On any failure (non-zero exit or error field in JSON) print ⚠️ <message> and stop.

Steps

Context. Run .agents/skills/pflow-skill-add/scripts/skill-add-context.sh → JSON {skills_dir, skills:[{name,description,scripts}]}.
Clarify. You MUST know (a) what the skill does and (b) when it triggers. If ambiguous, ask — never assume. Propose a kebab-case NAME.
Build. Write skills/<NAME>/SKILL.md and optional skills/<NAME>/scripts/ following the template below. Add a row for the new skill to the Skills table in README.md.
Validate. Run .agents/skills/pflow-skill-add/scripts/skill-add-finalize.sh --name "<NAME>" → {status:"ok",skill_dir,scripts_chmod}.

Template

New skills use a 3-phase structure (phases 1 and 3 are optional):

Phase 1 — Preparation (optional): read-only script gathers data → JSON. SKILL.md step: 1. Run <path> → {json}.

Phase 2 — Agent work: agent makes creative/judgment decisions. SKILL.md step: 2. <description of the decision>.

Phase 3 — Finalization (optional): action script takes agent output, mutates state → JSON. SKILL.md step: 3. Run <path> --arg "VALUE" → {json}.