paste this link into your ai. it will know what to do.
Maintain progress on multi-step, long-running, or open-ended work. Use for migrations, rollouts, audits, recovery, infrastructure, curation, or any goal with several dependent verification units. Keep explicit remaining state, choose the next causal unit, verify it, and continue until the agreed outcome is complete or genuinely blocked.
Keep going
Maintain one outcome and an explicit work model. Do not ask the user to choose a next step that follows directly from the stated goal and repository authority.
Pairs with the loop skills: if the task already runs under execution-rigor (execution contract) or truth-loop (reasoning contract), defer to their contract; keep-going is the lighter alternative for when that weight is not needed.
Loop
- Restate the outcome and current falsifier in one sentence.
- Split remaining work into the smallest units with an owner, observable done condition, and proof command.
- Choose the unit that removes the largest causal blocker or risk.
- Load only the owning package spec and skills triggered by that unit.
- Implement, run the proof, inspect the artifact, and update remaining state.
- Review the cumulative delta at meaningful milestones.
- Continue while a safe, in-scope unit remains.
Use the plan tool for multi-step state when available. Do not create a markdown status file unless that repository declares it as the canonical work record.
Decision boundary
Continue through reversible implementation choices. Ask the user only when a missing choice changes product direction, canonical ownership, safety, external communication, or irreversible state. Lead with the recommended choice and its consequence.
When blocked:
- reproduce the exact blocker;
- finish independent work around it;
- name the missing authority, access, artifact, or decision;
- provide one concrete unblock action.
Do not replace forward work with repeated status summaries. Do not turn nearby, unrelated debt into the task.
Milestone review
Use adversarial-reviewer on the raw cumulative diff and verification evidence. Use verification-hygiene before any completion claim. Load a domain skill (runtime, storage, infrastructure, or feature-family) only when its boundary is actually touched.
Handoff
Hand off result-first: the answer, up to three proved changes, one next action, and one bounded human decision only if required. Completion means the agreed outcome is observed in the real artifact, not merely that the work list became tidy.
keep it where your ai can reach it.
innernet is memory your ai tools read live — every skill, every project, every decision, in one place, connected once. save this skill to yours, or publish one of your own as a link like this.