Agent Skills › pony-maggie/code_minions › implement-with-tdd

implement-with-tdd

GitHub

基于 TDD 流程实现单个任务，结合 AI 代码审查循环迭代直至收敛。通过强制测试执行和有限轮次的人工介入机制，确保代码质量与功能交付。

src/code_minions/builtin/skills/implement-with-tdd/SKILL.md pony-maggie/code_minions

Trigger Scenarios

需要按照 TDD 规范进行编码的任务希望利用 AI 辅助进行代码审查和修复的开发场景

Install

npx skills add pony-maggie/code_minions --skill implement-with-tdd -g -y

More Options

Non-standard path

npx skills add https://github.com/pony-maggie/code_minions/tree/main/src/code_minions/builtin/skills/implement-with-tdd -g -y

Use without installing

npx skills use pony-maggie/code_minions@implement-with-tdd

指定 Agent (Claude Code)

npx skills add pony-maggie/code_minions --skill implement-with-tdd -a claude-code -g -y

安装 repo 全部 skill

npx skills add pony-maggie/code_minions --all -g -y

预览 repo 内 skill

npx skills add pony-maggie/code_minions --list

SKILL.md

Frontmatter

{
    "llm": {
        "max_tokens": 16000,
        "temperature": 0.2,
        "max_iterations": 30
    },
    "name": "implement-with-tdd",
    "role": "implementer",
    "inputs": {
        "ticket": {
            "type": "object",
            "required": true
        }
    },
    "outputs": {
        "commit_sha": {
            "type": "string"
        },
        "rounds_used": {
            "type": "integer"
        },
        "test_result": {
            "type": "object"
        },
        "files_changed": {
            "type": "array"
        },
        "review_report": {
            "type": "object"
        }
    },
    "policies": {
        "require_tests": true,
        "reviewer_max_rounds": 2,
        "self_heal_max_rounds": 3
    },
    "description": "TDD implementation with AI code-review loop for a single ticket.",
    "allowed-tools": [],
    "required-mcps": [],
    "invokes-skills": [
        "ai-code-review"
    ],
    "entrypoint-script": "scripts\/run.py"
}

implement-with-tdd

Implement a single ticket using a TDD cycle and iterate with AI code review until converged or max-rounds reached.

Inputs

ticket (object, required): one task from plan-tasks / create-jira-tickets.

Outputs

files_changed (string[])
test_result (object): {passed: boolean, output: string}
commit_sha (string): commit created at end of the loop
review_report (object): final ai-code-review output
rounds_used (integer)

Policies

require_tests: true means a green command is not enough: the runtime must also detect at least one executable test in the workspace before the ticket can be reported as implemented.
reviewer_max_rounds: 2 means the first blocker/major review finding is fed back to the implementer for one repair attempt before the task falls back to human intervention.