prompt-authoring-workflow
prompt-author
Shorthand: prompt-author. Turn a developer's stated intent into a useful prompt for working with agents, as the planner Technician: a Builder Garage workflow invocation, a question, an exploration request, or a reusable template with marked placeholders. Name the action, location, and canonical skill for an invocation, add only context the durable records lack, and report missing requirements as gaps that belong in the issue. Display the prompt by default; save it only when explicitly asked, to the Builder's path or .garage/prompts/, never overwriting without authorization and never committing. A prompt never declares that work is ready to begin.
- Technician role
- planner
Technical details
- Source revision
- v0.4.0 · f078359
- Files
- SKILL.md
Download the ZIP below, preserve the prompt-authoring-workflow/ directory, and place it in a skill directory supported by your harness.
Hand-placed skills are unmanaged. Canonical skills resolve project choices through garage config validate, so configuration-dependent steps still need a valid project configuration and garage on PATH. Directly requested discovery does not require configuration for its discovery-applicability decision.
From a project with valid configuration and declared skill targets, install and verify the complete embedded canonical skill set, together with the manuals the project adopts:
garage config validate garage skills install garage skills verify
The CLI supports user-scope targets for Claude Code, Codex, and Copilot CLI, not individual-skill or project-scope installs. Harness path behavior was confirmed on macOS; Linux and Windows harness behavior remains unconfirmed. Existing unmanaged or modified copies require explicit replacement consent and a backup. See skill installation.
