This is a multi-agent orchestration workflow that breaks research tasks into parallel subtasks, spins up child processes via `codex exec` in sandboxed environments, and assembles their outputs into a polished report file. It enforces a strict execution flow: reconnaissance to scope the problem, decomposition into concrete subtargets, parallel execution with skills or MCP tools for data gathering, programmatic aggregation, then chapter-by-chapter refinement. The workflow prioritizes file artifacts over chat responses and includes detailed logging at every step. Use it when you need systematic competitive analysis, batch link processing, or evidence-driven long-form writing where you want the orchestration logic explicit and auditable rather than handled in a single monolithic prompt.

skills.sh installs
128
GitHub stars
249
last updated
Mar 9, 2026

Install

$ npx -y skills add feiskyer/codex-settings --skill deep-research --agent claude-code

Installs into .claude/skills of the current project.

Files

Deep Research

主控负责界定问题、拆分、核验与综合;子任务负责互不重叠的采集或局部分析。完成标准是用户需要的报告和证据覆盖,不是子进程退出或材料聚合成功。

核心约束

  1. 保持用户当前模型和推理配置,不传 --model,不覆盖无关配置。
  2. 子进程默认使用 workspace-write;只有确实需要 shell 网络访问时才启用 runner 的 --network。
  3. 先检查当前会话可用的 Skills、连接器和 MCP,再按来源适配能力;不要假设固定服务或工具名存在。
  4. 不使用 --dangerously-bypass-approvals-and-sandbox。
  5. 所有运行产物写入独立的 .research/<name>/ 目录。
  6. 只在委派已获授权时启动子任务。执行前简述拆分与成本;超出已有预算、访问范围或权限时再征求同意,不因正常使用已授权的研究来源重复暂停。

按需加载

准备运行子任务时读 references/runner.md,其中包含目录、manifest、命令与恢复约束。两个脚本均只依赖 Python 标准库;--help 给出参数真源。

Workflow

  1. 从用户请求中提取目标、时间范围、来源边界和交付格式,只追问会改变研究结论或执行范围的缺口。用代表性真实样本校准拆分,复用已有材料。
  2. 为每个子任务明确问题、输入、允许访问的范围、输出和证据要求。要求结论旁附来源,无法确认的内容说明缺口。将采集材料保存在运行目录内,避免重复抓取。
  3. 选择与任务成本相称的并发和超时,陌生或高成本批次先小规模验证。主控同步推进非重叠工作,不重复子任务正在做的采集。
  4. 检查 results.json、子报告和关键原始来源。失败、空输出、缺失引用或未覆盖的问题仍是未完成;按 runner 参考中的恢复约束有限重试,不重跑已验证成功的任务。缺少权限或需要扩大范围时停止相应操作并说明阻塞。
  5. 使用 aggregate.py 整理内部材料,再综合为 polished_report.md(用户指定位置优先)。消除重复、核对关键数据、区分来源冲突和证据强弱;aggregated_raw.md 不是成品。

交付报告绝对路径、关键结论和剩余缺口,说明实际使用的权限例外。必要证据仍缺失时,报告已完成范围、具体阻塞及恢复条件,不把部分材料称为完整报告;不要发送内部 Prompt 或调度日志充当成品。

Sponsors

Details

Repository
github.com/feiskyer/codex-settings
Categories
AI & Agent Building, Data Science & ML, CLI & Terminal
License
MIT
First seen
Jun 11, 2026

More from feiskyer/codex-settings

All 7
1proffesor-for-testing avatarquality-metricsproffesor-for-testing/agentic-qeMeasure quality effectively with actionable metrics. Use when establishing quality dashboards, defining KPIs, or evaluating test effectiveness.126skills.sh installs
2khazp avatarvibe-buildkhazp/vibe-coding-prompt-templateBuild your MVP following the AGENTS.md plan. Use when the user wants to start building, implement features, or says "build my MVP", "start coding", or "implemen124skills.sh installs
3vladm3105 avatargoogle-adkvladm3105/aidoc-flow-frameworkDevelop agentic software and multi-agent systems using Google ADK in Python119skills.sh installs
4openai avatarcreate-planopenai/skillsCreate a concise plan. Use when a user explicitly asks for a plan related to a coding task.118skills.sh installs
5danielmiessler avatarAgentsdanielmiessler/personal_ai_infrastructureagents117skills.sh installs
6masondelan avatarselvedgemasondelan/selvedgeUse before editing a tracked entity (a DB column, table, function, API endpoint, dependency, or env var) and after any substantive change. Call prior_attempts f2skills.sh installs