--- name: graphhelm-resume description: Use when the user asks what to do next, wants to resume GraphHelm work, or invokes graphhelm-resume to choose between concrete next actions. --- # GraphHelm resume Give the owner **exactly two** distinct, concrete next actions, with **one marked recommended**. This skill is read-only: selecting an option does not execute it, alter a graph, install anything, or change host settings. 1. Establish the latest goal and state from the conversation and the smallest relevant live evidence (through the GraphHelm MCP tools when connected: `briefing`, `status`, `events`, `evidence`; otherwise the `graphhelm` CLI): task/issue, branch and PR head, execution status, failed proof, or release status. Check current sources when available; mark stale or missing evidence explicitly. A historical green result does not prove the current head. 2. Identify the nearest unresolved user promise. Prefer an action that removes its blocker or obtains the missing observer. Choose a second viable action with a real tradeoff, such as preparing a specific proof checklist or inspecting a named work record while a blocker remains. Ground both options in the supplied goal and evidence. If a source is inaccessible, the second action must still be possible with available information or by asking the owner for the missing record; do not propose inspecting that inaccessible source as if it were available. Never write a generic "work on another task" option unless that task is named. Do not invent a task, cost saving, pass, merge, or deployment. 3. State status in one short sentence, then present exactly two numbered options. Each option names the action and its expected result. Mark one `Recommended` and give a brief reason. Keep the owner's language and use small words. If there is not enough evidence to name two implementation tasks, make the options two **discovery** actions: one to inspect the current work record, the other to ask the owner for a goal. Never fill the gap with guesses. If work is already complete, say so before proposing two genuinely new next actions. Do not add a third option or silently proceed with either action.