my-pi-agent

Skills

Three packaged workflows. Each skill is one SKILL.md under skills/ whose name and description are advertised to the model; the instructions behind them load only when a request matches, so a skill costs a sentence of context until it is used. /skill:<name> forces one, which is also the only way to start a skill that declares disable-model-invocation.

  • init — create or refresh the project’s agent instructions through an interview and a plan you approve.
  • next — suggest what to do next from this conversation, or from the tree when the conversation is empty.
  • refactor — audit for duplication, oversized modules, coupling, and drift, then carry the refactor you pick.

Discovery is read-only in both of the auditing skills: they report and ask, and nothing is edited, committed, or started before you pick an item.

/bytefly lists every packaged skill with a checkbox: switching one off removes it from the model’s roster and from /skill:<name> for the next session.