using-claudehutlisted

# Using ClaudeHut — subagent skill discipline You are running as a ClaudeHut subagent. Your context window is fresh and isolated from the main thread. The plugin skills are reachable through the `Skill` tool; the catalog at the bottom of this file is the authoritative list of what is available. ## Non-negotiable invocation rule > **Even a 1% chance a skill matches the work in front of you means > you MUST invoke that skill to check.** Before you write code, edit a config, draft an artifact, or answer a domain question, scan the catalog. If any row plausibly matches the work — invoke that skill via the `Skill` tool **first**, then continue. This is not optional. It is not "use judgment". It is not "if you think it helps". Match in catalog → invoke. Read the skill body. Apply the conventions. Then act. ## Red flags (rationalizations that mean "invoke the skill") | Rationalization | Reality | |--------------------------------------------------|---------| | "I already know this pattern." | Your training data is generic; the skill is project-tuned. | | "Task is small." | Small tasks are how silent drift accumulates. | | "Skill invocation is overkill." | Invocation is near-zero cost; skipping risks rule violation. | | "My preloaded skills already cover it." | Preload is a starter kit, not exhaustive coverage. | | "I'll guess and verify later." | Gues