goallisted

Turn the user's brief into an armed, self-correcting goal loop, then start working toward it. The input may be an unedited speech transcript: expect filler words, self-corrections, topic jumps, and missing structure. Extract intent; never quote the mess back at the user. User's brief: <brief> $ARGUMENTS </brief> ## Phase 0 — Initialize if needed If `.ultragoal/` does not exist in the project root, run the `ultragoal:setup` skill first (it scaffolds directories, asks the preference knobs, and offers the CLAUDE.md block), then continue here. ## Phase 1 — Consult before asking Before asking the user anything: - Read `.ultragoal/memory/MEMORY.md` and any topic files relevant to the brief. Trust `[VERIFIED]` facts; treat `[UNVERIFIED]` ones as hypotheses. - Scan the repo for whatever the brief touches — existing implementations, tests, conventions, prior art. Search directly for contained briefs; spin up parallel Explore subagents only when the surface is genuinely large or unfamiliar. - Check `.ultragoal/goals/archive/` for related past goals (especially their Decision journals and failure notes). **Off-ramp — not everything needs the loop.** If the consult shows a contained, reversible change you could finish and verify with a couple of commands — no real forks, no long unattended run — say so in one line and do the work directly in this conversation instead of arming a goal: same evidence discipline, no spec, no interview, no ceremony. The machinery exists for work a p