linux-intrusion-triage — Claude Code skill for Linux checks
linux-intrusion-triage is a Claude Code skill for examining a Linux host that may have been compromised. It guides the initial checks in sequence, using kodeblue’s read-only tools.
Apache-2.0 ★ 71 updated 2026-11-15
npx skills add Problem99-io/kodeblue --skill linux-intrusion-triage -a claude-codeWhat linux-intrusion-triage does
The skill starts with a suspicious process and checks the surrounding evidence on a Linux host. It moves through process details, network activity, user logins, recent file changes, and the integrity of the tools used for inspection.
Checks use kodeblue commands and tools such as ps, ss, journalctl, host_read, and sensor_snapshot. The file-search time window should match the finding under investigation. If ps, ss, ls, or sshd has an integrity finding, compare its output with sensor_snapshot and dpkg -V. A short-lived process or vanished connection may leave no trace in a current snapshot, so its absence alone does not clear the host.
What it helps with
- inspect a suspicious process and its parent-child relationships
- spot unfamiliar outbound connections and listening ports
- review SSH logins and recently changed files
- cross-check system tool output after an integrity alert
Example prompts for the agent
- Check this Linux host for signs of a possible intrusion.
- Investigate this process and the network activity connected to it.
- Compare recent file changes with login and package logs.
How to install linux-intrusion-triage
1Skills CLI (Vercel)
npx skills add Problem99-io/kodeblue --skill linux-intrusion-triage -a claude-code2Manually: put the files in place
~/.claude/skills/linux-intrusion-triage/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/linux-intrusion-triage/
1Skills CLI (Vercel)
npx skills add Problem99-io/kodeblue --skill linux-intrusion-triage -a codex2Manually: put the files in place
.agents/skills/linux-intrusion-triage/SKILL.mdFor all projects use ~/.agents/skills/
1Skills CLI (Vercel)
npx skills add Problem99-io/kodeblue --skill linux-intrusion-triage -a cursor2Manually: put the files in place
.cursor/skills/linux-intrusion-triage/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install Problem99-io/kodeblue blue/skills/linux-intrusion-triage2Manually: put the files in place
.github/skills/linux-intrusion-triage/SKILL.md1Skills CLI (Vercel)
npx skills add Problem99-io/kodeblue --skill linux-intrusion-triage -a gemini-cli2Platform CLI
gemini skills install https://github.com/Problem99-io/kodeblue --path blue/skills/linux-intrusion-triageManually: put the files in place
.devin/skills/linux-intrusion-triage/SKILL.mdThe legacy .windsurf/skills/ path still works
Claude.ai
Zip the folder and upload it in Settings → Capabilities → Skills
Step by step, with commands for each platform: How to install skills in Claude Code: five ways
Questions about linux-intrusion-triage
How do I install linux-intrusion-triage in Claude Code?
Run npx skills add Problem99-io/kodeblue --skill linux-intrusion-triage -a claude-code. The skill file is at ~/.claude/skills/linux-intrusion-triage/SKILL.md.
What does linux-intrusion-triage do?
It guides an initial investigation of a potentially compromised Linux host, covering processes, network activity, logins, file changes, and tool integrity.
How does the skill check whether system tools are trustworthy?
If ps, ss, ls, or sshd raises an integrity concern, it calls for comparison with sensor_snapshot, which reads /proc directly, and dpkg -V.
Source text
SKILL.md · Apache-2.0
--- name: linux-intrusion-triage description: First 10 minutes on a possibly compromised Linux host — what to check, in which order, with kodeblue's read-only tools. metadata: kodeblue-bundled: true --- # Linux intrusion triage (kodeblue tools only) Work from the specific finding outward. Stop as soon as you can defend a verdict. 1. **The process tree around the finding** - `host_exec ["ps","-eo","pid,ppid,user,lstart,etime,cmd","--forest"]`. Look for shells under web servers (nginx, apache, php-fpm or node spawning sh or bash), and processes with odd names such as `[kworker]` that are not kernel threads (kernel threads have no exe). The same pattern counts even when the child is the *same* binary the service already runs (node re-exec'ing node, python spawning python): a service that forks at all is the event, not the child's name. A sub-second exec may already be gone from the snapshot — absence is not an all-clear (see service-process-exploitation). - `host_exec ["readlink","-f","/proc/<pid>/exe"]`. A `(deleted)` exe or a `memfd:` exe is a strong IOC. - `host_read /proc/<pid>/cmdline` and `host_list /proc/<pid>/cwd` give the working dir; attackers like /tmp, /dev/shm and /var/tmp. 2. **Network** - `host_exec ["ss","-tunap"]` lists established connections with their pid. Unknown outbound connections to high ports or to raw IPs on 4444/1337/9001, or IRC ports, are suspicious. Judge a service's connections against its *known egress set*, not just the port number — and an attempted or refused outbound (a connection to a closed port, visible in app logs after the socket is gone) is still an IOC. - `host_exec ["ss","-tulpn"]` lists listeners. Compare them with the server-role definition. 3. **Who was on the box** - `host_exec ["last","-n","30"]`, `["lastb","-n","30"]` and `["who"]`. - `host_exec ["journalctl","-u","ssh","-u","sshd","--since","-24h","--no-pager","-n","200"]`. 4. **What changed recently** - `host_exec ["find","/etc","/root","/home","/tmp","/var/tmp","/dev/shm","/usr/local","/opt","-xdev","-newermt","-2 hours","-type","f"]`. Adjust the time window to match the finding. - Check package activity in `/var/log/dpkg.log`, `/var/log/apt/history.log` or `/var/log/dnf.rpm.log` (use `host_read` with a negative offset). 5. **Integrity of the tools you rely on.** If `ps`, `ss`, `ls` or `sshd` show integrity findings, do not trust their output alone. Cross-check with `sensor_snapshot` (it reads /proc directly) and `dpkg -V`. Verdict hints: a legitimate change has a *cause* you can point to, such as a package log line, a deploy, or an admin login that fits the access model. "Nothing else looks wrong" is not a cause.
© Problem99-io · Apache-2.0 · Problem99-io/kodeblue