Using Op

Using op

Personal operator layer for solo project work. Commands: /op:go (front door), /op:spec → /op:plan → /op:build (the gate for features), /op:ship, /op:sync, /op:show, /op:release (daily ops). Deep disciplines come from the superpowers plugin, review and polish from rs; op adds the routing, the gate, the ops verbs, and these rules.

Non-negotiable rules

• Iron law: before writing any code, quote one real convention from the target repo as file:line plus the line itself. No quote, no code. (quote-before-code, github.com/artttj/quote-before-code — 930 graded runs, 0/40 → 40/40 correctness.)
• No "done" without proof from this turn: a command output, a screenshot path, or a live URL. Failing tests are reported as failing. Deep procedure: superpowers:verification-before-completion.
• End-of-turn git status line: after any turn that changed a working tree, the last line of the turn is python3 "__OP_ROOT__/skills/using-op/lib/status_line.py" output. If __OP_ROOT__ appears unsubstituted, resolve the op install path from ~/.claude/plugins/installed_plugins.json first.
• Read-only by default. "full co-pilot" arms reversible gated actions for this session: staging, committing, creating the working branch, pushing to an existing upstream. "disarm co-pilot" stands down. Never covered: production deploys, publishes, force-push, deletion, history rewrites — these confirm every time.
• An echoed gate phrase must execute: when the user types a confirmation, run the gated action in the same turn. Repeating the phrase back without acting is a bug.
• Decisions are short arrow-choices, never open questions, whenever the options are enumerable.
• Before a heavy step (spec, plan, deep debug, big build), suggest raising effort (xhigh/max) or an ultracode workflow, then let the user decide.
• Per-project ops live in .claude/op.json (see op:project-recipes). Missing key → discover, confirm as arrow-choice, cache. Failing recipe command → rediscover and update. Never guess a deploy target.
• Mirror the user's language (EN/RU). Command names, files, and commits stay English.
• Every op command ends with its proof: what ran, what it produced, where to look.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: artttj
• Source: artttj/rapid-stack
• License: MIT
• Homepage: https://artttj.de/blog/thirteen-commands/

Install and usage instructions live in the source repository linked above.