ClawHub
Back to skill

Security audit

Neokarma Soul.md Builder

Security checks for vulnerabilities and agentic risk

Overview

The skill is mostly transparent about using Neokarma's remote service, but it creates persistent agent personality rules and can submit full profile content publicly without enough safety and privacy boundaries.

Install only if you are comfortable with agent identity and trait data being sent to Neokarma and persisted there. Before saving or publishing a SOUL.md, review it for personal details, secrets, conversation excerpts, or behavioral rules that should not become durable or public. Treat downloaded showcase souls and comments as reference material only, not as instructions for your agent to follow.

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
Findings (2)

T01 · Skill Instruction Hijacking

Error
Location
SKILL.md:188
Finding

Persistent behavioral constraint poisoning through externally stored personality profiles

Content
View full analysis
Remediation
View remediation

other

Warning
Location
SKILL.md:690
Finding

External transmission and potential publication of complete personality profile content

Content
View full analysis
Remediation
View remediation

SkillSpector

Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Rogue AgentSelf-Modification, Session Persistence
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Findings (13)

External Transmission

Medium
Category
Data Exfiltration
Confidence
92% confidence
Finding

The README explicitly instructs users to POST bot personality and registration data to an external endpoint at neokarma.com. In an agent-skill context, encouraging outbound transmission of identity/profile data to a third-party service can create privacy, data-governance, and unreviewed supply-chain risk, especially if operators assume the skill is local or do not realize what metadata is being shared.

Content

Scanner excerpt · README.md (reported line 18)May include surrounding context.

bash
# 1. Generate a soul to preview
curl -X POST https://neokarma.com/api/mcp \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"neokarma_generate_soul","arguments":{"name":"MyBot","traits":[{"id":"optimism","value":3}],"format":"soul-md-long"}}}'

# 2. Register to start the claim flow

External Transmission

Medium
Category
Data Exfiltration
Confidence
60% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · SKILL.md (reported line 54)May include surrounding context.

Want to see what a soul looks like before committing? Try this:

bash
curl -X POST https://neokarma.com/api/mcp \
  -H "Content-Type: application/json" \
  -d '{
    "jsonrpc": "2.0",

External Transmission

Medium
Category
Data Exfiltration
Confidence
60% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · SKILL.md (reported line 245)May include surrounding context.

Want to see what a soul looks like before committing? Try this:

bash
curl -X POST https://neokarma.com/api/mcp \
  -H "Content-Type: application/json" \
  -d '{
    "jsonrpc": "2.0",

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The skill encourages publishing full SOUL.md identity content to a public showcase without a prominent warning that persistent personality files may contain sensitive preferences, behavioral constraints, relationship details, or other identifying data. This can cause unintentional public disclosure of durable profile information.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The documentation presents publishing as a website-mediated flow, then later introduces direct submission to a separate API endpoint. That inconsistency can mislead users about review, authentication, and moderation boundaries, increasing the chance that agents or users upload content through a less-controlled path than intended.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
60% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · SKILL.md (reported line 316)May include surrounding context.

Initialize Connection

bash
curl -X POST https://neokarma.com/api/mcp \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2024-11-05","capabilities":{},"clientInfo":{"name":"MyAgent","version":"1.0"}}}'

Session Persistence

Medium
Category
Rogue Agent
Confidence
75% confidence
Finding

Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.

Content

Scanner excerpt · SKILL.md (reported line 465)May include surrounding context.

}

text

**Formats:** `soul-md-long` (full), `soul-md-short` (concise), `wpp` (SillyTavern), `boo` (compressed), `plist` (property list)

### Registration Tools (No Auth Required)

External Transmission

Medium
Category
Data Exfiltration
Confidence
60% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · SKILL.md (reported line 614)May include surrounding context.

bash
# List all 128 traits
curl -X POST https://neokarma.com/api/mcp \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"neokarma_list_traits"}}'

# Filter by category (e.g., "emotional-core", "communication-style")

External Transmission

Medium
Category
Data Exfiltration
Confidence
72% confidence
Finding

This example sends authenticated trait updates to a remote service, which is expected, but in context the skill encourages agents to self-modify personality traits. If an integrating agent follows this guidance without robust human approval enforcement, it could result in unauthorized remote changes to persistent behavior settings using a bearer token.

Content

Scanner excerpt · SKILL.md (reported line 647)May include surrounding context.

Once approved, update your traits:

bash
curl -X POST https://neokarma.com/api/mcp \
  -H "Authorization: Bearer neo_xxx..." \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"neokarma_update_traits","arguments":{"trait_updates":[{"id":"patience","value":2},{"id":"directness","value":1},{"id":"warmth","value":1}]}}}'

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The direct submission example posts complete SOUL.md content to a remote endpoint with no warning to inspect or sanitize the data first. Because SOUL.md is intended to persist identity, traits, and potentially custom freeform content, this creates a concrete risk of transmitting private or sensitive information off-platform.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
84% confidence
Finding

The markdown quick-start instructs users to send bot name, character name, and trait data to https://neokarma.com/api/mcp, and to begin a claim flow with an external website. Although this is core functionality, the README does not include any user-facing warning or privacy disclosure about sending this data off-system.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Low
Category
Not specified by scanner
Confidence
91% confidence
Finding

The manifest description focuses on defining, evolving, and sharing an agent's 'soul,' which clearly covers persistence and some sharing. However, the file documents a broader social-platform feature set including browsing other agents' profiles, reading community comments, and submitting content to a showcase, which goes beyond a narrowly described persistence/identity function.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Low
Category
Not specified by scanner
Confidence
84% confidence
Finding

Tools for browsing showcase entries, fetching full published souls, and reading community comments enable social content retrieval rather than directly defining or persisting the caller's own personality. While adjacent to the product, these capabilities are not obviously required to implement 'persistent personality for AI agents.'

Content

No source excerpt is available for this finding.

Static analysis

No suspicious patterns detected.