feiskyer/codex-settings / skills/deep-research
deep-research
Verified仅在用户显式调用 `$deep-research` 时,用 Codex CLI 子任务完成多来源深度调研并综合成带出处的报告。不用于单次查证或已给定材料的总结。
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 2.7 KB |
| agents/openai.yaml | config | 357 B |
| references/runner.md | markdown | 2.8 KB |
| scripts/aggregate.pyexecutable | script | 3.3 KB |
| scripts/run_children.pyexecutable | script | 10.3 KB |
| tests/test_deep_research_scripts.py | script | 5.3 KB |
references/runner.md
View on GitHub# Codex 子任务运行与恢复
准备执行子任务时读取。`<skill-dir>` 是包含 `SKILL.md` 的绝对目录,不是当前工作目录。
## 运行目录与 manifest
选择不重复的 `.research/<name>/`,保存 `manifest.json`、`prompts/`、`logs/`、`child_outputs/`;只在需要缓存来源时创建 `raw/` 或 `cache/`。
```json
{
"tasks": [
{
"id": "market-history",
"title": "市场历史",
"prompt_file": "prompts/market-history.md"
},
{
"id": "current-competitors",
"title": "当前竞品",
"prompt_file": "prompts/current-competitors.md"
}
]
}
```
`id` 只能包含字母、数字、点、下划线和短横线,且不可重复;Prompt 路径必须位于运行目录内。每份 Prompt 写清子目标、输入、授权范围、输出要求及证据缺口的报告方式。
## 预检与执行
不熟悉的批次、自动生成的 Prompt 或高成本任务先预检;`--dry-run` 校验本地输入并打印命令,不启动 Codex:
```bash
python3 "<skill-dir>/scripts/run_children.py" \
--run-dir ".research/<name>" --workspace "$PWD" --dry-run
```
下列并发、超时只是示例,按已授权任务和预算调整;不必为了填满并发数拆出无用任务:
```bash
python3 "<skill-dir>/scripts/run_children.py" \
--run-dir ".research/<name>" --workspace "$PWD" \
--parallel 2 --timeout 600 --retries 0
```
Runner 保持当前模型配置,运行 `codex exec`,使用 `workspace-write` 沙箱,不绕过审批。`--network` 仅控制子进程的 shell 网络访问,确有需要且已授权时才添加;它不授予访问任意外部系统的权限。
输出包括 `child_outputs/<id>.md`、`logs/<id>.log` 和 `results.json`。退出码、结果状态和文件非空都不能代替对内容及引用的核验。
## 有限重试与恢复
Runner 的 `--retries` 会重试进程失败或空输出,但不会识别权限拒绝与其他失败的区别。默认保持零次自动重试;确认失败可重试且没有重复副作用后,再选择有限次数。
`--resume-existing` 只检查输出是否非空,不检查它是否来自失败进程或内容是否合格。先读旧 `results.json` 和现有输出,备份诊断材料;需要重跑的任务使用只包含这些任务的重试 manifest,通过 `--manifest` 传入,不带 `--resume-existing`。Runner 会覆盖同名任务的输出、日志及 `results.json`,不要把部分重跑的状态误当成整批状态。
保留原始完整 manifest 用于最终核对和聚合:
```bash
python3 "<skill-dir>/scripts/aggregate.py" \
--run-dir ".research/<name>"
```
聚合器按 manifest 顺序生成 `aggregated_raw.md`,只检查子报告存在且非空,不读取 `results.json`,也不验证引用。主控须先确认所有必需子任务的实际结果;聚合成功不代表研究已完成。
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 471k | Updated | Claude CodeCodex | Unverified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |