skill-system-gate

技能系统门禁

在注册实验前实施默认拒绝的门禁，并提供只读的注册表安全性／状态检查。

决策矩阵

任务意图	操作	主要后端	适用场景
根据可配置的门禁规则验证仓库产物	validate	scripts/gate_validate.py	在做出关闭门禁的决定或执行确定性策略检查前
根据门禁规则验证实验实现和日志	validate-experiment	preprocess_lib.gate_engine	检查旧版 GNN/FraudDetect 流水线
检查注册表状态是否适合重新运行／重置	verify-registry	db_registry.DBExperimentsDB	执行任何手动重置／重新运行操作前
查看队列健康状况和状态计数	status	db_registry.DBExperimentsDB	需要快速查看当前实验状态概览时

核心模式

模式 1：验证仓库是否符合门禁要求

python3 scripts/gate_validate.py --rules config/gate_rules.yaml --artifact-root .

• 加载可配置的 gate_rules.yaml。
• 支持规则类型：command、structural、eda-contract。
• 输出 Markdown 判定结果，并在最后一行输出 JSON 摘要。

模式 2：验证旧版实验是否符合门禁要求

scripts/validate_exp.sh <exp_name> [phase_root]

• 加载 gate_bank.json 规则。
• 运行所有规则类别（source_contains、source_not_contains、stderr_scan、file_exists、file_min_size）。
• 输出 Markdown 判定结果，并在最后一行输出 JSON 摘要。

模式 3：在决定重置前验证注册表安全性

scripts/check_registry.sh <exp_name> [phase_root]

• 读取实验的注册表条目。
• 对状态为 RUNNING 的实验禁止给出重置建议。
• 返回状态详情和 safe_to_reset 决策 JSON。

模式 4：显示当前注册表健康状况概览

scripts/status.sh [phase_root]

• 汇总活动集／已完成集中的计数。
• 显示正在运行的实验及其执行位置。
• 在最后一行返回简洁的 JSON 计数器。

指南

• 始终将验证视为默认拒绝：任何错误级别的规则违规都会导致门禁失败。
• 在执行面向注册表的操作前，始终先运行 validate-experiment。
• 检查期间，切勿修改实验源文件／配置文件。
• 切勿通过此技能执行重置／重新运行；只能报告建议。
• 切勿将状态为 RUNNING 的实验判定为可安全重置。
• 对注册表和实验产物的所有操作都必须保持只读。

{
  "schema_version": "2.0",
  "id": "skill-system-gate",
  "version": "1.0.0",
  "capabilities": ["configurable-gate", "experiment-gate", "registry-safety", "experiment-status"],
  "effects": ["fs.read", "db.read", "proc.exec"],
  "operations": {
    "validate-experiment": {
      "description": "Validate experiment source artifacts and logs against gate_bank rules and return a pass/fail verdict.",
      "input": {
        "exp_name": { "type": "string", "required": true, "description": "Experiment directory name under experiments/" },
        "phase_root": { "type": "string", "required": false, "description": "Phase directory path", "default": "/datas/store162/arthur0824hao/Study/GNN/FraudDetect/SubProject/Phase3" }
      },
      "output": {
        "description": "Gate validation markdown plus machine-readable verdict.",
        "fields": { "passed": "boolean", "errors": "number", "warnings": "number" }
      },
      "entrypoints": {
        "unix": ["bash", "scripts/validate_exp.sh", "{exp_name}", "{phase_root}"]
      }
    },
    "validate": {
      "description": "Run configurable gate_rules.yaml checks and emit pass/fail with per-rule results.",
      "input": {
        "rules": { "type": "string", "required": false, "description": "Path to gate rules file", "default": "config/gate_rules.yaml" },
        "artifact_root": { "type": "string", "required": false, "description": "Working directory for rule execution", "default": "." }
      },
      "output": {
        "description": "Gate verdict with per-rule details.",
        "fields": { "passed": "boolean", "errors": "number", "warnings": "number" }
      },
      "entrypoints": {
        "unix": ["python3", "scripts/gate_validate.py", "--rules", "{rules}", "--artifact-root", "{artifact_root}"]
      }
    },
    "verify-registry": {
      "description": "Check registry state safety and recommend whether reset/rerun is safe without mutating state.",
      "input": {
        "exp_name": { "type": "string", "required": true, "description": "Experiment name in registry" },
        "phase_root": { "type": "string", "required": false, "description": "Phase directory path", "default": "/datas/store162/arthur0824hao/Study/GNN/FraudDetect/SubProject/Phase3" }
      },
      "output": {
        "description": "Registry safety markdown plus recommendation payload.",
        "fields": { "found": "boolean", "status": "string", "safe_to_reset": "boolean" }
      },
      "entrypoints": {
        "unix": ["bash", "scripts/check_registry.sh", "{exp_name}", "{phase_root}"]
      }
    },
    "status": {
      "description": "Show current registry dashboard with counts by status and running experiment details.",
      "input": {
        "phase_root": { "type": "string", "required": false, "description": "Phase directory path", "default": "/datas/store162/arthur0824hao/Study/GNN/FraudDetect/SubProject/Phase3" }
      },
      "output": {
        "description": "Registry status dashboard and aggregate counters.",
        "fields": { "running": "number", "needs_rerun": "number", "completed": "number", "total": "number" }
      },
      "entrypoints": {
        "unix": ["bash", "scripts/status.sh", "{phase_root}"]
      }
    }
  },
  "stdout_contract": {
    "last_line_json": true
  }
}