| name | ship-tasks |
| description | Run the CyberOS ship-tasks workflow in any repo: drive a task (product or improvement class) through implement -> review -> test -> done, with HITL required at the two human-acceptance gates and gates from the repo's own build/lint/test. Use when asked to ship, implement, or harden a task (backlog class:improvement), or to drive a docs/tasks backlog. Do NOT use for full-repo inspection remediation — that is /inspect then /harden on an inspection-report@1. |
ship-tasks (portable)
The full, normative workflow is in cuo/ship-tasks.md (bundled beside this file), with cuo/EXECUTION-DISCIPLINE.md (the halt and HITL doctrine) and cuo/STATUS-REFERENCE.md (the status lifecycle). Read those; they are the source of truth.
Not /harden: "Harden a task" here means shipping a backlog item with class: improvement. Full-repo inspection remediation (consume inspection-report@1, emit hardening-record@1) is the /inspect → /harden skill chain — distinct, never merged into this conductor.
Resume on restart (ship-manifest@1, TASK-CUO-206): before starting step 1, check for docs/tasks/.workflow/<task-ID>.ship.json (gitignored run state). If present with a matching workflow_version, re-hash its recorded artefacts and resume at the first unproven step per the workflow doc's Resume semantics; a stale or missing manifest just means a fresh run. Write the manifest back after every step, and always re-ask the HITL gates - the manifest never carries an approval.
In one paragraph: pick the first eligible task in docs/tasks/BACKLOG.md (ready_to_implement, dependencies done). Deep-map the repo, write the edge-case matrix, implement with observability and at least 90% coverage on touched files, review the diff against every section-1 clause, and run the gates (.cyberos/cuo/gates/run-gates.sh = the repo's own build/lint/test + coverage; caf and awh only if present). HITL is required: halt at review acceptance and at final acceptance for a recorded human verdict, and never set done yourself. On any gate failure, route the task back to ready_to_implement. Improvement and hardening work is the same workflow with class: improvement. For inspect-report remediation use /inspect and /harden instead.