Skip to main content

Providers

Workloom renders canonical package assets into provider-specific files. Each AI coding tool reads configuration from different locations in different formats.

Supported providers​

ProviderStatusOutput locations
Claude CodeSupported.claude/, CLAUDE.md, .claude/settings.json
CodexSupportedAGENTS.md, .codex/config.toml
GitHub CopilotPlanned.github/copilot-instructions.md, .github/instructions/
GeminiPlannedExtensions, GEMINI.md

Claude Code​

Claude Code reads project configuration from .claude/ and CLAUDE.md. Workloom renders:

Asset typeClaude output
InstructionCLAUDE.md, .claude/ files
Skill.claude/skills/<name>/SKILL.md
Agent.claude/agents/<name>.md
MCP server.claude/settings.json mcpServers section
Hook.claude/settings.json hooks section
Provider config.claude/settings.json

Multiple packages contributing to settings.json are merged deterministically.

Codex​

Codex reads AGENTS.md and .codex/config.toml. It supports layered discovery — AGENTS.md files in nested directories take precedence over root.

Asset typeCodex output
InstructionAGENTS.md
SkillAGENTS.md section
AgentAGENTS.md persona section
MCP server.codex/config.toml
Provider config.codex/config.toml

For monorepos, Workloom renders separate AGENTS.md files per target path, matching Codex's layered discovery model.

GitHub Copilot (planned)​

Copilot requires committed repo files for some features. Workloom will use the committed_projection policy for:

  • .github/copilot-instructions.md — repo-wide instructions
  • .github/instructions/*.instructions.md — path-specific instructions

Gemini (planned)​

Gemini CLI uses an extension model. Workloom will generate Gemini-compatible extension configs or install/link through a Gemini adapter.

Provider capability matrix​

Not all assets can be rendered to all providers. The renderer checks provider compatibility before generating output. Assets with no compatible provider for a given target are skipped with a warning.