writing-skillslisted

# Writing better-dev skills Every better-dev practice ships as a `SKILL.md` on the agentskills.io standard, so one skill runs unchanged across Claude Code, Codex, pi, and hermes. This is the bar to author against. ## Frontmatter - minimal on purpose Required - two keys, nothing else load-bearing: - `name` - kebab-case, ≤ 64 chars, matches the folder name. - `description` - the *only* text an agent reads before deciding whether to load the body. Write it as **triggering conditions**, third person, starting "Use when …". Do **not** summarize the workflow here: a description that reads like a summary tempts the agent to act on it and skip the body. Each trigger names a **distinct** situation; a synonym that only renames a branch already listed is duplication - collapse it into the branch it renames. Optional - add only when a skill actually needs it: - `disable-model-invocation: true` - makes the skill **user-invoked** (`/name` only), with zero standing context cost. Use it *only* when a skill should fire at a human's deliberate command: destructive or expensive actions, or a reference a person pulls up by hand. Do **not** use it for disciplines the *agent* must apply on its own - like this authoring standard, which the self-extension flow needs when it writes a skill. Those stay **model-invoked** so the agent can reach them. Omit the key to auto-fire. - `argument-hint` - e.g. `[feature-slug]`, shown at the `/name` prompt. - `allowed-tools` - grant exactly