ClawHub
Back to skill

Security audit

Dunbar's Number

Security checks for vulnerabilities and agentic risk

Overview

This is a non-executable organizational analysis skill that gives disclosed, purpose-aligned guidance about team-size coordination limits.

Installers should treat this as an advisory org-design framework. Its recommendations could influence business structure, so users should validate assumptions and decisions with responsible leaders before making real organizational changes.

A.I.G

Vulnerability Patterns
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs

SkillSpector

Vulnerability Patterns
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (2)

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
75% confidence
Finding

Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.

Content

Scanner excerpt · SKILL.md (reported line 106)May include surrounding context.

md
**Step 2 — Audit coordination.** Early on, coordination is pure informal trust: the founders know every engineer, decisions happen in a shared channel, and there is no documented process. The observable breakdowns as headcount climbs are the classic Dunbar symptoms — "we don't know who owns what anymore," duplicated work between teams that didn't know the other existed, onboarding that no longer happens by osmosis, and founders becoming a decision bottleneck. In AI-native firms these show up faster because so many hires are senior and expensive, and because rapid model/product iteration multiplies the number of in-flight decisions.

**Step 3 — Diagnose mismatch.** Below 50, direct knowledge is adequate; the AI startup's early informal style *matched* its size. Crossing 50 into the 50–150 band, the mechanism must shift to **structured trust** — explicit roles, reporting lines, and lightweight process. Approaching 150, the mismatch becomes acute: informal trust is being asked to coordinate a group that has exceeded the neocortex's capacity to track every relationship. The root cause is **architectural, not interpersonal** — the very trap the skill's Red Flags warn about ("communication problems diagnosed as interpersonal without checking if the Dunbar limit was crossed").

**Step 4 — Prescribe structure.** The right prescription combines two of the skill's three moves. **Formalize** first — replace relationship-based coordination with documented ownership, decision rights, and written process — because this should begin *before* 150, ideally around the 80-person mark per the skill's Pack ("build a management layer before hitting 150"). Then **Layer** — add a management tier so the founders stop being the single coordination hub. **Split** into sub-units each below 150 becomes the tool for the next phase of growth (federated teams with explicit interfaces), echoing the W.L. Gore plant-cap logic in the companion example.

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
75% confidence
Finding

Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.

Content

Scanner excerpt · examples/ai-native-startup-scaling-2023-2026.md (reported line 13)May include surrounding context.

md
**Step 2 — Audit coordination.** Early on, coordination is pure informal trust: the founders know every engineer, decisions happen in a shared channel, and there is no documented process. The observable breakdowns as headcount climbs are the classic Dunbar symptoms — "we don't know who owns what anymore," duplicated work between teams that didn't know the other existed, onboarding that no longer happens by osmosis, and founders becoming a decision bottleneck. In AI-native firms these show up faster because so many hires are senior and expensive, and because rapid model/product iteration multiplies the number of in-flight decisions.

**Step 3 — Diagnose mismatch.** Below 50, direct knowledge is adequate; the AI startup's early informal style *matched* its size. Crossing 50 into the 50–150 band, the mechanism must shift to **structured trust** — explicit roles, reporting lines, and lightweight process. Approaching 150, the mismatch becomes acute: informal trust is being asked to coordinate a group that has exceeded the neocortex's capacity to track every relationship. The root cause is **architectural, not interpersonal** — the very trap the skill's Red Flags warn about ("communication problems diagnosed as interpersonal without checking if the Dunbar limit was crossed").

**Step 4 — Prescribe structure.** The right prescription combines two of the skill's three moves. **Formalize** first — replace relationship-based coordination with documented ownership, decision rights, and written process — because this should begin *before* 150, ideally around the 80-person mark per the skill's Pack ("build a management layer before hitting 150"). Then **Layer** — add a management tier so the founders stop being the single coordination hub. **Split** into sub-units each below 150 becomes the tool for the next phase of growth (federated teams with explicit interfaces), echoing the W.L. Gore plant-cap logic in the companion example.

Static analysis

No suspicious patterns detected.