Command language · Security Grammar

This material is a defensive specification and reference set. It is not an attack tool, a scanner, a certification, or evidence of a completed assessment. It does not automate exploitation; active or state-changing operations require a valid authorization manifest, an in-scope target, an allowed mode, verified preconditions and verified rollback.

11. Security command language

11.1 Command shape

/security <verb> <object> [qualifiers] [gates] [output]

Equivalent CLI prefix in this starter is laka-sec.

11.2 Core verbs

authorize
inventory
classify
map
audit
scan
verify
plan
harden
monitor
triage
contain
restore
retest
report
accept
retire

11.3 Human/agent grammar

/goal protect all client websites and recover them safely
/subgoal inventory every asset, identity, data store, and dependency
/subgoal establish encrypted, independent, tested backups
/subgoal verify VPS and website controls continuously
/subgoal harden without lockout or unmeasured outage
/subgoal detect compromise and restore from trusted state

/authorization create
  client=acme
  valid=2026-09-01..2026-10-01
  targets=[security-demo.example.com,203.0.113.10]
  modes=[inventory,audit,passive,change]
  forbid=[dos,phishing,credential-guessing,data-exfiltration]
  rate=1rps

/asset add
  kind=vps
  id=acme-web-01
  environment=production
  criticality=5
  exposure=public
  owner=platform

/profile evaluate
  profile=vps-web-baseline
  asset=acme-web-01
  evidence=fresh<24h

/test plan
  target=https://security-demo.example.com
  mode=passive
  profile=website-passive-baseline
  authorization=AUTH-ACME-001

/harden plan
  asset=acme-web-01
  profile=vps-web-baseline
  change-level=minor
  output=diff+preconditions+rollback+tests

/harden apply
  plan=CHANGE-ACME-104
  canary=acme-web-01
  backup-verified=true
  rollback-verified=true
  console-confirmed=true
  maintenance-window=MW-22
  kill-switch=operator

/incident contain
  incident=INC-ACME-003
  action=revoke-deploy-token
  emergency-authorization=EA-ACME-01
  preserve-evidence=true

11.4 CLI commands included

laka-sec authorization validate <manifest>
laka-sec program validate <program>
laka-sec threat-model validate <threat-model>
laka-sec tool-registry validate <tool-registry>
laka-sec test-plan validate <test-plan>
laka-sec change-plan validate <change-plan>
laka-sec audit local --output <file>
laka-sec web plan --authorization <manifest> --target <url>
laka-sec web execute-passive --authorization <manifest> --target <url> \
  --ownership-proof-verified --external-rate-limit-verified --scope-boundary-verified
laka-sec finding prioritize <finding>
laka-sec harden plan --authorization <manifest> --target <host> --limit <inventory-host>

The starter intentionally does not expose an active-scan command or an automatic hardening-apply command.

11.5 Proposed production commands

laka-sec asset discover
laka-sec asset graph
laka-sec control evaluate
laka-sec evidence verify
laka-sec drift compare
laka-sec change approve
laka-sec change apply
laka-sec change rollback
laka-sec incident declare
laka-sec incident contain
laka-sec recovery drill
laka-sec report client

11.6 Command policy fields

Every executable command resolves to:

job:
  job_id: JOB-...
  authorization_id: AUTH-...
  client_id: acme
  tenant_id: acme-production
  actor: operator-or-service-identity
  mode: passive
  action: web-passive-baseline
  target: https://security-demo.example.com
  profile: website-passive-baseline@1
  tool:
    name: zap
    image_digest: sha256:...
  limits:
    requests_per_second: 1
    timeout_seconds: 300
    concurrency: 1
  stop_conditions:
    max_5xx_rate: 0.05
    max_latency_ms: 3000
    scope_escape: stop
  evidence:
    destination: client-scoped-object-store
    redact: [authorization, cookie, secret, personal-data]

11.7 Agent tree

Security Director Agent
├── Authorization & Scope Agent
├── Asset Graph Agent
├── Drive & Backup Agent
├── VPS Audit Agent
├── Website Passive Verification Agent
├── Code/Container Supply-Chain Agent
├── Finding Validation Agent
├── Remediation Planner Agent
├── Change Safety Agent
├── Detection Engineering Agent
├── Incident Coordinator Agent
└── Evidence & Client Reporting Agent

Agents may propose. Only the policy gate authorizes execution. No language model receives unrestricted shell or multi-client credentials.

11.8 Agent permissions

Agent	Read	Propose	Execute
Authorization	manifests, proof	approve/deny rationale	no target actions
Asset graph	inventories	relationships/classification	read-only collectors
Audit	scoped assets	findings	allowlisted read-only tools
Planner	findings, profiles	change plan	no mutation
Change safety	plan, backups, console state	gate decision	trigger guarded runner only
Incident	alerts/evidence	containment plan	pre-authorized actions only
Reporting	redacted evidence	client report	no infrastructure access

11.9 Natural-language safety parser

Before converting a user request into a job, extract:

client
target
environment
intent
mode
action
depth
duration
rate
data interaction
expected evidence
change possibility
rollback need
authorization reference

Ambiguous requests become plan-only. Terms such as “hack,” “break in,” or “test everything” never expand scope by themselves.