Agent Claude Code
An independent reviewer for changes to project rules. It checks whether adding, changing, or removing a rule leaves all references, automated hooks, evaluation links, and related documentation connected correctly.
Agent Claude Code
An independent reviewer for sandbox setup scripts and their verification records. A sandbox is an isolated test environment; the review checks whether it really starts, stops, reports status, handles repeated commands safely, and uses the right data definitions.
Agent Claude Code
A worker that turns written test cases into executable test scripts and runs them in a sandbox. It follows the project's existing language and test style and keeps end-to-end and API test identifiers distinct.
Agent Claude Code
An independent reviewer for executable test scripts. It checks whether each script matches its documented test case and whether the reported results are supported by actual run output.
Agent Claude Code
A review agent for examining and improving the rules and workflows that guide coding agents. It works from evidence and records findings without changing business code.
Agent Claude Code
A skeptical reviewer for technical designs and API contracts—the documents that explain how a software change should be built and how its interfaces behave. It checks the proposal against the real project and the original requirements.
Agent Claude Code
An agent that turns collected requirements into user stories and acceptance criteria. User stories describe what a person needs, while acceptance criteria state observable results that can be checked.
Hook Claude Code
Runs before the agent uses a tool for Bash tool calls, executing hook-policy.sh. From harness-base/harness-control.
Hook Claude Code
Runs when the agent finishes a response, executing stop-check.sh via bash. From harness-base/harness-control.
Hook Claude Code
Runs when you submit a prompt, before the agent sees it, executing correction-nudge.sh via bash. From harness-base/harness-control.
Settings file Claude Code
Agent settings declaring 3 hook events (PreToolUse, Stop, UserPromptSubmit).
Instructions file CodexOpenCode
Instructions for harness-base/harness-control, covering agent harness 控制面, 启动顺序, 规则, 验证 and eval.
Instructions file
Instructions for harness-base/harness-control.
Skill Claude CodeCodex
A project-specific skill template written in Chinese that describes scope, steps, hard rules, and how to update the skill as the project changes.