linux-ops
/linux-ops
Operates Linux servers: systemd services, permissions, disks, networking, firewalls and restores. Use when a service works in a shell but fails under systemd, a disk stays full after deleting files, DNS, TLS or firewall rules misbehave on a VPS or bare-metal host, SSH access is at risk during a network change, or a backup restore must be proven.
- Author
- lynricsy ↗
- Repository
- hyperskills ↗
- Licence
- MIT
- Last repository change
- 04.11.2026
Assessment and security
- Files in this package
- 13
- Scripts
- 0
- HTML comments
- 0
- URLs in SKILL.md
- 0
No signals detected. Repository changed on 04.11.2026, without references to retired models or APIs.
MIT, open: use and modify while retaining author attribution.
What it does and when to use it
Operates Linux servers: systemd services, permissions, disks, networking, firewalls and restores. Use when a service works in a shell but fails under systemd, a disk stays full after deleting files, DNS, TLS or firewall rules misbehave on a VPS or bare-metal host, SSH access is at risk during a network change, or a backup restore must be proven.
Original author description
Operates Linux servers: systemd services, permissions, disks, networking, firewalls and restores. Use when a service works in a shell but fails under systemd, a disk stays full after deleting files, DNS, TLS or firewall rules misbehave on a VPS or bare-metal host, SSH access is at risk during a network change, or a backup restore must be proven.
How it works
- Scope
- Core rules
- Workflows
- Topic router
- Output format
- Environment
SKILL.md preview
---
name: linux-ops
description: "Operates Linux servers: systemd services, permissions, disks, networking, firewalls and restores. Use when a service works in a shell but fails under systemd, a disk stays full after deleting files, DNS, TLS or firewall rules misbehave on a VPS or bare-metal host, SSH access is at risk during a network change, or a backup restore must be proven."
license: MIT (upstream attributions in NOTICE.md)
metadata:
author: HyperSkills
version: "2026.11.04"
category: platform
---
# linux-ops
Paths below are relative to this skill's directory.
## Scope
Operate at the Linux host boundary: identify which process, mount, listener,
manager or persistence layer owns the failure; change that layer without losing
management access, evidence or recoverable data. Cover systemd-based bare-metal
and VPS hosts, including distribution differences in service names and managers.
Keep these boundaries:
- Cloud IAM, provider networking and VM control planes belong to aws, azure or
gcp. A provider console may be a recovery prerequisite, not a deployment task.
- Container images and Kubernetes resources belong to containers. Do not apply
host firewall or namespace recipes inside a container as though it were a host.
- Application logic belongs to its language/framework skill; Linux merely being
the runtime is not a reason to diagnose it here.
- Database-consistent capture and replay require the database's supported
procedure. Own the host cutover, not an invented database backup protocol.
- Production telemetry pipelines belong to observability; shell program design
is not this skill's purpose.
A request to analyze supplied evidence authorizes analysis, not connecting to
machines or applying repairs. Write an executable plan only after its target
values and access boundary are known; do not silently execute that plan.
## Core rules
1. Bind every observation to host identity, time, boot, process and namespace where relevant; identical paths can name different objects.
2. Separate observation, proposed mutation and authorized execution; a diagnostic request does not authorize a restart or remote login.
3. Record the owner of runtime and persistent configuration before editing; two independent managers can immediately undo each other.
4. Require an external recovery path for changes that can remove access or destroy data, plus a tested rollback independent of the connection being changed; a second shell alone is not recovery.
5. Preserve volatile evidence before restarting, rotating, deleting or resetting failure state; the repair may erase the only explanation.
6. Distinguish disk configuration, manager-loaded configuration and live process state; daemon-reload does not retrofit namespaces, credentials or resource limits into MainPID.
7. Match readiness and reload claims to the daemon's actual protocol; active state and a successful signal helper are not proof that clients see the new configuration.
8. Keep non-root identity and isolation while fixing the specific denied operation; chmod 777, broad capabilities and disabling MAC or sandboxing are not diagnostic fixes.
9. Treat capability bounding as a ceiling, not a grant; inspect effective and ambient sets and exec-time privilege restrictions before changing them.
10. Count storage by filesystem and object identity, not path strings or FD rows; blocks, inodes and per-process descriptors exhaust independently.
11. Recheck PID identity and device/inode immediately before a signal or file action; process and descriptor numbers can be reused.
12. Verify new connections on every relevant address family after an access change; established sessions and IPv4 success can conceal a broken IPv6 or new-SSH path.
13. Validate TLS with the intended hostname, SNI, trust chain and endpoint; bypassing verification proves connectivity, not certificate correctness.
14. Select a recoverable point by explicit identity and consistency evidence, not unqualified latest; mirrors and successful backup jobs do not establish RPO or RTO.
15. Restore into an isolated target before cutover and coordinate all writers; an intact repository can still restore an unusable application state.
16. State exactly what each check proves and leaves untested; do not promote syntax checks, samples or repository checks into end-to-end recovery evidence.
## Workflows
### Diagnose a host incident
- [ ] Confirm the supplied or authorized host identity, boot/time interval,
distribution, kernel, relevant tool version and impact. Record whether the
evidence is live, a snapshot, or a synthetic incident.
- [ ] Choose the reference matching the observed boundary in the topic router;
do not run a full-host inventory to answer a single-service question.
- [ ] Collect the smallest discriminating evidence: effective unit and journal
for service failure; mount plus block/inode figures for storage; listener,
route, resolver and endpoint identity for network failure.
- [ ] Mark access-denied or unavailable observations as unknown, not empty.
Begin unprivileged; request only the privilege necessary to read the
missing evidence, excluding private keys and secret environment values.
- [ ] Write a causal hypothesis and the observation that would falsify it.
Compare the actual daemon context with a successful shell context rather
than treating a root-shell success as an application health test.
- [ ] Before an action that loses evidence, identify its owner, retention duty
and a protected off-host or unaffected-volume evidence destination.
- [ ] **Gate — localized cause:** cite evidence for the failing layer and one
discriminating check. If the check cannot be performed, label the proposed
cause as a hypothesis and provide the exact missing observation.
### Change a service or its execution environment
- [ ] Identify the exact unit instance, drop-ins, activation sockets/timers,
MainPID and worker processes. Read `references/service-privileges-and-limits.md`.
- [ ] Capture the current effective settings and rollback copies with ownership,
permissions and secrets protected. Specify availability budget, drain
method, health probe and external recovery access before disruption.
- [ ] Determine whether the change is daemon configuration, unit configuration,
socket configuration or live cgroup state. Specify the activation required
for each; do not use reload-or-restart to conceal that decision.
- [ ] Check syntax with the installed daemon's validator and, for unit changes,
`systemd-analyze verify` against the candidate unit. Treat unknown settings
as a compatibility problem, not a harmless warning.
- [ ] For denied access, resolve identity, ancestor traversal, ACL/MAC,
capability, filesystem and namespace evidence before changing one policy.
- [ ] For a namespace or execution-context change, drain and replace the affected
process under an approved window; retain the non-root and read-only design.
- [ ] Activate only the authorized change, then read the live process state.
Watch readiness, worker turnover and restart counters over a meaningful
interval, not only the instant after the command returns.
- [ ] **Gate — live activation:** the intended process observes the new object
or setting, a fresh client exercises it, isolation remains effective, and
the documented rollback remains available until acceptance.
### Recover capacity without erasing the incident
- [ ] Identify the filesystem behind the failing path and classify the failure
as blocks, inodes, quota, read-only/I/O, or descriptors. Read
`references/storage-and-processes.md`.
- [ ] Compare like-for-like `df` and scoped `du` observations; retain collection
times, traversal errors and mount boundaries.
- [ ] For deleted-open files, group holders by device/inode, estimate allocated
space once per object, and preserve required evidence before the last close.
- [ ] Identify the producer responsible for continuing growth. Propose its
supported reopen, retention or bounded shutdown mechanism rather than a
generic truncate, kill or package-cache cleanup.
- [ ] State exact targets, authorization, recovery copies and capacity required
for evidence preservation. If preservation is impossible, escalate the
loss tradeoff explicitly rather than silently deleting.
- [ ] **Gate — sustainable headroom:** remeasure blocks and inodes separately,
confirm the intended holders released the object, and demonstrate that a
representative write works without a continuing exhaustion trend.
### Change remote access, routing or filtering
- [ ] Read `references/network-and-remote-changes.md`; identify the real SSH
port, source network, destination addresses, jump path, firewall manager,
routing manager and persistence loader for both IP families.
- [ ] Record current runtime and persistent state, and test the independent
console/rescue path before touching the management path.
- [ ] Prepare the complete candidate and rollback transaction. Arm rollback
outside the SSH process tree and prove its status, deadline and privileges.
If reboot is in scope, account for how recovery survives reboot too.
- [ ] Validate configuration without activation; preserve management access and
required network control traffic in the candidate before restrictive policy.
- [ ] Apply one boundary at a time under explicit authorization; keep the old
session and do not remove the previous access path yet.
- [ ] **Gate — fresh remote access:** establish a non-multiplexed new SSH session
from the expected client path, exercise required IPv4/IPv6 business traffic
and TLS, inspect persistence, then cancel rollback and retire old access.
### Prove a backup and perform a recovery
- [ ] Read `references/backup-and-restore.md`; state the recovery point objective
(maximum data loss) and recovery time objective (maximum service outage).
- [ ] Select repository and snapshot ID by host, paths, time and application
consistency evidence. Compute recoverable age at the incident time; do not
substitute backup schedule frequency for observed recovery-point age.
- [ ] Establish isolated target identity, free blocks/inodes, metadata support,
restricted access and network containment before restoring untrusted data.
- [ ] Restore the selected scope, check content and metadata, then exercise the
recovered service without contacting production dependencies or writers.
- [ ] Time retrieval, provisioning, restore, replay, validation and switch-over.
Mark RTO unproven if only download speed or a small sample was timed.
- [ ] For cutover, stop or fence every writer, complete the approved final delta
or replay, retain the pre-cutover state, and allow one authoritative writer.
- [ ] **Gate — usable recovery:** the recovered application passes its own
integrity and client checks at the selected point, measured outage meets
RTO, observed data loss meets RPO, and rollback handles post-cutover writes.
## Topic router
| Topic | Read when | File |
|---|---|---|
| Service lifecycle, permissions, capabilities, namespaces and limits | A unit fails only under systemd, a reload does not activate data, or execution policy changes | `references/service-privileges-and-limits.md` |
| Storage and process ownership | A write fails, df and du disagree, deleted files retain space, or a process exhausts FDs | `references/storage-and-processes.md` |
| DNS, TLS, routes, SSH and host firewall | Reachability differs by client/address family, or a change may disrupt remote management | `references/network-and-remote-changes.md` |
| Consistent backup and controlled restoration | Evaluating recoverability, copying metadata-sensitive data, or planning a restore/cutover | `references/backup-and-restore.md` |
## Output format
Use this template for an incident or change review; omit inapplicable rows rather
than claiming they passed.
```text
Target: host / unit or filesystem / relevant version / observation time
Authority: analysis only | authorized observations | authorized mutation scope
Impact and invariants: availability, evidence retention, identity and isolation
Evidence: observed facts with source; hypotheses and missing observations separate
Cause: mechanism and falsifying check
Change: exact target, current -> proposed state, owner, activation and disruption
Recovery: external access, saved state, executor/deadline, rollback trigger
Verification: check -> expected result -> observed result -> limitation
Status: proposed | applied and verified | rolled back | blocked
```
For configuration findings, add `path:line - finding` and a minimal before/after
fragment. Separate a suggested command from one actually run. Never include
private keys, passwords or raw secret-bearing environment dumps in the report.
## Environment
Use the target's existing tools and managers. Common observations use systemd,
procfs, util-linux, coreutils, iproute2 and the installed OpenSSH; storage details
may require lsof, ACL/xattr tooling or the filesystem's native utilities.
Do not install diagnostic packages during an incident without approval: package
installation changes disk pressure and state. Request existing evidence instead.
Consult installed manuals (`man systemd.exec`, `man sshd`, `man rsync`) and
`restic help restore`/`restic help check` before using version-sensitive flags.
Files in this package (13)
- NOTICE.md 2 KB
- SKILL.md 13 KB
- SOURCES.yaml 6 KB
- evals/evals.json 7 KB
- evals/files/capacity-incident.txt 1 KB
- evals/files/network-change.txt 2 KB
- evals/files/restore-evidence.txt 1 KB
- evals/files/service-incident.txt 2 KB
- evals/files/tls-namespace-renewal.txt 4 KB
- references/backup-and-restore.md 9 KB
- references/network-and-remote-changes.md 8 KB
- references/service-privileges-and-limits.md 9 KB
- references/storage-and-processes.md 5 KB
Versions
The database has no other skills named linux-ops.
Database snapshot: 13.11.2026 · package version baf419b7da5c (SHA-256 digest of the entire package)
Downloads in the last 30 days: 0 · ranking
Report a problem with this skill: harmful content, incorrect descriptions or an author’s removal request.
Similar skills
- 100linux-service-ops
Configure and troubleshoot application services on an authorized Linux host with explicit service ownership. Use when managing systemd units, startup, logs, users, permissions or service resources; preserve the administrator existing distro configuration. - 100linux-systemd-service-management
Use this skill when authoring, sandboxing, and administering background daemon services using systemd unit files. - 86linux-server-ops
Use when diagnosing or managing a Linux server over SSH. - 100linux-service-triage
Diagnoses common Linux service issues using logs, systemd/PM2, file permissions, Nginx reverse proxy checks, and DNS sanity checks. Use when a server app is failing, unreachable, or misconfigured. - 90Linux
Debugs and hardens Linux hosts: permissions, disk full, OOM kills, stuck processes, systemd units, cron, networking, SSH, and boot failures. Use when a service starts by hand but fails at boot, a process ignores kill -9, df and du disagree, a box runs out of memory or inodes,… - 100net-ops
Cross-platform network troubleshooting (Windows, macOS, Linux) via local or remote shell. Use for: DNS broken, can't resolve hostnames, nslookup/dig works but apps fail, NRPT, WFP, scutil, /etc/resolver, systemd-resolved, resolv.conf, NetworkManager, VPN DNS leak residue…