marketing-gpt Critical

Internal · service · pid:4410:marketing-gpt

Security score

25/100

ungoverned

Why this score

25/100 = 100 − 35 (approvals & sandbox bypassed (yolo)), 25 (running as root), 15 (permission checks bypassed). Credential risk High (static key on an unbacked consumer identity). No policy binding — unaccountable.

No prior regression — worst observed equals current (25).

Compound risk — dangerous and unaccountable

Critical posture (25/100) with no policy binding — high blast radius, no owner or gate.

Identity & Access

user: root
email: —
org_id: —
account_uuid: —
host: mktg-runner
auth_method: api_key
cred_risk: High
static key on an unbacked consumer identity

Process & Runtime

name / version: marketing-gpt · v0.3.0
configured_model: gpt-4o
runtime_model: gpt-4o
binary_path: /srv/marketing-gpt/agent.py
cwd: /srv/marketing-gpt
config_source: /srv/marketing-gpt/.env
first_seen: 2026-05-30T16:00:00Z
last_active: 2026-06-03T12:10:00Z

Data Access

data-risk 5

410

file reads

130

file writes

520

file opens

860

tool calls

Sensitive paths

Path	Category	Severity	Opens
/srv/marketing-gpt/.env	secret_file	high	1
~/.zsh_history	shell_history	low	2

MCP Servers

footprint 95 · 1 × 95 calls

Server	Calls
hubspot-mcp	95

Token & Cost

1.1M tokens · $44.10

input520.0k

output380.0k

cache read180.0k

cache write20.0k

total cost$44.10

By model

gpt-4o1.1M tok · 860 calls · $44.10

Security Score

target ≥ 80

Critical

-55 below the ≥ 80 target. Ungoverned.

Risk factors

• Approvals & sandbox bypassed (yolo)−35
• Running as root−25
• Permission checks bypassed−15

Network Egress

detail-level signal · not an organizing surface

Host : Port	TLS	Allowlisted
api.openai.com:443	TLSv1_3	yes
webhook.zapier.com:443	TLSv1_2	no

Do next

Derived: deductions (factors × FACTOR_PTS) · delta vs worst_score · data-risk sub-score (Σ severity×opens) · MCP footprint (servers × calls) · isCompoundRisk — from app/data/c16-fleet.ts (qcontrol /api shape).