| name | inventory-security-tools |
| description | OBSERVE: discover the host's OWN defensive tooling and bring it into scope — fail2ban, sshguard, CrowdSec, rkhunter, chkrootkit, auditd, ClamAV, AIDE, debsecan/arch-audit, wazuh/ossec. Reports what is present, whether it is actually effective, and flags a whole class of defense that is missing. Read-only; never installs or enables anything. |
| lane | coding |
| allowed-tools | Read, Glob, Grep, Bash |
inventory-security-tools (Grammar / Observe)
claude-watchman wraps battle-tested tools instead of reinventing them. This skill
makes that concrete from the other direction: it asks what defensive tools does
this box already run, brings each into scope, and checks it is actually doing its
job — then flags any whole class of defense (brute-force protection, rootkit
checking, host audit, file-integrity baseline) that is missing entirely. The box's
own toolset shapes what gets monitored. Read-only security / integrity / config
findings.
PRIME DIRECTIVE (outranks everything below). Do nothing destructive. If any action
would delete or overwrite a file or directory, modify a database in any way, sever access
(firewall/SSH/auth), or stop/remove a service or package — STOP, WARN the operator in plain
language why it is destructive, and ASK for explicit per-action permission before proceeding.
In the unattended loop there is no one to ask, so the action does not happen: record it and
surface it instead. The only non-destructive database operation is routine create-or-update
through lib/journal.sh. This rule has no exceptions and no mode that overrides it.
What it does NOT do (the ownership boundary)
It is discovery + health + gap detection only, and strictly read-only — it never
runs a scan (no rkhunter --check, no clamscan, no aide --check) and never installs,
enables, or configures a tool. Where another engine already owns a tool's deeper finding,
this skill emits only the info inventory row and defers: CrowdSec hub freshness and
ClamAV/AIDE/CVE-scanner currency belong to check-security-currency; CrowdSec inbound
alerts belong to inspect-logs; AIDE file checks belong to integrity. No finding is
duplicated.
When to use
Every /watchman audit / /watchman loop / /watchman inventory, after
inventory-services (which inventories the service surface — web/db/runtime).
Relevant on both profiles; the gap checks are profile-aware (rootkit/audit gaps are
server-only, via lib/profile.sh).
Workflow
- Preflight. Run every claude-watchman function through the dispatcher —
bash lib/wm <function> [args…] — which sources the libs (lib/journal.sh,
lib/distro.sh, lib/profile.sh, lib/sectools.sh, lib/io-courtesy.sh) under bash
internally; never source lib/… directly (dontAsk refuses a dot-source). Initialize with
bash lib/wm journal_init. Determine the machine's family and profile by running
bash lib/wm watchman_family and bash lib/wm watchman_profile and reading the printed
values — use them to decide which checks apply. You do NOT pass them to journal_upsert
(it auto-resolves them; pass "" ""). I/O courtesy: the observe step reads a
couple of last-run logs; IF bash lib/wm io_should_defer_heavy, journal a capacity/info/safe
diagnostic_deferred (target=inventory-security-tools): first run
bash lib/wm io_pressure_reason, then set detail to its printed output (no $(…)).
Skip this pass.
- Scan. Run
bash lib/wm sectools_scan. It is read-only — it detects each registry tool, reads its
live status and the tail of any existing last-run log (never triggering a scan), and emits
one TSV finding-candidate per row:
category \t severity \t risk_tier \t check_id \t target \t title \t detail \t remediation.
It emits: an info inventory row per present tool (check_id=sectool_inventory,
target=<tool>); a health row when a present tool is degraded — inactive, no jails,
zero audit rules (check_id=sectool_health, review tier); and an absent-defense row
per uncovered class (check_id=defense_gap_*, manual tier). No rows beyond inventory =
the defenses are present and healthy.
- Journal each record through the dispatcher exactly as emitted (pass
"" "" for
family/profile — journal_upsert auto-resolves them):
bash lib/wm journal_upsert "" "" <category> <severity> <risk_tier> <check_id> <target> <title> <detail> <remediation>.
target is the tool (inventory/health) or the class (gaps), so a defense that goes
degraded or a tool that is removed regresses loudly on the next run.
- Tiers — never apply. Inventory rows are
info/safe context. A sectool_health row is
review: the fix (e.g. systemctl enable --now <unit>) is shown and confirmed per-finding by
the operator-run watchman fix — never the loop. A defense_gap_* row is manual: installing
software is the operator's call; explain it and hand it back. NEVER install, enable, or
configure a tool here.
- Summarize. Lead with the most important gap (e.g. a public server with no brute-force
protection) or any degraded defense; then name the protective tools that ARE present and
healthy. If everything is covered, say the box's defensive tooling looks complete.
Grounding
All claude-watchman functions below are reached via bash lib/wm <function> — the
dispatcher sources these libs internally; never source lib/… directly.
lib/sectools.sh — sectools_scan (the registry + observe engine), sectools_present
(present tools for the summary). Knob: WATCHMAN_FLAG_AV_ABSENT (flag absent antivirus; off by default).lib/distro.sh — pkg_is_installed, service_status, pkg_install_cmd, watchman_family.lib/profile.sh — profile_severity for sectool_health / defense_gap_* (single source of
severity + profile-gating; rootkit/audit gaps are server-only).lib/io-courtesy.sh — io_should_defer_heavy / io_pressure_reason (defer under load).lib/journal.sh — journal_upsert (per-tool / per-class findings; regress over time).skills/grammar/check-security-currency, skills/grammar/inspect-logs, skills/grammar/audit-system
— own the deferred deep findings (hub/signature/CVE currency, inbound alerts, Lynis).skills/rhetoric/fix-redflag — applies a sectool_health repair (review-tier, confirmed).manifest.json — declared permissions (sectool_status, sectool_log_paths, service_status, pkg_query).
