skill-forgelisted

# Skill forge Maintains this skill set. Invoked by hand, because writing a skill is deliberate work. ## Before writing Ask whether a new skill is the right answer at all. Three cheaper options usually are: - **A line in `CLAUDE.md`** — for a fact the agent needs on every run - **A section in an existing skill** — for a branch of something already covered - **A wiki page** — for an answer, rather than a process A skill earns its own file when it is a **process with steps and a completion bar** that fires in a recognisable situation. Anything else is reference, and reference belongs in the skill that reaches it. ## What makes these skills work The set has a house style, and it is the style that carries the autonomy: **A named failure mode up front.** Each skill opens by saying what goes wrong without it. That is what makes the steps feel necessary rather than ceremonial, and it is what an agent reaches for when deciding whether the skill applies. **Completion criteria that are checkable and exhaustive.** Every skill ends in a `## Done when` that a run can verify against. "Every gate has exited zero and has an entry in gates.md" bounds the work; "verification is complete" does not, and the difference shows up as premature completion. **Positive targets, not prohibitions.** A ban drags the banned behaviour into context and half-reads as an instruction. Say what to do — `agentic-guardrails`' table is the pattern: the shortcut in the left column, the real fix in the right