Y4rd13

Agent Harness Template — AI skill for Claude Code

AI community

Templates, a Claude Code plugin and lessons for making an AI coding assistant follow your repository's conventions, with deterministic and LLM CI layers.

How to install Agent Harness Template

This entry records only its repository, not the path inside it, so there is no exact command to give. Open Y4rd13/agent-harness-template and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Agent Harness Template does

Templates, a Claude Code plugin and lessons for making an AI coding assistant follow your repository's conventions, with deterministic and LLM CI layers.

Alternatives in AI

Surface — Show ranked attack surface for a target from its recon manifest + hunt memory 3.8k ★
LongHorizon Harness — The long-horizon computer-use harness 1.3k ★
Loushang — AI-native agent harness for coding workflows by python: multi-model LLM orchestration, stateful sessions, tool 1.3k ★

README

agent-harness-template

[Español](README.es.md)

Templates, a Claude Code plugin and short lessons for making an AI coding assistant follow your repository's conventions, with checks that do not depend on it remembering them.

The repository is mostly Markdown. It holds almost no code: where a check needs logic, you get pseudocode to port, and code appears only where nothing works without it (one hook, one shell step, one review gate).

The idea: layers, from advice to enforcement

  advice                                                                    enforcement
  ------------------------------------------------------------------------------------>
  AGENTS.md          .claude/rules/      skills          settings + hooks   CI checks         AI review gate     human
  CLAUDE.md          (path-scoped)       subagents       (client-side)      (deterministic)   (model + code)     review
  every session      when a path         on request      every tool call    every PR          every PR           merge
                     is touched

Instruction files tell the agent how the repository works. It usually follows them; nothing forces it.
Settings and hooks are applied by Claude Code whatever the model decides.
CI fails the pull request for everyone: humans, any agent, any tool.
The AI review catches what no rule states, inside a sandbox, and only blocks on evidence a script verified.
A person owns the merge.

Rule of thumb: if a machine can check it, write the check; if it must always hold, use a hook or a required check; only what is left goes into prose.

Quickstart A: let the plugin set it up

claude plugin marketplace add /agent-harness-template
claude plugin install agent-harness@agent-harness-template

Then, inside Claude Code in your repository:

Command	What it does
`/agent-harness:init dry-run`	Reads your repository and prints the harness it would create, with yo