Assistant Support¶
.repo.yaml names the assistants a repository configures: claude, copilot, both, or none.
Both receive the same rules; the rest differs with what each assistant supports.
| Content | Claude Code | GitHub Copilot |
|---|---|---|
| Entry file, written once for the developer | CLAUDE.md |
— |
| Core rule, loaded in every session | .claude/rules/core.md |
.github/copilot-instructions.md |
| Rules per file type, extracted from the guidelines | .claude/rules/*.md with paths: |
.github/instructions/*.instructions.md with applyTo: |
| Reusable workflows: commit, review, TDD, debugging, ... | .claude/skills/, .claude/commands/ |
.github/prompts/*.prompt.md |
| Reviewers in an isolated context | .claude/agents/ |
.github/chatmodes/, in the current context |
| Command approval | .claude/settings.json |
Per user in VS Code |
| Session start reminder | .claude/hooks/session-start.sh |
— |
| MCP servers | .mcp.json |
.vscode/mcp.json |
Which rules a profile gets and which files they load for is listed with the guidelines; capability 004 sets a word budget for every file an assistant loads.
Copilot's Limits¶
- No isolated reviewers: a chat mode runs in the current conversation, so it knows how the code came about; a Claude Code agent judges the code without that context. The review prompts say so.
- No checked-in command approval: what Copilot may run without asking is configured per user.
- No session hook: nothing reminds Copilot of the skills at the start of a session; the core
rule in
.github/copilot-instructions.mdhas to carry the workflow on its own. - Prompts are translations: the prompt files are generated from the Claude skills and commands; Claude-specific steps, such as plan mode, are dropped.