Make Skill

Create one new workspace Skill from the current conversation through planning, user approval, draft authoring, validation, and publication.

Resolve <workspace> from the runtime directory context: use the current agent's absolute workspace path (also the working directory when no separate project is configured). Pass that same value throughout the lifecycle, independently of the task's project directory and the script cwd. Lifecycle artifacts belong under <workspace>/.qwenpaw/make-skill/; published Skills belong under <workspace>/skills/.

Script interface

Run python scripts/<script> through execute_shell_command, setting cwd to this Skill's <dir> from the available-skills entry. Each script reads one JSON object from stdin (or --input <file>) and returns one JSON object. Every input includes workspace; the table lists the other top-level fields.

Operation	Script	Other input fields	Successful result
Create a plan	`create_plan.py`	`plan`	`plan_id`, normalized `plan`
Revise that plan	`create_plan.py`	`plan_id`, complete new `plan`	Same `plan_id`, normalized `plan`
Initialize after approval	`init_draft.py`	`plan_id`	`draft_id`, `skill_dir`
Validate a draft	`validate_skill.py`	`draft_id`	`digest`
Publish a validated draft	`publish_skill.py`	`draft_id`, `expected_digest` from validation	Publication result

plan_id identifies one editable plan; keep it across revisions, including renames. draft_id identifies the initialized draft for validation, testing, and publication; skill_dir is where its package files belong. Use the returned values unchanged, not IDs inferred from names or paths.