keepworkinglisted

# keepworking Use this skill to keep non-trivial work moving until there is evidence. ## Core Goal Move from request to verified closure: ```text goal -> plan -> execute -> verify -> repair -> re-verify -> close ``` ## Routing Before doing work, classify the next useful step. | Tier | Use for | Default behavior | | --- | --- | --- | | `simple` | search, classification, summaries, status checks | read-only, low-risk discovery | | `medium` | bounded implementation, repair, deterministic checks | scoped edits and verification | | `complex` | architecture, deep debugging, multi-stage workflows | staged checkpoints and explicit risks | Escalate when the current tier can no longer close the task safely: - `simple -> medium` when edits or deterministic checks are needed - `medium -> complex` when the cause is still unclear after one verification cycle - `complex -> simple` when cheap discovery can reduce ambiguity first ## Workflow 1. Restate the current goal in one line. 2. Separate discoverable facts from preferences. 3. Pick the smallest useful next slice. 4. Execute or inspect that slice. 5. Verify with evidence. 6. If verification fails, repair and re-run the same or stronger check. 7. Close only when the evidence and unresolved risks are explicit. ## Evidence Prefer evidence that another person or tool can inspect: - file paths - command output - test or build output - logs - browser or UI checks - screenshots - structured manifests - audit events Do not treat a