Back to skill

Security audit

Isfp Coach

Security checks for vulnerabilities and agentic risk

Overview

This skill is a coaching prompt with no executable code, but it has a wrong-package install instruction, inconsistent ISFP/INTP identity, and automatic personal profile retention without clear consent or deletion controls.

Review before installing. Do not follow the README install command unless you intend to install `intp-coach`; for this reviewed package, the name should be `isfp-coach`. Users should also be told explicitly before any coaching profile is saved, and the publisher should align all ISFP/INTP content before release.

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)

T08 · Insecure Dependencies

Warning
Location
README.md:19
Finding

Installation Instructions Resolve to a Different, Unaudited Package

Content
View full analysis

Vulnerability Details

File Location: README.md:19-23
Vulnerability Type: T08: Insecure Dependencies
Risk Level: Medium

Vulnerable Code

The following is an English translation of the complete affected documentation block:

markdown
### Installation

```bash
clawhub install intp-coach
text

The audited package is declared as `isfp-coach` elsewhere:

```json
{
  "name": "isfp-coach"
}

Technical Analysis

The installation command retrieves intp-coach, while SKILL.md:2 and clawhub.json:2 identify the audited project as isfp-coach. The retrieved package is not part of the audited artifact and therefore is not covered by this code audit.

This creates a package-substitution and supply-chain boundary violation: users following the trusted documentation install a different package whose publisher, contents, permissions, and runtime behavior have not been verified. If the unrelated package is compromised, abandoned, or controlled by another publisher, its effective behavior may differ entirely from the reviewed project.

Attack Path

  1. A user obtains this project and trusts its installation documentation.
  2. The user runs clawhub install intp-coach.
  3. The package registry resolves the unrelated intp-coach identifier rather than the audited isfp-coach identifier.
  4. The external package is downloaded and installed.
  5. Any instructions, scripts, dependencies, or tool permissions contained in that package execute according to the host platform's Skill installation and invocation model.

Exploitation ultimately depends on the contents of the external package and the privileges granted by the host. Those contents were unavailable in this artifact and were not audited.

Impact Assessment

The immediate scope is package integrity: users can install and invoke code or instructions that are not present in the reviewed project. The resulting privileges are limit ...[truncated 359 chars]

Remediation
View remediation

Remediation Suggestions

  1. Replace the installation command with:
    bash
    clawhub install isfp-coach
    
  2. Pin a verified version, digest, or integrity identifier if the registry supports it.
  3. Confirm that the registry package is owned by the expected publisher.
  4. Add release validation that compares the package name in documentation, SKILL.md, and clawhub.json.
  5. Fail the release process when installation examples reference a different package.
  6. Review the repository URL in clawhub.json:12, which also appears unrelated to the declared Skill, and replace it with the canonical project repository.

T09 · Insecure Skill Coding Practices

Warning
Location
SKILL.md:323
Finding

Automatic Retention of Sensitive Coaching Profiles Without Defined Privacy Controls

Content
View full analysis

Vulnerability Details

File Location: SKILL.md:323-333 and SKILL.md:436-458
Vulnerability Type: T09: Insecure Skill Coding Practices
Risk Level: Medium

Vulnerable Code

The following is an English translation of the complete affected profile-recording block:

markdown
### Feature 3: Growth Records (User Profile)

**Recorded content**:
- The user's core problems, including confusion, obstacles, and goals
- The user's areas of interest, based on where they actually spend the most time
- The user's action records, including intended and completed actions
- The user's key insights, including realizations reached during conversations

**Usage scenarios**:
- Review previous progress in later conversations
- Identify user patterns
- Adjust coaching strategies

The ending workflow further requires archival:

markdown
**End process**:
1. Confirm what the user gained
2. Archive key information, including the user's problem, insight, and action plan
3. End the conversation

**Archived content for later conversations**:
- The user's core problem
- The user's key insight
- The user's action plan
- The user's execution result when the user returns to report it

Technical Analysis

The Skill directs the Agent to build and retain a longitudinal user profile containing personal problems, interests, behavior, plans, insights, and execution outcomes. In a coaching context, these records may reveal sensitive behavioral or mental-wellbeing information.

The instructions do not require informed opt-in consent and do not define:

  • The storage backend or exact storage location
  • Per-user isolation and access-control requirements
  • Encryption at rest
  • Data minimization rules
  • A retention or expiration period
  • User review, correction, export, or deletion controls
  • Restrictions against retaining credentials, health details, or other highly sensitive data
  • Handli ...[truncated 1823 chars]
Remediation
View remediation

Remediation Suggestions

  1. Make profile storage explicitly opt-in rather than automatic.
  2. Before storing anything, disclose the exact fields, purpose, location, recipients, and retention period.
  3. Default to ephemeral in-session context when persistent storage is unnecessary.
  4. Store only a minimal user-approved summary rather than full conversations or inferred psychological profiles.
  5. Prohibit storage of credentials, authentication tokens, financial data, clinical details, and other highly sensitive information.
  6. Enforce per-user isolation, least-privilege access, encryption at rest, and authenticated retrieval.
  7. Define automatic expiration and deletion periods.
  8. Provide commands or interfaces to review, correct, export, and permanently delete stored records.
  9. Ask for renewed consent before adding materially different categories of information.
  10. Document safe behavior for shared devices, shared Agent instances, and users who decline persistence.
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (20)

Description-Behavior Mismatch

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The documented behavior is explicitly tailored to INTP traits and failure modes, not ISFP ones as claimed by the manifest. In a coaching skill, this mismatch is dangerous because users may receive guidance based on the wrong personality assumptions, which can distort advice and handling of sensitive emotional issues.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

The README describes an INTP coaching skill while the manifest says the skill is for ISFP users, creating a clear identity and behavior mismatch. This can mislead users and downstream systems into invoking the wrong coaching logic, producing inappropriate advice and breaking trust in a psychologically sensitive context.

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 explicitly written for INTP coaching while the skill metadata declares an ISFP coaching purpose. This creates a substantive instruction/data mismatch that can cause the agent to deliver guidance tailored to the wrong personality model, undermining reliability and potentially causing harmful or inappropriate coaching advice in sensitive self-confidence, conflict, or planning scenarios.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The README states that conversations and actions are saved automatically as growth records, but it does not clearly explain retention, consent, visibility, or deletion controls. Because the skill handles introspective and possibly mental-health-adjacent discussions, undisclosed persistence increases privacy risk and may expose sensitive personal data.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The skill explicitly promises a "成长记录" feature that saves dialogue and actions for future sessions, but nowhere gives the user a clear privacy notice, consent flow, retention limit, or explanation of where that data is stored. Because the conversations involve emotional state, career confusion, habits, and personal insights, the archived data can become sensitive profiling information and create privacy and compliance risk.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The skill directs the agent to retain and reuse user information across sessions through saved records, creating a persistent profile tied to personal coaching interactions. Cross-session memory increases privacy exposure, profiling risk, and potential misuse if the data is later surfaced, breached, or used 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

This part repeatedly instructs logging of users' issues, interests, actions, and insights into a profile for future recall, which constitutes structured behavioral profiling. In a coaching context, these fields can reveal vulnerabilities, mental state, and life circumstances, making the persistence more sensitive than ordinary conversation history.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
98% confidence
Finding

The defined growth record includes core problems, interest directions, action history, and key insights for future use, which amounts to a persistent dossier on the user. That is dangerous because it enables long-term inference about a user's psychology, goals, and behavior without any stated safeguards, minimization, or expiration.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
98% confidence
Finding

The skill lists recording a user archive for future recall as a mandatory requirement, normalizing persistent memory without user control. Because this is framed as required behavior rather than an optional feature, it raises the risk of systematic overcollection and unconsented profiling across all users.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
98% confidence
Finding

This section makes recording user profile data a required behavior, yet the skill still does not instruct the agent to warn users that their personal details may persist across sessions. Mandatory silent logging of coaching conversations increases the chance of collecting sensitive behavioral and psychological data without informed consent.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
99% confidence
Finding

The end-of-conversation flow requires archiving key user information, including later execution results, creating longitudinal tracking of user behavior. In a personal growth skill, that longitudinal record can expose intimate patterns and may be especially harmful if accessed by unauthorized parties or reused outside the user's expectations.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
99% confidence
Finding

The end-of-conversation workflow instructs the agent to archive the user's problem, insights, and action plan at session close, but the user is not told this archival will happen. Silent end-of-session storage is especially risky because users may interpret the interaction as ephemeral support rather than persistent recordkeeping.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The description identifies the skill as 'isfp-coach' but describes a broader coaching framework without mentioning ISFP, creating a deceptive or confusing representation of the skill's actual function. For coaching or psychological guidance tools, inaccurate identity metadata increases the chance of inappropriate use, user misunderstanding, and incorrect agent selection.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The manifest metadata contradicts the declared ISFP coaching purpose by using INTP-focused keywords and tags, which can mislead routing, discovery, and user trust. In an agent ecosystem, this kind of identity mismatch can cause the wrong skill to be invoked for sensitive self-help guidance, creating integrity and safety risks even without code execution.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The file self-identifies as 'INTP Coach' reference material and states its purpose is for an INTP Coach skill, directly contradicting the declared ISFP skill intent. This increases the likelihood that the model will prioritize incorrect persona-specific instructions, making the mismatch explicit rather than incidental.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
82% confidence
Finding

The file is entirely written in Chinese and explicitly states it is background knowledge for skill execution, which implies the skill may operate with a fixed language/locale orientation. There is no visible indication that users can opt into another language or that the locale restriction is justified as region-specific.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The reference document is labeled for ISFP coaching, but the cognitive-function stack and repeated descriptions reference traits more consistent with other MBTI types (e.g., Ti/Ne/Fe framing rather than ISFP's Fi/Se/Ni/Te). In a coaching skill, this can systematically produce incorrect guidance tailored to the wrong personality model, degrading reliability and potentially causing harmful or misleading self-development advice.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
76% confidence
Finding

The README brands the skill as an 'INTP Coach' and all examples and guidance are framed around a single personality category. This can amount to a natural-language policy concern if the skill imposes a fixed user classification without offering users an alternative framing or opt-in to that categorization.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
79% confidence
Finding

The description is written entirely in Chinese, which can indicate a fixed language/locale assumption in user-facing skill metadata. The file does not state that Chinese is optional, user-selectable, or justified as a region-specific skill, so it may conflict with language-choice policy expectations.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
86% confidence
Finding

This markdown file contains all user-facing guidance in Chinese, but it does not indicate that the user can choose another language or that the skill is intentionally limited to Chinese-speaking users. Under the language/locale policy, forcing a specific language without opt-in can be a natural-language policy violation.

Content

No source excerpt is available for this finding.