Skills
CodeOps ships 11 skills. Each is a multi-step protocol Claude runs on request — they auto-trigger from natural language and can also be invoked explicitly as /codeops:<name>.
| Skill | What it does |
|---|---|
| make_plan | Clarifying interview → Zero-Ambiguity Gate → plans/<feature>/ document set |
| exec_plan | Implements a plan task-by-task, verifying and committing per mode |
| make_requirements | Requirements discovery / add one RD / health-check |
| retro_requirements | Reverse-engineer a codebase into a reconstruction brief |
| grill_me | Relentless design-disambiguation interview |
| preflight | Adversarial, codebase-grounded quality audit |
| techdocs | Create/maintain VitePress architecture docs + ADRs |
| roadmap | Track a whole feature-set across its lifecycle |
| upgrade_plan | Bring an outdated plan/requirements set to current standards |
| setup_routing | Wire per-project model & effort routing (Opus/Sonnet by task tag) into CLAUDE.md + .claude/agents/ |
| setup_codeops | Scaffold or auto-migrate a repo into the nested codeops/ layout (one confirmation, git mv) |
See also the Commands page for the 19 slash commands.
How they compose
grill_me → make_requirements → preflight → make_plan → preflight → exec_planwith roadmap tracking the whole feature-set and techdocs keeping architecture docs current. The tutorials walk these pipelines end to end.
Skill vs. alias commands
The consolidated skills cover several verbs each, and thin alias commands make each verb directly typeable (they delegate to the parent skill in the right mode) — e.g. /add_requirement, /review_requirements → make_requirements. Aliases are manual-only; only the parent skills auto-trigger from natural language. See Commands.