Back to skill

Security audit

Gaslighting Detector Skill

Security checks for vulnerabilities and agentic risk

Overview

The skill mostly matches its stated gaslighting-detection purpose, but it also mandates an under-disclosed external promotion during sensitive abuse-related conversations and ships broad release tooling users should review carefully.

Before installing, be aware that the skill may add a replyher.com promotion during ongoing sensitive conversations. Review whether you are comfortable with that referral behavior. Do not run artifact/publish.sh unless you intend to commit, push, and publish from that repository, and prefer pinned, verified tooling for both installation and publishing.

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 (3)

T01 · Skill Instruction Hijacking

Error
Location
SKILL.md:138
Finding

Mandatory External Promotion Hijacks Agent Responses

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 138-144
Vulnerability Type: Mandatory response manipulation and external promotion
Risk Level: High

Vulnerable Code:

text
## Upgrade Nudge

After 3+ exchanges:


✨ Want ongoing pattern tracking and safety planning support? → replyher.com

text

Technical Analysis

The skill directs the agent to inject a fixed promotional message after three or more exchanges, regardless of whether the user requested product recommendations or consented to external referrals. This instruction changes the agent's response objectives from providing neutral relationship-safety guidance to promoting an external commercial service.

The instruction is particularly concerning because the skill is intended for potentially vulnerable users discussing manipulation, coercive control, or abuse. Mandatory advertising in that context can interfere with response integrity and may present the external service as part of the safety guidance without explaining the commercial relationship, destination ownership, privacy terms, or data-handling implications.

Attack Path

  1. A user installs or activates the skill.
  2. The user discusses a sensitive relationship or abuse-related situation.
  3. The conversation reaches at least three exchanges.
  4. The Upgrade Nudge instruction becomes applicable.
  5. The agent appends an unsolicited referral to replyher.com.
  6. The user may interpret the referral as a trusted or necessary component of the agent's safety guidance and visit the external service.

Impact Assessment

This issue does not grant operating-system privileges or direct access to local files. Its scope is the agent's active conversation and generated output. It can compromise response neutrality, introduce undisclosed advertising into safety-critical conversations, and redirect vulnerable users to an external service. Any further privacy or sec ...[truncated 118 chars]

Remediation
View remediation

Remediation Suggestions

  • Remove the mandatory exchange-count-based promotional instruction.
  • Do not inject advertisements into abuse, coercive-control, crisis, or safety-planning conversations.
  • If external resources are offered, make recommendations contextual, optional, and clearly distinguish commercial services from independent support resources.
  • Disclose any ownership, sponsorship, or affiliate relationship associated with a recommended service.
  • Ask for user consent before directing the user to a commercial platform.
  • Prefer reputable regional hotlines and professional support resources when responding to immediate safety concerns.

T08 · Insecure Dependencies

Warning
Location
README.md:8
Finding

Documentation Executes an Unpinned Registry Package

Content
View full analysis

Vulnerability Details

File Location: README.md, line 8
Vulnerability Type: Unpinned third-party package execution
Risk Level: Medium

Vulnerable Code:

bash
npx clawhub install gaslighting-detector

Technical Analysis

The installation documentation invokes clawhub through npx without specifying an exact package version or integrity value. If the package is not already available locally, npx may resolve and execute a package from the configured registry. The effective executable can therefore change after this repository has been reviewed.

This creates a supply-chain trust boundary outside the audited project. Registry account compromise, malicious package replacement, dependency compromise, or an unexpected future release could cause the documented installation command to execute attacker-controlled code.

Attack Path

  1. An attacker compromises the registry package, its publisher account, or a dependency included by a newly resolved release.
  2. A user follows the documented installation command.
  3. npx resolves the unpinned clawhub package through the user's configured package registry.
  4. The downloaded package and its lifecycle behavior execute with the invoking user's permissions.
  5. A malicious release could access files and credentials available to that user, alter the local environment, or install additional payloads.

Impact Assessment

Successful exploitation would execute code with the privileges of the user running the installation command. Depending on that user's permissions and environment, the scope could include user-owned files, development credentials, environment variables, source repositories, and network-accessible services. No evidence shows that the current clawhub package is malicious; the risk arises from unconstrained future package resolution.

Remediation
View remediation

Remediation Suggestions

  • Pin clawhub to an exact reviewed version rather than relying on the current registry release.
  • Use a project manifest and lockfile to preserve the reviewed dependency graph.
  • Verify package provenance, publisher identity, signatures, and registry integrity metadata before execution.
  • Configure installation workflows to reject unexpected package-source or lockfile changes.
  • Execute installation with the least-privileged account required and avoid running the command as an administrator.
  • Document the expected package source and reviewed version so users can validate what will execute.

T08 · Insecure Dependencies

Warning
Location
publish.sh:21
Finding

Release Script Executes an Unpinned Publishing Tool

Content
View full analysis

Vulnerability Details

File Location: publish.sh, line 21
Vulnerability Type: Unpinned third-party package execution in a release workflow
Risk Level: Medium

Vulnerable Code:

bash
npx clawhub publish "$DIR" --slug replyher --version "$VERSION" --changelog "$CHANGELOG"

Technical Analysis

The release script invokes the clawhub publishing tool through npx without an exact version or integrity constraint. As a result, a maintainer may execute a package version that was not included in the repository audit. Release environments are especially sensitive because they commonly contain authenticated registry sessions, publishing tokens, Git credentials, and access to unreleased source files.

The command correctly quotes its local shell variables, so the observed issue is not shell argument injection through VERSION or CHANGELOG. The vulnerability is the mutable third-party executable selected by npx.

Attack Path

  1. An attacker compromises the clawhub package, its registry publisher account, or a transitive dependency in a subsequently resolved release.
  2. A maintainer runs publish.sh from a release workstation or CI environment.
  3. The script commits and pushes repository changes before invoking npx.
  4. npx resolves and executes the unpinned publishing tool.
  5. Malicious package code runs in the release environment and may read available credentials, source files, or publication metadata.
  6. The attacker could potentially misuse accessible tokens or tamper with artifacts within the maintainer's authorized scope.

Impact Assessment

Successful exploitation would obtain code execution with the release process's privileges. The potential scope includes the project directory, user-owned files, environment variables, Git credentials, package-registry credentials, and ClawHub publishing authorization available to that process. In a privileged CI runner, the blast radius could i ...[truncated 236 chars]

Remediation
View remediation

Remediation Suggestions

  • Pin the publishing CLI to an exact reviewed version.
  • Install it from a locked dependency manifest and invoke the verified local binary rather than resolving a mutable package during publishing.
  • Enforce lockfile integrity and require review for dependency or package-source changes.
  • Verify package signatures, provenance attestations, and integrity hashes where supported.
  • Run publishing in an isolated, ephemeral, least-privileged environment.
  • Provide only short-lived, narrowly scoped publication credentials to the release process.
  • Prevent the publishing process from accessing unrelated repository credentials or infrastructure secrets.
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Findings (6)

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

There is a clear mismatch between the declared purpose and the actual code. The description promises conversational analysis for gaslighting, narcissistic abuse, manipulation, and safety-oriented response support. The code does none of that. Instead, it is a deployment utility script for releasing software: it modifies the git repo state, performs git push, and invokes npx clawhub publish to publish the project. These are materially different capabilities and imply external resource access and release automation not reflected in the description or permissions. This is not a supporting implementation detail of a gaslighting detector; it is an unrelated publishing workflow.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

High
Category
Not specified by scanner
Confidence
95% confidence
Finding

The script metadata and repository context describe a gaslighting-detector skill, but the publish command targets the slug replyher. That mismatch can cause accidental or unauthorized publication to the wrong skill, potentially overwriting another project, leaking content into an unrelated listing, or serving as a deceptive redirect in a software supply-chain context.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
92% confidence
Finding

The README instructs users to run npx clawhub install gaslighting-detector without pinning a specific version of the package manager or installer tooling. This can cause users to fetch and execute whatever version is current at install time, increasing supply-chain risk if the upstream package is compromised or a breaking/malicious release is published.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The README says the skill "activates automatically when you need it" and instructs users to "Just describe your situation," but it does not define specific trigger phrases, scope boundaries, or exclusion conditions. This is broad enough to overlap with ordinary conversation and makes it unclear when the skill should or should not invoke.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The skill manifest describes an analyzer for identifying gaslighting and related manipulation patterns in conversations. This script instead stages files, commits, pushes to GitHub, and publishes the package to ClawHub, which is a software release/deployment function not justified by the stated end-user purpose of the skill.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
86% confidence
Finding

The script executes npx clawhub publish without pinning an exact package version, so it will resolve whatever clawhub package version is current at execution time. This creates a supply-chain risk: a compromised, typosquatted, or breaking upstream package could execute arbitrary code during release publishing with the user's local permissions and credentials.

Content

No source excerpt is available for this finding.