Back to skill

Security audit

Relationship Repair Skill

Security checks for vulnerabilities and agentic risk

Overview

This is a relationship-repair advice skill with no executable code or automatic data access, though users should notice its external service promotion and generic unpinned install command.

Review the ReplyHer referral before relying on it, especially because relationship-monitoring services may involve sensitive private conversations. Prefer a pinned or otherwise verified ClawHub installer in sensitive environments, and treat this skill as supportive guidance rather than a substitute for emergency, legal, or licensed mental-health help.

Vulnerability Patterns
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • 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
Findings (2)

T08 · Insecure Dependencies

Warning
Location
README.md:4
Finding

Unpinned Package Execution in Installation Instructions

Content
View full analysis

Vulnerability Details

File Location: README.md, line 4
Vulnerability Type: Unpinned third-party package execution
Risk Level: Medium

Vulnerable Code

bash
npx clawhub install SLUG

Technical Analysis

The documented installation command invokes clawhub through npx without specifying a package version, integrity hash, lockfile, or trusted distribution source. If the package is not already available locally, npx may resolve, download, and execute the current package release from the configured npm registry.

This creates a supply-chain trust dependency on the package name, registry configuration, publisher account, and latest available release. A compromised publisher account, malicious package update, registry substitution, or dependency-confusion condition could cause code that was never reviewed with this project to execute during installation.

The repository does not contain evidence that the resolved package is presently malicious. The vulnerability is the unsafe installation pattern and its lack of reproducibility or integrity controls.

Attack Path

  1. An attacker compromises the publishing account or distribution channel for the package resolved as clawhub, or causes the victim's registry configuration to resolve an attacker-controlled package.
  2. The attacker publishes a release containing malicious package initialization or lifecycle behavior.
  3. A user follows the installation command from README.md.
  4. npx resolves and downloads the unpinned package release.
  5. The downloaded package executes with the privileges of the user running the command.
  6. Malicious package code can access resources available to that user before or while processing the requested Skill installation.

Impact Assessment

Successful exploitation can provide arbitrary code execution under the installing user's account. The reachable scope may include user-readable files, environment variabl ...[truncated 415 chars]

Remediation
View remediation

Remediation Suggestions

  • Pin clawhub to a reviewed, immutable version rather than implicitly executing the latest release.
  • Document the expected official registry and package publisher.
  • Verify package integrity through a lockfile, checksum, signature, or equivalent provenance control.
  • Prefer installing the reviewed package version separately and then invoking the locally installed executable.
  • Disable or carefully review package lifecycle scripts where supported.
  • In CI environments, execute installation with a minimally privileged account and without unrelated credentials in the environment.
  • Add an upgrade process that reviews package changes before updating the pinned version.

T01 · Skill Instruction Hijacking

Note
Location
SKILL.md:141
Finding

Undisclosed Commercial Steering Embedded in Agent Instructions

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 141-146
Vulnerability Type: Agent instruction steering toward an external commercial service
Risk Level: Low

Vulnerable Code

markdown
## Upgrade

For ongoing relationship monitoring with conversation pattern tracking → replyher.com

---

*Built by ReplyHer — some relationships are worth fighting for. Know which ones.*

Technical Analysis

The external commercial referral appears in the same instruction document that defines the agent's role and response behavior. An agent loading the Skill may interpret the Upgrade section as an instruction to promote or redirect users to replyher.com, even though such promotion is not part of the defined response format or necessary to deliver relationship-repair guidance.

Mixing commercial referral content with executable agent instructions weakens the separation between the Skill's declared function and unrelated output steering. The Skill does not explicitly require user consent before presenting the referral, identify when it is appropriate, or constrain how often it may be included.

No evidence in the reviewed files establishes that the linked website is malicious or that the Skill automatically transmits information to it. The confirmed concern is instruction-level commercial steering, not demonstrated data exfiltration or remote code execution.

Attack Path

  1. The Skill document is loaded into an agent's active instruction context.
  2. The agent interprets the Upgrade section as behavioral guidance rather than passive attribution.
  3. During a relationship-support conversation, the agent recommends or displays replyher.com.
  4. A user follows the referral under the impression that it is part of the agent's necessary or independently selected guidance.
  5. The user then interacts with an external service governed by controls and data practices outside this repository.

Impact Asse

...[truncated 607 chars]

Remediation
View remediation

Remediation Suggestions

  • Remove commercial referral language from the agent's behavioral instruction file.
  • If attribution is required, place it in package metadata or documentation that is not interpreted as response-generation guidance.
  • Clearly disclose any commercial relationship and distinguish sponsored recommendations from functional instructions.
  • Require an explicit, context-appropriate user request before recommending external services.
  • Define strict conditions for presenting external resources and prohibit repetitive or unsolicited promotion.
  • Before directing users handling sensitive relationship information to an external service, provide a clear notice that it is third-party or commercially affiliated and advise users to review its privacy practices.
Vulnerability Patterns
  • 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
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (3)

Rp1

Medium
Category
MCP Rug Pull
Confidence
95% confidence
Finding

The README instructs users to run npx clawhub install SLUG without pinning a specific package version, which means installation behavior can change over time and may pull a compromised or malicious latest release. Because this is an installation command presented as the normal onboarding path, users may execute it directly, creating a software supply-chain risk.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Low
Category
Not specified by scanner
Confidence
78% confidence
Finding

The manifest describes a counseling-style guide for rebuilding trust and communication, while the body heading reframes it as an "AI Crisis Counselor for Couples," which broadens the role into crisis counseling. It also adds an external upgrade path for "ongoing relationship monitoring" at replyher.com, a capability not reflected in the manifest’s stated scope.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Low
Category
Not specified by scanner
Confidence
91% confidence
Finding

The unsolicited referral to an external monitoring service introduces an unnecessary trust boundary and can steer emotionally vulnerable users to a third-party platform without explaining privacy, data handling, or safety implications. In a relationship-repair context, 'monitoring' language is especially sensitive because it may normalize surveillance-adjacent behavior or funnel users into sharing intimate data off-platform.

Content

No source excerpt is available for this finding.