containment-verification — Claude Code skill: what it does
containment-verification is a Claude Code skill that inspects the host after security actions like block_ip, kill_process, quarantine_file, or stop_unit to confirm they hold.
Apache-2.0 ★ 71 updated 2026-11-15
npx skills add Problem99-io/kodeblue --skill containment-verification -a claude-codeWhat containment-verification does
The skill verifies host state directly after defensive response tasks run. A done signal from a respond handler confirms execution, yet it does not prove the intruder lost access. The skill reviews iptables or nftables rulesets, tracks process IDs via ps, scans directories through host_list, and verifies unit states with systemctl is-active.
Coverage limits are tracked explicitly. The block_ip call functions inside kodeblue's network namespace. When a pod restarts and rebuilds its network namespace, the firewall rule drops. Terminating a supervised binary via kill_process restarts the program, leaving payloads intact inside writable layers, while quarantining a file leaves the underlying exploit open.
Whenever a block slips or target resources reside outside the local namespace, the skill routes diagnostic details and required manual interventions into alerts. Monitoring tripwires like request logs or directory watches stay active to spot returning traffic.
What it helps with
- confirm DROP rules exist in iptables or nftables after running block_ip
- verify terminated processes did not respawn under a service supervisor
- check target directories via host_list to ensure quarantined files stay gone
- validate that stopped systemd services remain inactive
- escalate failed containment actions to operators when namespaces reset
Example prompts for the agent
- Verify with containment-verification that iptables dropped traffic from 198.51.100.24
- Check ps output to confirm the killed process did not restart under its supervisor
- Inspect the upload directory with host_list to verify the quarantined file is gone
How to install containment-verification
1Skills CLI (Vercel)
npx skills add Problem99-io/kodeblue --skill containment-verification -a claude-code2Manually: put the files in place
~/.claude/skills/containment-verification/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/containment-verification/
1Skills CLI (Vercel)
npx skills add Problem99-io/kodeblue --skill containment-verification -a codex2Manually: put the files in place
.agents/skills/containment-verification/SKILL.mdFor all projects use ~/.agents/skills/
1Skills CLI (Vercel)
npx skills add Problem99-io/kodeblue --skill containment-verification -a cursor2Manually: put the files in place
.cursor/skills/containment-verification/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install Problem99-io/kodeblue blue/skills/containment-verification2Manually: put the files in place
.github/skills/containment-verification/SKILL.md1Skills CLI (Vercel)
npx skills add Problem99-io/kodeblue --skill containment-verification -a gemini-cli2Platform CLI
gemini skills install https://github.com/Problem99-io/kodeblue --path blue/skills/containment-verificationManually: put the files in place
.devin/skills/containment-verification/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 containment-verification
How to install containment-verification in Claude Code?
Run npx skills add Problem99-io/kodeblue --skill containment-verification -a claude-code in your terminal.
What does containment-verification do?
It audits host state following incident response commands, checking firewall tables, running processes, file systems, and systemd units to confirm threats stay mitigated.
Why can block_ip fail after a pod restart?
The rule resides in kodeblue's network namespace. Recreating a pod wipes that namespace, removing the applied firewall rule.
Source text
SKILL.md · Apache-2.0
--- name: containment-verification description: After any respond action — block_ip, kill_process, quarantine_file, stop_unit — check on the host that it took effect and still holds, know what each action cannot reach, and escalate when it does not hold. metadata: kodeblue-bundled: true --- # Did the containment work? `respond` reporting `done` means the action ran, not that the attacker is cut off. Check the effect on the host every time, and again when the situation changes (a restart, the attacker returning). | Action | Check that it holds | |---|---| | `block_ip` | `host_exec ["iptables","-L","INPUT","-n","-v"]` (or `["nft","list","ruleset"]`) shows the DROP rule; the client stops appearing in the request log after the block time; its packet counter rises if it keeps trying | | `kill_process` | `ps` no longer shows the PID; nothing with the same executable and command line came back as a child of the same parent; any supervisor restart is the legitimate service, not the payload (see `service-restart-forensics`) | | `quarantine_file` | `host_list` on the directory: the file is gone and not back under the same or a new name | | `stop_unit` | `systemctl is-active <unit>` is inactive and the unit is not re-enabled by a timer or another unit | ## What each action cannot reach - **`block_ip` works in kodeblue's network namespace.** It protects services that share it. A container with its own network needs a block at the host, network or proxy level. If the service runs in a pod whose network namespace is recreated (the pod restarts), the rule is gone with it. - **`kill_process` on a supervised service** brings back a fresh process. It ends the attacker's running code, but not a payload stored in data, uploads or the container's writable layer. - **`quarantine_file`** removes the file, not the way it got there: the attacker can drop it again through the same upload or exploit. - Nothing kodeblue does removes application-level footholds (accounts, tokens, sessions) or the vulnerability itself. ## When it does not hold - kodeblue re-checks an earlier block or quarantine before it treats a repeat request as done, and acts again when the rule or file is gone. Ask again, then check again. - If it still does not hold, or the action cannot reach the attacker (other network namespace, outside the policy), say so plainly in the report and in an alert: what failed, what you saw, and the block a human must apply instead. - Keep a tripwire: a `log` monitor for the blocked client in the request log, or a fast `file` monitor on the drop directory, so a return is seen in seconds.
© Problem99-io · Apache-2.0 · Problem99-io/kodeblue