Use this skill when exploring an unfamiliar project or subsystem, onboarding to a new area, or preparing for significant or cross-module changes. Do not trigger it automatically for every new session or for simple localized tasks.

Goal: Establish a concise, graph-backed context for the current task without scanning the entire repository.

Shared tool policy:

Follow ../../reference/codegraph-tool-policy.md for common CodeGraph tool selection, fallback and unavailable handling.

Task-specific tools:

Prefer codegraph_context, codegraph_search, codegraph_files and codegraph_explore for unfamiliar areas.

Workflow:

Call codegraph_status (or codegraph status .) to check if a graph index exists and is fresh. If missing or stale, request the user run codegraph init -i or codegraph sync.
Use codegraph_context to build a focused context for the current task (for example, "refactor auth module"), limiting the number of nodes to a manageable size (e.g. 20).
Supplement with codegraph_search for specific keywords, classes, functions or services relevant to the task. Disambiguate by module, file path or symbol kind when multiple matches appear.
For broad unfamiliar areas, use codegraph_explore after search has identified concrete symbols, filenames or short code terms. Avoid natural-language-only explore queries.
Use codegraph_files to understand the repository structure and identify important directories and modules.
Summarise the context: key files, modules, classes, functions, entry points, external integrations, and potential risks. Use lists or tables for clarity.
For tasks that trigger this skill, build context before broad file reading or editing.

Rules:

Do not read large numbers of files without first using CodeGraph to narrow down the scope.
Keep the main session lightweight: prefer codegraph_context for focused work and reserve codegraph_explore for genuine surveys.
Clearly state when the context is incomplete or when CodeGraph analysis is unavailable.

Más de este repositorio

mismo repositorio

codex-scheduler-watcher

MC-and-his-Agents/MC-SKILLS

创建或更新 Codex meta-scheduler watcher automation。用于为一个项目、milestone、parent issue、issue set、pull request、PR set、release、repo carrier 或手工目标维护 coordination unit graph、scheduler pool、candidate graph/pool、shared lane lock table、scheduler waiting queue 和调度线程生命周期；在安全并行时创建多个 scheduler thread，并在 shared lane 争用时执行 lane grant / wait / release / recovery。默认中文说明和回报，可适配用户语言；协议字段、状态枚举、工具名和结构化模板字段保持机器可读。

2026-06-156

codex-thread-orchestration

MC-and-his-Agents/MC-SKILLS

Codex 线程编排的 scheduler 与 worker 协作协议。用于 Codex 作为 scheduler 拆分全局目标、创建或恢复 worker、管理 branch/worktree、发送跨线程指令、运行 gate、完成 closeout；也用于 Codex 作为 worker 确认分配现场、创建并自检 scoped goal、只执行指定 scope、回报 scheduler-readable 关键节点，并等待 scheduler-owned gate。默认使用中文说明和回报，可根据用户语言适配；协议字段、状态枚举、工具名和结构化模板字段保持机器可读。

2026-06-156

write-a-goal

MC-and-his-Agents/MC-SKILLS

起草、优化或设置符合 OpenAI《Follow a goal》指南的 Codex goal。起草前先根据最新用户请求确定输出语言。为用户起草可复制的单行 /goal 命令；用户明确要求时，通过 goal 工具/API 创建自包含的 active goal objective。用户使用中文、混合语言或语言不明确时，输出中文。保留命令、路径、代码标识符、URL、issue ID、精确引用，以及 /goal、API、objective、copyable_goal_command、active_goal_api 等标识原文。

2026-06-046

architecture-map

MC-and-his-Agents/MC-SKILLS

Map the architecture of a repository using CodeGraph.

2026-05-226

bug-localizer

MC-and-his-Agents/MC-SKILLS

Narrow down likely code locations from bug descriptions using CodeGraph.

2026-05-226

change-plan-with-graph

MC-and-his-Agents/MC-SKILLS

Generate a graph-backed implementation plan before modifying repository code.

2026-05-226

name	codegraph-context-bootstrap
description	Build repository context using CodeGraph before starting tasks.

Goal: Establish a concise, graph-backed context for the current task without scanning the entire repository.

Shared tool policy:

Follow ../../reference/codegraph-tool-policy.md for common CodeGraph tool selection, fallback and unavailable handling.

Task-specific tools:

Prefer codegraph_context, codegraph_search, codegraph_files and codegraph_explore for unfamiliar areas.

Workflow:

Call codegraph_status (or codegraph status .) to check if a graph index exists and is fresh. If missing or stale, request the user run codegraph init -i or codegraph sync.
Use codegraph_context to build a focused context for the current task (for example, "refactor auth module"), limiting the number of nodes to a manageable size (e.g. 20).
Supplement with codegraph_search for specific keywords, classes, functions or services relevant to the task. Disambiguate by module, file path or symbol kind when multiple matches appear.
For broad unfamiliar areas, use codegraph_explore after search has identified concrete symbols, filenames or short code terms. Avoid natural-language-only explore queries.
Use codegraph_files to understand the repository structure and identify important directories and modules.
Summarise the context: key files, modules, classes, functions, entry points, external integrations, and potential risks. Use lists or tables for clarity.
For tasks that trigger this skill, build context before broad file reading or editing.

Rules:

Do not read large numbers of files without first using CodeGraph to narrow down the scope.
Keep the main session lightweight: prefer codegraph_context for focused work and reserve codegraph_explore for genuine surveys.
Clearly state when the context is incomplete or when CodeGraph analysis is unavailable.