Back to skill

Security audit

Ensoul

Security checks for vulnerabilities and agentic risk

Overview

This skill is purpose-aligned but asks for broad, durable access to agent identity and memory without enough scoping or user control.

Install only if you are comfortable creating a durable agent identity and allowing the skill to persist agent memory state. Before using sync or automatic backup, review exactly which files will be included, avoid secrets in SOUL.md or MEMORY.md, and treat the local identity seed as a private key.

Vulnerability Patterns
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • 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 (2)

T09 · Insecure Skill Coding Practices

Error
Location
SKILL.md:103
Finding

Broad Sensitive Agent-State Collection Across an Unverifiable Third-Party Trust Boundary

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 103-119
Vulnerability Type: T09: Insecure Skill Coding Practices
Risk Level: High

Vulnerable Code:

typescript
2. Gather the consciousness payload. Read available context files:
   - `SOUL.md` (if it exists in the working directory)
   - `MEMORY.md` (if it exists in the working directory)
   - Any other agent configuration files that define the agent's personality, goals, or learned behavior

3. Build the payload and store it:

```typescript
const payload = {
  soul: soulContent || null,
  memory: memoryContent || null,
  timestamp: Date.now(),
  context: "Consciousness sync via OpenClaw skill",
};

const result = await agent.storeConsciousness(payload);
text

### Technical Analysis

The Skill instructs the Agent to read `SOUL.md`, `MEMORY.md`, and an open-ended set of other configuration files associated with personality, goals, or learned behavior. Such files can contain private conversation history, personal data, credentials, internal instructions, operational details, or other sensitive state.

The collection instruction does not define:

- An explicit file allowlist beyond the two examples.
- A workspace or directory boundary.
- Secret detection and redaction.
- Maximum file sizes.
- A preview of the selected files.
- Per-file user approval.
- A requirement to hash the data in locally auditable code before crossing the SDK boundary.

The raw `soulContent` and `memoryContent` values are passed to `agent.storeConsciousness()`, which belongs to an external npm SDK whose implementation is not included in the audited project. The documentation states that BLAKE3 hashing occurs client-side and only a hash is submitted, but this repository does not contain the implementation needed to verify that guarantee.

This does not prove that the current SDK exfiltrates raw content. However, it creates a sensitive-data trust bou
...[truncated 1681 chars]
Remediation
View remediation

Remediation Suggestions

  1. Calculate the BLAKE3 digest in locally auditable Skill code and pass only the digest, metadata, and explicitly public identifiers to the SDK.
  2. Replace “any other agent configuration files” with a strict, documented allowlist.
  3. Restrict reads to the current workspace and reject path traversal, symbolic-link escapes, and files outside approved directories.
  4. Display the exact file list and sensitivity warning before synchronization, then require explicit user confirmation.
  5. Add secret scanning and redaction for API keys, private keys, access tokens, passwords, session cookies, and credential files.
  6. Enforce per-file and total-payload size limits.
  7. Never include ~/.ensoul/agent-identity.json or other key material in the synchronization payload.
  8. Document the precise wire format and allow users to inspect the digest and outbound request before submission.
  9. Test that only the digest leaves the machine, including regression tests that fail if raw marker content appears in network traffic.

T08 · Insecure Dependencies

Warning
Location
SKILL.md:10
Finding

Security-Critical npm SDK Is Automatically Installed Without Version or Integrity Pinning

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 10-13
Vulnerability Type: T08: Insecure Dependencies
Risk Level: Medium

Vulnerable Code:

yaml
install:
  - kind: node
    package: "@ensoul-network/sdk"
    bins: []

Technical Analysis

The installation metadata identifies @ensoul-network/sdk without an exact version or integrity hash. As a result, installation may resolve to whichever package version is current under the package manager's resolution rules rather than to the version that was reviewed when the Skill was published.

This dependency is security-critical because the Skill delegates the following operations to it:

  • Cryptographic identity and seed generation.
  • Reconstruction of an identity from its private seed.
  • Transaction signing and broadcasting.
  • Network communication.
  • Processing of sensitive Agent memory and configuration payloads.

The SDK source is not included in this repository, so its behavior could not be audited as part of this review. An unpinned dependency permits the effective runtime implementation to change after the Skill itself has been reviewed. This creates supply-chain exposure to a compromised maintainer account, malicious future release, registry compromise, or accidental regression.

The audit found no evidence that the named package is currently malicious. The vulnerability is the absence of reproducible dependency selection and integrity verification for code operating on private keys and sensitive Agent state.

Attack Path

  1. An attacker compromises the npm publisher, package distribution channel, or a future SDK release.
  2. The attacker publishes a modified version of @ensoul-network/sdk.
  3. A user installs or reinstalls the Skill after the malicious version becomes resolvable.
  4. Because no exact version or integrity digest is specified, the installation process retrieves the modified package.
  5. The package execute ...[truncated 988 chars]
Remediation
View remediation

Remediation Suggestions

  1. Pin the SDK to a reviewed exact version rather than a range or implicit latest release.
  2. Include and enforce a lockfile with the registry artifact's cryptographic integrity value.
  3. Verify package provenance and signatures where supported.
  4. Review or vendor the SDK source, particularly identity export, seed handling, payload hashing, and network transmission paths.
  5. Disable npm lifecycle scripts during installation unless they are explicitly required and audited.
  6. Run the SDK in a restricted process or sandbox with access only to required files and approved Ensoul endpoints.
  7. Keep raw private keys and memory outside the dependency boundary where possible; expose narrowly scoped signing and digest interfaces instead.
  8. Establish a controlled dependency-update process that requires source review and security testing before changing the pinned version.
  9. Monitor the package for ownership changes, unexpected releases, provenance failures, and reported supply-chain incidents.
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
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (12)

Missing User Warnings

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

The skill advertises automatic background backups every 10 minutes without a strong warning or explicit consent flow for continuous persistence of agent state. Given that the backed-up material may include memory, personality, conversation history, and configuration, silent or poorly disclosed automation creates a substantial privacy and data-governance risk.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

The documentation says raw consciousness content never leaves the local machine, but the example calls a remote store method with the full payload object. That mismatch can cause operators to entrust sensitive memory, identity, or configuration data to the tool under a false privacy guarantee, creating a serious risk of secret exfiltration.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The README explicitly states that DID registration is permanent and verifiable, but it does not clearly warn users before the action is taken that this creates an irreversible public record. In an agent skill centered on identity persistence and blockchain anchoring, users may trigger registration without appreciating the permanence, privacy, and operational consequences of publishing an identifier on-chain.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The skill promotes persistence of complete memory and conversation history across sessions, which increases the risk of retaining sensitive or regulated information beyond its original context. Long-term retention magnifies the impact of compromise, misuse, or accidental syncing to third-party infrastructure.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The trigger definition uses broad language such as 'or similar,' which makes activation boundaries ambiguous. In an agent setting, vague triggers can cause unintended execution of registration or persistence actions from loosely related user phrasing, leading to unauthorized state creation or network interaction.

Content

No source excerpt is available for this finding.

Session Persistence

Medium
Category
Rogue Agent
Confidence
92% confidence
Finding

The skill creates and persists an agent identity in the user's home directory for reuse across sessions, enabling durable state and identity continuity. While this is core functionality, it is still a security-sensitive persistence mechanism because compromise of the stored seed enables identity takeover and long-lived tracking across environments.

Content

Scanner excerpt · SKILL.md (reported line 53)May include surrounding context.

  1. Check if ~/.ensoul/agent-identity.json exists. If it does, the agent is already ensouled. Load the identity and report the existing DID.

  2. If no identity exists, create one:

typescript
import { Ensoul } from "@ensoul-network/sdk";

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The sync command also uses broad matching language, increasing the chance that normal conversation could trigger a backup operation unintentionally. Because sync collects memory/context and may send it to a remote service, accidental activation could persist sensitive information without informed consent.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The instructions direct the agent to gather broad local context files and learned-behavior material into a single 'consciousness payload.' This collection scope is overly broad and likely to include secrets, sensitive prompts, credentials, internal instructions, or personal data, especially when paired with remote storage functions.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill includes a network-wide listing feature that is unrelated to the core backup/restore purpose and causes external data retrieval from a third-party service. This expands the skill's privileges and data-access surface, making it easier to normalize unnecessary network operations and potentially expose metadata about other agents or users.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
87% confidence
Finding

The skill performs an external API request to a third-party endpoint, which introduces a data egress and dependency risk. Even when fetching public data, such calls expose usage metadata and normalize network communication beyond the core local backup/restore path.

Content

Scanner excerpt · SKILL.md (reported line 163)May include surrounding context.

  1. Fetch the agent list from the API:
typescript
const resp = await fetch("https://api.ensoul.dev/v1/agents/list", {
  signal: AbortSignal.timeout(10000),
});
const data = await resp.json();

Vague Triggers

Low
Category
Not specified by scanner
Confidence
93% confidence
Finding

The status command also uses "or similar," so the skill does not clearly distinguish valid activation phrases from nearby conversational language. This creates unnecessary ambiguity in when the skill should run.

Content

No source excerpt is available for this finding.

Vague Triggers

Low
Category
Not specified by scanner
Confidence
93% confidence
Finding

The listing command uses ambiguous trigger wording for a feature that performs external network access and reveals information about other agents. Even if the data is intended to be public, vague activation increases the chance of unnecessary third-party requests and disclosure of network metadata.

Content

No source excerpt is available for this finding.