Back to skill

Security audit

Estp Coach

Security checks for vulnerabilities and agentic risk

Overview

This coaching skill has no executable payload, but it needs review because its identity is inconsistent, its install command points to a different package, and it automatically stores personal coaching history without clear controls.

Review this package carefully before installing. It appears to be a coaching prompt package rather than malware, but the package identity should be corrected, the install command should match the audited package, and persistent coaching records should be opt-in with clear ways to review and delete stored data.

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

T08 · Insecure Dependencies

Warning
Location
README.md:19
Finding

Installation Instructions Resolve to a Different Skill Package

Content
View full analysis

Vulnerability Details

File Location: README.md:19-23
Vulnerability Type: Supply-chain package substitution
Risk Level: Medium

Vulnerable Code

markdown
### Installation

```bash
clawhub install intp-coach
text

### Technical Analysis

The audited package identifies itself as `estp-coach` in both `SKILL.md` and `clawhub.json`, but the documented installation command requests `intp-coach`. As a result, following the official installation instructions does not install the package that was audited.

Package installation is a trust-boundary operation. The behavior and permissions of the package resolved under `intp-coach` cannot be inferred from this repository. If that registry name is controlled by another publisher, compromised, or later transferred, it could distribute arbitrary instructions or executable components under the trust users place in this project.

### Attack Path

1. A user obtains this repository and trusts its README.
2. The user runs `clawhub install intp-coach`.
3. The package registry resolves a package other than the audited `estp-coach` project.
4. The unrelated package is installed and loaded with the permissions normally granted to a Skill.
5. Any malicious instructions, tools, or scripts in that external package execute within the permissions of the host Agent.

This path depends on the contents and ownership of the externally resolved package; the audit did not establish that the current `intp-coach` package is malicious.

### Impact Assessment

The immediate impact is installation of unintended third-party content. The ultimate privilege and scope depend on the host's Skill permission model and the installed package's capabilities. Potential exposure can include Agent session context, enabled tools, accessible files, network access, or other permissions granted to installed Skills. This repository itself contains no executable malicious payload.
Remediation
View remediation

Remediation Suggestions

  1. Replace the command with:
    bash
    clawhub install estp-coach
    
  2. Pin the package to a verified version or integrity digest where the registry supports it.
  3. Add an automated release check that compares the installation target with the name field in clawhub.json.
  4. Verify package ownership and publisher identity before publishing installation instructions.
  5. Correct the README's remaining INTP references so users can reliably identify the package they are installing.

other

Warning
Location
README.md:48
Finding

Automatic Persistence of Sensitive Coaching Profiles Without Privacy Controls

Content
View full analysis

Vulnerability Details

File Location: README.md:48-51
Vulnerability Type: Privacy-unsafe persistent profiling
Risk Level: Medium

Vulnerable Code

The following source excerpt is translated into English to satisfy the report language requirement:

markdown
| Feature | Description | Command |
|---------|-------------|---------|
| **Growth records** | Save conversations and actions for continuation next time | Execute automatically |

Related persistence instructions also appear in SKILL.md:323-337 and SKILL.md:431-459, where the Skill directs the Agent to retain the user's core problems, interests, insights, action plans, and execution results.

Technical Analysis

The Skill instructs the Agent to create and retain a behavioral profile automatically. The retained information may include personal concerns, career or learning difficulties, inferred behavioral patterns, plans, and reported outcomes.

No consent prompt, storage boundary, retention period, deletion mechanism, access-control requirement, encryption requirement, or data-minimization policy is defined. Persistent storage is not necessary for a single coaching interaction, and automatic retention exceeds what a user may reasonably expect from a conversational response.

This is not classified as T02: Agent Memory Poisoning because the stored content is not described as attacker-controlled rules intended to alter future Agent behavior. It is instead a privacy and data-governance weakness involving sensitive user profiling.

Attack Path

  1. A user discusses personal concerns, behavioral difficulties, goals, or execution results with the Skill.
  2. The Skill automatically records selected details for future sessions.
  3. The profile persists beyond the immediate conversation without explicit informed consent.
  4. A later session, another component sharing the same memory store, or an operator with storage access retrieves the profi ...[truncated 882 chars]
Remediation
View remediation

Remediation Suggestions

  1. Make persistent memory strictly opt-in and obtain informed consent before writing any profile data.
  2. Explain what fields will be stored, why they are needed, where they are stored, and who can access them.
  3. Default to session-only processing when consent is absent.
  4. Store only the minimum necessary summary; do not retain raw conversations or unnecessary psychological inferences.
  5. Define a short retention period and automatically delete expired records.
  6. Provide commands to review, correct, export, and permanently delete stored information.
  7. Require per-user storage isolation, access control, and encryption appropriate to the host environment.
  8. Avoid storing highly sensitive mental-health details unless necessary and explicitly requested by the user.
  9. Update the documentation so it no longer represents profile creation as an automatic operation.
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • 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
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (18)

Description-Behavior Mismatch

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

The README documents an INTP coaching skill while the manifest metadata says this is an ESTP coaching skill. This mismatch can cause the wrong skill to be invoked, reviewed, or trusted under false assumptions, which is especially risky for a coaching workflow that may store personal conversations and give behavior-shaping advice.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

The body of the supposed ESTP reference describes traits and cognitive-function issues that align with a different MBTI profile, including Ti-Ne patterns and analysis paralysis rather than the action-oriented ESTP profile advertised in the manifest. Because this is background knowledge the agent may rely on during responses, the mismatch can produce consistently wrong coaching advice for vulnerable users seeking behavioral guidance.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

The reference file is for INTP coaching while the skill is declared as an ESTP coach, creating a clear skill-data mismatch. This can cause the agent to deliver systematically incorrect personality-specific guidance, weakening safety and reliability and potentially steering users toward unsuitable advice in growth, risk, and decision-making contexts.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The README is entirely written in Chinese and all example interactions assume Chinese output, but there is no statement that users may choose another language or that the language restriction is intentional and justified. This can indicate a language policy violation because the skill appears to force a specific language without user opt-in.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The example prompts are common conversational statements and are not specific enough to constrain activation to deliberate use. In practice, this can make the skill activate on routine chat about confusion, focus, or decision-making, expanding its operational scope beyond what users expect.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The skill relies on broad natural-language activation such as automatic recognition rather than narrowly scoped triggers. That increases the chance of accidental invocation during ordinary conversation, causing unsolicited coaching behavior or unintentional handling of sensitive personal content.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The README states that conversations and actions are saved as growth records, but it does not clearly warn users or describe consent, retention, or deletion behavior. Because this skill handles self-reflection and potentially mental-health-adjacent content, undisclosed persistence increases privacy risk and user harm.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The skill explicitly advertises a 'growth record' that saves dialogue and action history, but it does not provide a user-facing notice about what data is stored, for how long, where it is kept, or how the user can opt out. Because the content includes personal struggles, goals, and behavioral patterns, silent retention creates a meaningful privacy risk and can expose sensitive profiling data if mishandled.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill states that it will preserve conversation and action history for future sessions, which is a direct instruction to retain and reuse user-provided data over time. In a coaching context, that data can reveal intimate habits, vulnerabilities, and decision patterns, making cross-session persistence materially risky without clear consent and safeguards.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

These instructions direct the system to record user profile information such as core problems, interests, actions, and insights without any embedded consent flow or visible disclosure to the user. This is dangerous because it normalizes hidden profiling of potentially sensitive self-improvement and mental-state information, which can later be reused in ways the user did not expect.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The skill instructs creation of a persistent user profile from dialogue content and recurring behavior, including patterns like repeatedly selecting a plan but failing to execute it. This goes beyond transient assistance into behavioral profiling, which increases privacy and misuse risk if the data is exposed, repurposed, or retained indefinitely.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
98% confidence
Finding

The end-of-conversation archiving workflow requires repeated storage of user problems, insights, action plans, and execution results, creating a broad and accumulating personal-data trail. In this coaching setting, the archived material can contain sensitive emotional, professional, and behavioral information, so repeated retention substantially enlarges privacy exposure over time.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The manifest claims this is an ESTP coaching skill, but the metadata keywords and tags describe INTP traits such as analysis paralysis and focus issues. This mismatch can misroute users, downstream agents, or review systems into applying the wrong behavioral guidance, undermining trust and potentially causing harmful advice in a coaching context.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
84% confidence
Finding

This markdown file is entirely written in Chinese and explicitly framed as internal guidance for the skill, with no indication that the user can choose another language or that the skill is intentionally limited to a Chinese-language or region-specific context. Under the language/locale policy rule, forcing a specific language without user opt-in is a natural-language policy concern.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The reference file for an ESTP coaching skill points to a different personality profile/path (intp-coach/references/estp-traits.md), which indicates the skill may be wired to the wrong background material. In a coaching skill, this can systematically deliver incorrect guidance to users, undermining trust and potentially reinforcing the wrong behavioral interventions.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

This reference file presents all coaching guidance, examples, and usage context exclusively in Chinese. Under the policy, forcing a specific language without user opt-in can be a natural-language policy violation unless the locale restriction is explicitly documented and justified, which is not present here.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
84% confidence
Finding

The manifest description is written entirely in Chinese, which can signal a language-specific skill experience without any indication that users may choose another language. Under the policy, language constraints should either offer user opt-in or be clearly documented and justified as locale-specific.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Low
Category
Not specified by scanner
Confidence
93% confidence
Finding

The repository URL points to an apparently unrelated project name, which breaks provenance and makes it harder to verify the source, maintenance, and intent of the skill. In supply-chain or trust-sensitive environments, unrelated repository metadata can conceal abandoned, repurposed, or deceptive packaging.

Content

No source excerpt is available for this finding.