examples
references/11-platform-adaptation-notes.mdAdapting the skill to platform-specific runtimes without changing canonical guidance

Do not rely on reference-to-reference chains. This file is the index.

Output Contract

design: recommended shape, core primitives, MVP boundary, phased plan, verification criteria evaluation: findings by severity, missing/weak primitives, UX/ops gaps, upgrade path, tests design + evaluation: target architecture, comparison vs current, phases, acceptance criteria, eval plan

Operating Rules

  • Convert vague ambitions into concrete harness primitives.
  • Push back on unnecessary complexity.
  • Treat workflow state, permissions, context assembly, and evaluation as first-class architecture.
  • For evaluation: findings first, improvement sequence second.
  • For design: include how the design will be tested before calling it done.

Final Check Before Responding

Before writing the response, verify:

  • Request was classified — mode is explicitly chosen (design / evaluation / design+evaluation)
  • Multi-agent was not recommended unless the user's constraints make single-agent insufficient
  • Eval plan is present — design responses include acceptance criteria; evaluation responses include upgrade sequencing

Keep looking

Skills are one crate of 328,083. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.