skill-forgelisted

# Skill Forge ## Outcome Design one narrow capability as a portable Agent Skills package with explicit activation boundaries, then compile and prove it when a Foundry runtime is available. ## Use this skill when - A successful workflow should become reusable. - A user asks to create, improve, or package a skill. - A repeated task needs stable instructions, tools, guardrails, or evaluation. - The capability should work in ChatGPT, Codex, or a skills-based plugin. Do not use it when the work requires persistent decision rights, a distinct durable memory scope, or an ongoing trigger owned by an autonomous actor. Use `$agent-forge` only after proving one of those boundaries. ## Procedure 1. Define the decision contract. - Capture the objective, audience, deliverables, stakes, reversibility, permissions, constraints, and completion tests. - Ask only for missing information that materially changes the capability. 2. Discover before creating. - Inspect the capability graph and existing skills. - Reuse or depend on an existing capability when it already owns the behavior. - Record required, preferred, and forbidden capabilities explicitly. 3. Bound activation. - Write concrete positive triggers and at least one negative trigger. - Keep one skill responsible for one coherent outcome. - Avoid broad adjectives as requirements; translate them into observable gates. 4. Author the Skill Pack. - Give every procedure step a proof obligation. - Declare inp