Automate me

Use when the user wants agents to follow their durable working conventions.
For one task-specific workflow, use authoring-a-skill
directly. Resolve an existing mode Skill before drafting: an update preserves
uncontradicted rules and considers history since its last edit. A new Skill
needs the user's chosen name or a simple suggested handle; do not infer a
personal identity from private records.

Inspect only authorized current-workspace history. Divide a sizeable period
into independent slices if available, and collect recurring corrections with
task citations. One isolated remark is weak evidence; contradictions stay
visible. Ask the user briefly about areas history cannot reveal, such as
autonomy, verification, format, and publication boundaries. Do not mine other
projects or personal accounts by default.

Cluster only non-default, actionable preferences. Draft the Skill using
authoring-a-skill. Keep its trigger narrow to
the user or chosen handle; do not make ordinary coding requests activate a
personal mode. Link to existing workflows rather than duplicating them. Use
unslop to cut generic prose. For an update, show which
rules were retained, revised, or removed and why.

Show the draft and its evidence for user correction before treating a subjective
mode as finished. Validate frontmatter, links, and trigger behavior. Put it in
the runtime's chosen personal or project Skill location; do not overwrite
unrelated configuration. Commit or publish only when requested. Return path,
trigger, key rules, weak or disputed findings, and the review status.