Multi-Agent Setup

buildlog renders rules to multiple AI coding agent formats. The rules are natural language — only the file format and path change between agents.

Supported targets

Agent Target flag Output path
Claude Code claude_md CLAUDE.md
Claude Code settings_json .claude/settings.json
Claude Code skill .claude/skills/*.md
Cursor cursor .cursor/rules/buildlog-rules.mdc
GitHub Copilot copilot .github/copilot-instructions.md
Windsurf windsurf .windsurf/rules/buildlog-rules.md
Continue.dev continue .continue/rules/buildlog-rules.md

Promoting rules to an agent

# Promote to Claude Code's CLAUDE.md
buildlog promote <skill-ids> --target claude_md

# Promote to Cursor
buildlog promote <skill-ids> --target cursor

# Promote to multiple targets at once
buildlog promote <skill-ids> --target claude_md --target cursor

Format differences

Agent Format
Claude Code (CLAUDE.md) Plain Markdown, appended to file
Claude Code (settings.json) JSON configuration
Claude Code (skill) Markdown with SKILL.md convention
Cursor YAML frontmatter + Markdown (.mdc)
GitHub Copilot Plain Markdown, appended to instructions file
Windsurf Plain Markdown
Continue.dev YAML frontmatter + Markdown

Adding a new agent

The renderer registry pattern makes this straightforward. Each renderer is a function that takes a list of rules and writes them to the appropriate file format and path. See the Contributing guide for details.