scribelisted

# Authoring Claude Code Instruction Artifacts ## When this skill applies Load this skill whenever authoring or editing any of: - A `SKILL.md` (new or edit) - A `CLAUDE.md` (managed, user, or project level) - A plan file produced via plan mode - A subagent definition (`~/.claude/agents/*.md`, `.claude/agents/*.md`, or `<plugin>/agents/*.md`) - A slash command (`.claude/commands/*.md`) — note: merged into skills, see `references/slash-commands.md` - A hook script or its documentation - Any markdown whose purpose is to steer Claude Code behavior in a future session This skill does NOT apply when *executing* inside Claude Code (writing code, running builds). It governs only the text of instruction artifacts that Claude Code will consume later. ## Preload manifest — load these references BEFORE writing Per artifact type, `Read` the listed reference files BEFORE drafting. Loading them after the fact catches mistakes only on the proofreader pass — by then you have already encoded the wrong shape across multiple files. The cost of a four-file `Read` upfront is one round-trip; the cost of a re-audit loop is multiple proofreader dispatches plus rewrites. | Authoring … | MUST preload | SHOULD preload when relevant | |---|---|---| | `SKILL.md` | `references/skill-authoring.md`, `references/tools.md` | `references/dynamic-context.md` (if injection used), `references/permissions.md` (if `allowed-tools` used) | | Subagent definition (`agents/*.md`) | `references/subagents.md`, `refer