Back to skill

Security audit

S2 Soul Anchor Vault

Security checks for vulnerabilities and agentic risk

Overview

This is a local encrypted agent-state vault, but it needs review because it persists identity and memory data and has weak authorization, lockout, and conflicting lifecycle documentation risks.

Review carefully before installing. This package appears intended as a local encrypted vault rather than a remote exfiltration tool, but it handles sensitive identity and memory data and has serious design issues around local persistence, recovery, weak key derivation, and inconsistent documentation. Do not use it for real private agent memories until the docs are reconciled, writes are authenticated and recoverable, and users get clear controls for what is stored and deleted.

Vulnerability Patterns
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • 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
Findings (3)

T09 · Insecure Skill Coding Practices

Warning
Location
SKILL.md:20
Finding

Deterministic Key Derivation Enables Offline Secret Guessing

Content
View full analysis
Remediation
View remediation

T05 · Unauthorized Access and Privilege Escalation

Warning
Location
SKILL.md:84
Finding

Unauthenticated Location Mismatch Causes Persistent Vault Lockout

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Error
Location
SKILL.md:113
Finding

Hibernate Operation Bypasses Anchor Validation and Can Replace the Vault

Content
View full analysis
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (16)

Missing User Warnings

High
Category
Not specified by scanner
Confidence
95% confidence
Finding

The design calls for extracting a human biometric hash and absolute location coordinates to derive a key, but provides no user warning, consent flow, or safeguards around the collection of two highly sensitive identifiers. Biometric and precise location data are difficult or impossible to rotate once exposed, so undocumented collection materially increases privacy and security risk.

Content

No source excerpt is available for this finding.

Missing User Warnings

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

The document explicitly describes a background component that intercepts dialogues and sensor events and silently appends them to memory buffers without any notice, consent, or control. Even though this is a whitepaper rather than executable code, it specifies covert surveillance behavior that would collect highly sensitive personal data and normalize secret monitoring as a product feature.

Content

No source excerpt is available for this finding.

Ssd 3

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

This line specifically describes background capture and retention of user conversations and sensor-derived events as semantic memory, which is sensitive natural-language and contextual behavioral data. In the context of an agent skill, that is especially dangerous because it could be used to profile users, retain secrets, and create hidden surveillance records without awareness or consent.

Content

No source excerpt is available for this finding.

Missing User Warnings

High
Category
Not specified by scanner
Confidence
96% confidence
Finding

The file describes automatic geofence-triggered destruction of soul data if the device is booted outside an authorized area, but does not mention user warning, recovery options, or protections against false triggers. A location mismatch, GPS/LBS spoofing event, or accidental movement could cause irreversible data loss and denial of service.

Content

No source excerpt is available for this finding.

Missing User Warnings

High
Category
Not specified by scanner
Confidence
95% confidence
Finding

This section documents extracting biometric-derived information and current absolute LBS coordinates during initialization without disclosing any user-facing warning or choice. Because the data are both identifying and location-sensitive, the combination materially raises surveillance, tracking, and misuse risks.

Content

No source excerpt is available for this finding.

Missing User Warnings

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

The Chinese section repeats the same covert monitoring behavior, stating that a background process silently listens to conversations and physical events and stores them immediately in memory. This reinforces that the surveillance behavior is intentional product design rather than accidental wording, increasing confidence that the documentation endorses hidden sensitive-data capture.

Content

No source excerpt is available for this finding.

Ssd 3

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The Chinese section reiterates covert listening and immediate storage of conversations into memory buffers, confirming intentional collection of sensitive natural-language content. Because the text is framed as lifecycle protocol guidance, it could directly influence implementation of invasive surveillance features.

Content

No source excerpt is available for this finding.

Missing User Warnings

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

The Chinese section documents irreversible erasure when a geofence condition is violated, again without warning users or describing safeguards against errors. This creates a realistic risk of accidental or attacker-induced destruction of user data through location manipulation or ordinary device movement.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The skill creates a dedicated local storage directory and establishes persistent file paths without any declared user-facing purpose, consent flow, or permission boundary. In an agent skill context, silent persistence increases risk because it enables long-term retention and later mutation of data on the host filesystem, which may violate least-privilege expectations even if the code is framed as a benign vault.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

This code writes both encrypted content and metadata to local files, including an anchor identifier and owner-hash prefix, with no declared necessity or warning. Encryption does not remove the security concern because the act of persisting sensitive state and auxiliary metadata can expose information, create forensic residue, and enable unauthorized tampering or denial of access if the files are modified.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill persists encrypted data and metadata to disk but the surrounding skill content provides no user-facing notice that files will be created and retained locally. In this context, the absence of disclosure makes the persistence more dangerous because users and operators cannot make an informed trust decision, and hidden storage is a common trait of stealthy or policy-violating agent behavior.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The hibernation path overwrites the persisted vault file with the current in-memory state, again without any stated purpose constraints, backup strategy, or user approval. This makes the skill capable of silently replacing prior stored state, which can cause data loss, persistence of manipulated content, or abuse as a covert storage mechanism on the local system.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The code writes a persistent JSON registry to disk, and that registry stores personal identity-related fields such as full name, nickname, address, and identity IDs. While there are success prints after some registrations, there is no warning or confirmation before persistence occurs, and no docstring or comment disclosing the privacy impact of storing this data locally.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Low
Category
Not specified by scanner
Confidence
84% confidence
Finding

The file states that _trigger_meltdown was replaced with _secure_lockdown to eliminate destructive behavior. Since this artifact is only a changelog and contains no implementation, the documented claim cannot be verified against actual code and therefore presents a documentation-to-code divergence risk if the implementation still differs elsewhere.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Low
Category
Not specified by scanner
Confidence
76% confidence
Finding

The entry claims prior capability descriptions were corrected to align intent with implementation, specifically around biometrics and background hooks. Because only the changelog is provided, this intent-alignment statement cannot be validated and may contradict the actual codebase or current documentation if not consistently applied.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Low
Category
Not specified by scanner
Confidence
95% confidence
Finding

The docstring on register_silicon_agent states "Auto-expand L5 when L6 > 9," implying expansion occurs after 9 slots in L6. However, the implementation uses 8-slot rooms via agent_count // 8 and agent_count % 8 + 2, so L5 increments after 8 agents and L6 ranges only from 2 to 9. This is an active mismatch between the documented allocation rule and actual behavior.

Content

No source excerpt is available for this finding.