Configuration

Sherpa has project configuration, project state, and host integration files. These are related but separate.

Project files

Configuration examples are under templates/configuration/. The configuration workflow supports /sherpa config for a redacted overview and /sherpa config check for validation and secret-protection checks. The implementation is in core/configuration/manager.py, src/config_manager.py, and workflows/configuration.md.

Secrets

Before you pack credentials, check which values are project facts and which are secrets. Start from .env.sherpa.example or the relevant example under templates/configuration/. Put real credentials in ignored environment files and never commit tokens or keys. The configuration check reviews Git protections and scanned secret indicators; it does not make committing secrets safe.

Agent configuration

The repository has checked-in adapter files for Claude Code, OpenCode, and Kilo Code. They are not generated from one shared configuration schema. The global installer targets Claude Code and OpenCode only. No central per-agent JSON configuration is documented.

Installer environment variables

SHERPA_RELEASE_MANIFEST_URL overrides the release manifest URL. Installers also expose SHERPA_VERSION, SHERPA_INSTALL_CLAUDE, SHERPA_INSTALL_OPENCODE, SHERPA_INSTALL_CLI, and SHERPA_NO_PATH. Former sherpa_* prefixed variables are accepted as fallbacks for one release.

The manifest format does not currently resolve arbitrary historical versions, so version pinning is unsupported. For exact switches, see installation and configuration source reference.

Precedence and custom paths

The repository documents no layered global/project/agent configuration precedence system. Installer targets are fixed by the scripts; a project-local installer switch is not implemented. Do not infer custom paths or configuration overrides beyond the documented installer environment variables.