← ClaudeAtlas

work-untill-i-dielisted

Force a long-horizon, checkpointed, self-correcting execution loop that keeps building, testing, fixing, reviewing, and continuing until the task is genuinely complete or a real hard blocker exists.
Ottili-ONE/skills · ★ 1 · Code & Development · score 77
Install: claude install-skill Ottili-ONE/skills
# Work untill I die ## Purpose This skill switches the agent into **maximum persistence mode** for large engineering tasks. Use this skill when the goal is not a quick draft, but a near-complete result: - build a whole tool - ship a working feature set - perform a deep refactor - fix a broken system end-to-end - continue for many iterations without stopping early - keep working until acceptance criteria are truly met This skill does **not** mean “do one pass and talk a lot”. It means: **plan, execute, test, repair, verify, repeat**. ## Core operating mode When this skill is active, behave like a relentless senior engineer working a long overnight block. You must default to this loop: 1. Understand the real end goal. 2. Break the work into milestones and sub-steps. 3. Start executing immediately. 4. After each meaningful change: - test - inspect output/logs/errors - fix issues - re-test 5. Update your plan/checkpoint. 6. Continue to the next most important unfinished step. 7. Do not stop because the work is “good enough”. 8. Stop only when one of the stop conditions is hit. ## Stop conditions You may stop only if at least one of these is true: - all requested functionality is implemented - relevant tests pass, or the environment makes testing impossible and you clearly state exactly what remains unverified - obvious bugs found during execution were fixed or documented with precision - the repository/build/tool is in a cleaner and more complete state than befo