CLI interface
September 11, 2026 ยท View on GitHub
The npm and PyPI distributions expose the same red command and JSON semantics.
| Command | Effect |
|---|---|
version | Report release and protocol versions |
init | Create red.toml; fail if it already exists |
new research | Create the next Research artifact |
new evolve | Create the next Evolve artifact, optionally with --from references |
promote R-N --to evolve | Create an Evolve artifact referencing the Research source |
promote E-N --to document --accepted --verified | Validate the caller's explicit assertions and supplied Document paths, then record acceptance/synchronization |
status | Count mapped Document, Evolve, and Research files |
check | Validate configuration, artifact identities, duplicates, and references |
skill install/status/update/uninstall | Manage the release-matched RED Skill at repository or user scope |
instructions install/uninstall | Manage only the marked RED block in an instructions file |
instructions export | Create a standalone RED.md |
Commands that return data accept --json. JSON objects conform to output-schema.json. Paths in output use / separators even on Windows.
skill update and skill uninstall require .red-install.json with distribution: "red"; they refuse to alter an unmanaged directory. instructions uninstall similarly removes only the marked block.
One-shot execution
npx -y @exoticknight/red@latest <command>
pipx run --spec red-methodology red <command>
Agents should prefer an existing project-local or user-installed red. A Skill may try one approved, network-backed one-shot runner when no CLI exists, then continue manually if unavailable. It must not add dependencies or install globally without authorization.
The CLI performs deterministic file operations and structural checks. It does not decide that Research may enter Evolve or that Evolve may enter Document. An agent invokes promote only after a human or a project-authorized decision source supplies that transition. The --accepted and --verified flags record caller assertions; an agent must not set them from its own assessment alone.