--- name: github-kernel description: foundational definitions for GitHub skills, safety rules, tool escalation, and security boundaries --- ## What is it? The **GitHub Kernel** is the foundational skill that defines the "laws of physics" for all other GitHub capability skills. It provides the central source of truth for: - Tool selection strategy (Escalation Ladder). - Safety boundaries and non-negotiable rules. - Conventions for interactions with the GitHub platform. - Mapping of abstract capabilities to concrete MCP tools. ## Success Criteria - Foundational rules are applied across all other GitHub skills. - The Tool Escalation Ladder is followed consistently. - Safety rules for destructive operations are respected. - Interaction conventions (naming, commits) are maintained. ## When to use this skill - Consult this skill for **foundational knowledge** when initializing a GitHub-related session. - Refer to this skill whenever you are unsure about which tool to use (MCP vs. CLI vs. API). - Consult this skill to check if an action is considered "safe" or requires user confirmation. ## What this skill can do - Explain the [Tool Escalation Ladder](references/TOOL_ESCALATION.md). - Define [Safety Rules](references/SAFETY_RULES.md) for destructive operations. - Map capabilities to specific [MCP Tools](references/MCP_TOOL_MAP.md). - Establish consistent conventions for branch naming, commit messages, and PR titles (see [CONVENTIONAL_COMMITS.md](references/CONVENTIONAL_COMMITS.md)). ## What this skill will NOT do - **Execute any GitHub actions.** This skill is purely informational. - Create issues or PRs. - Modify repository state. - Interact with the API directly. ## How to use this skill This skill is a **passive reference**. 1. **Read** the reference documents to understand the operating parameters. 2. **Apply** the rules when executing other skills like `github-issues` or `github-pr-flow`. ## Tool usage rules All GitHub skills must adhere to the **Escalation Ladder**: 1. **Primary**: Use **GitHub MCP Tools** (fast, safe, context-aware). 2. **Fallback**: Use **`gh` CLI** (if MCP is missing/broken). 3. **Last Resort**: Use **REST API** (only for edge cases like metadata fields). See [TOOL_ESCALATION.md](references/TOOL_ESCALATION.md) for detailed logic. ## Examples ### Determining Tool Safety > "I need to delete a file. Is this safe?" > _Check `SAFETY_RULES.md` -> Destructive operation -> Requires explicit confirmation._ ### Selecting the Right Tool > "I need to comment on an issue." > _Check `MCP_TOOL_MAP.md` -> `add_issue_comment` is available -> Use MCP._ ## Limitations - Does not enforce rules programmatically; agents must voluntarily adhere to these guidelines. - Does not contain specific workflow logic for complex tasks (like "release management").