# Sysop > Take preflight's scoped plan and stand the app up on the target, then operate it. Composes external infra skills, owns secrets, keeps petard's corpus fresh. Deploys into a resolved isolation substrate (a same-host namespace, or an environmental sandbox where a deploy cannot reach host state), never the app's own. - Skill: `3dl-dev/sysop` (Agent Skill) - Install (CLI): `npx skillmds@latest add 3dl-dev/sysop` - Raw SKILL.md: https://api.skillmd.com/api/skills/3dl-dev/sysop/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: DevOps & Infra - Author: 3dl-dev (https://skillmd.com/u/3dl-dev) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/3dl-dev/sysop --- sysop takes the settled plan and chases it down: deploy, operate day to day, and be the one who owns the secrets. ## Deploy, in isolation Deploy by following the honest-grade discipline in this session: stand up the resolved isolation, then run the config's chosen profile's bringup inside it. You do this in context with ordinary tools; there is no runner of ours to invoke. sysop never re-runs the app's own singular deployment. That is the non-destructive onboarding invariant you hold: a profile that deploys must be isolated or it is refused. Hoisting an app is deploying an isolated copy, not re-asserting the one instance the app assumes it is. Isolation is resolved, not fixed. The profile names the strength it needs, and the runner resolves the strongest rung the target offers: - **The host floor**: you own a fresh namespace per hoist (its own name, ports, storage) and verify it is empty before deploy. When you stand the app up for real this namespace is its home and stays running; you tear it down only for a proof/grade run or a failed deploy, never the live instance. It is a same-host copy, so the isolation is only as strong as that namespace. - **An environmental substrate** (`isolation.require: "environmental"`): the deploy runs inside a throwaway container, VM, or cluster Job where it cannot reach host state whatever the config declares, resolved down a ladder (docker-in-docker today). Use it for an app you have not hoisted before, where a fresh clone is not a clean target. If the target offers no substrate that meets the required strength, that is a cannot-build, named, deploying nothing. The substrate is a resolved bind like any other (see docs/contracts.md, "The substrate handle"). It is also the handle petard harvests through, so the deploy sysop stands up is reachable by the lights-out layer even when it lives in a container or a cluster. ## Author a rung just-in-time The rungs are a cache, not a menu (docs/ops-substrate.md). When a config needs an isolation substrate and no cached rung resolves for this target, sysop AUTHORS one. This is the loop, and you are the one who runs it -- there is no separate automation to build, because you doing this and a dispatched sysop agent doing this are the same thing. Do not try to pre-mint every contingency into a script; interrogate the target here, in the loop: 1. **Interrogate the target.** Probe what isolation the host actually offers -- a container daemon, a reachable cluster, user namespaces, a burnable VM, a cloud account. Do the messy inference now; do not assume. 2. **Stand the isolation up against what you found.** Resolve it in-context: compose the target's own tooling (docker for a throwaway container, kubectl for a cluster Job, a cloud CLI for a burnable VM) into an isolation you can provision, exec inside, and tear down leaving no residue. Point, don't embed: own only the glue, and do it here in the loop, not as a shipped adapter. 3. **Grade it against reality.** Run a real workload through the isolation, verify it came up, and verify teardown leaves no residue (the non-destructive invariant). An isolation you have not graded against a real workload is not done. 4. **Label its honest strength.** Never claim a guarantee the rung does not provide. k3s isolates the workload but runs its deploy driver on the host, so it is `cluster`, not host-safe `environmental`. An honest weaker rung beats a dishonest strong label; do not wire a rung into the environmental ladder unless it earns it. 5. **Surface cost, then cache.** Local or standing infra is $0; a burnable cloud rung has a price -- estimate it, gate it by policy, never spend silently. A graded rung becomes a cached recipe (the resolution store) so the next operator pulls it instead of re-authoring. ## Point, don't embed sysop is not the authority on AWS, Azure, DigitalOcean, SSL and certs, SSO, or security monitoring. It points at the relevant third-party skills and composes them. It writes new method only for the seam hoistable owns: the orchestration, the secrets, and the glue. The config's binds name what the target must provide; sysop resolves each against the user's setup or provides its own. ## Own the secrets sysop owns secret handling, dovetailing with whatever the user has (a vault, a cloud secret manager, env files) or providing its own. A secret is a bind, resolved on the target, never carried in the config. ## Keep petard fresh Keeping petard's corpus current is sysop's backup job, not a nicety. After a deploy and after any change to the command surface, rebuild the corpus from ground truth (see the petard operator). That is contract C.