Back to skill

Security audit

Enfp Coach

Security checks for vulnerabilities and agentic risk

Overview

This is a coaching skill with no executable payload, but it has inconsistent ENFP/INTP identity metadata and asks to automatically retain personal coaching records without clear consent or deletion controls.

Review before installing. The skill does not contain executable code or evidence of exfiltration, but users should be aware that it may cause an agent to store personal coaching details across sessions. Install only if the ENFP/INTP identity mismatch is corrected and persistence is made explicit, opt-in, reviewable, and deletable.

Vulnerability Patterns
  • 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
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
Findings (2)

T09 · Insecure Skill Coding Practices

Warning
Location
SKILL.md:323
Finding

Automatic Retention of Potentially Sensitive Coaching Data Without Privacy Controls

Content
View full analysis

Vulnerability Details

File Location: SKILL.md:323-334, with related instructions at SKILL.md:99, SKILL.md:259, SKILL.md:416, and SKILL.md:431-459; also advertised at README.md:50
Vulnerability Type: Uncontrolled retention of potentially sensitive user data
Risk Level: Medium

Relevant instruction excerpt translated into English:

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

**Recorded content:**
- The user's core problems, such as confusion, obstacles, and goals
- The user's areas of interest
- The user's action records, including intended and completed actions
- The user's key insights from conversations

**Usage scenarios:**
- Review previous progress during the next conversation
- Identify patterns in the user's behavior
- Adjust coaching strategies

Related instructions require the Agent to save conversations and actions for later sessions, record insights in a user profile, and archive the user's problems, insights, action plans, and execution results. README.md:50 describes this behavior as automatic.

Technical Analysis

The Skill instructs the Agent to create a persistent behavioral profile containing personal problems, interests, plans, insights, and execution history. In a coaching context, these records may include mental-health concerns, relationship information, career circumstances, and other sensitive personal details.

No consent mechanism, storage boundary, retention period, deletion procedure, data-minimization rule, access-control requirement, or user-isolation requirement is defined. The instructions also do not distinguish temporary conversational context from persistent memory.

The package contains no storage implementation, so the audit cannot establish where the information would be written. The vulnerability arises when a host Agent follows these instructions using an available memory or profile facility. This is an insecure data-handling desi ...[truncated 1262 chars]

Remediation
View remediation

Remediation Suggestions

  1. Make all cross-session storage explicitly opt-in rather than automatic.
  2. Before saving anything, disclose what fields will be stored, why they are needed, where they are stored, and how long they will be retained.
  3. Default to ephemeral conversational context when the user has not consented to persistence.
  4. Store concise summaries instead of complete conversations, and exclude mental-health, relationship, identity, financial, and other sensitive details unless strictly necessary.
  5. Require per-user isolation and prohibit reading or writing another user's profile.
  6. Add commands or documented procedures to inspect, correct, export, and delete stored information.
  7. Define a short retention period and automatic expiration policy.
  8. Require encryption and access controls from any host memory implementation.
  9. Update README.md so that it does not describe storage as automatic.
  10. Clearly state that the Skill itself does not provide storage and that persistence must not occur unless the host supports the required privacy controls.

other

Note
Location
clawhub.json:2
Finding

Conflicting Package Identity, Version, and Repository Metadata

Content
View full analysis

Vulnerability Details

File Location: clawhub.json:2-12, with conflicting version information at SKILL.md:3
Vulnerability Type: Package metadata integrity and provenance failure
Risk Level: Low

json
"name": "enfp-coach",
"version": "1.0.0",
"tags": ["intp", "coaching", "growth", "career", "learning", "analysis-paralysis"],
"repository": {
  "type": "git",
  "url": "https://github.com/lj22503/women-awakening-museum.git"
}

The main Skill declares a different version:

yaml
version: 1.6.0

Technical Analysis

The package is named enfp-coach, but its tags identify it as an INTP Skill. The README and one reference file are also INTP-oriented, while the main Skill is ENFP-oriented. In addition, clawhub.json declares version 1.0.0, whereas SKILL.md declares version 1.6.0.

The repository URL points to a project whose name is unrelated to the coaching package. This prevents users and reviewers from reliably establishing source provenance, matching the installed artifact to a reviewed release, or determining which documentation describes the intended behavior.

The repository field is passive metadata. The project contains no instruction or code that downloads or executes content from that URL. Therefore, this finding is not classified as remote payload retrieval or an insecure dependency.

Attack Path

  1. A user or reviewer examines clawhub.json to identify the package source and intended personality type.
  2. The conflicting ENFP and INTP metadata creates uncertainty about which behavior and reference material are authoritative.
  3. The reviewer follows the repository URL expecting to inspect the source of this package.
  4. The unrelated repository prevents reliable source verification and could cause review, issue reporting, or update decisions to be made against the wrong project.
  5. If package distribution later relies on this metadata, version mismatches could als ...[truncated 560 chars]
Remediation
View remediation

Remediation Suggestions

  1. Choose a single Skill identity and consistently use either ENFP or INTP across SKILL.md, README.md, references, keywords, and tags.
  2. Synchronize the version in clawhub.json with the version declared in SKILL.md.
  3. Replace the repository URL with the actual source repository for this package, or remove the field until that repository exists.
  4. Remove unrelated reference files from the published artifact or clearly label them as non-operative comparative material.
  5. Add release tags or commit identifiers so users can match a package artifact to an immutable reviewed source revision.
  6. Add automated package-validation checks that reject inconsistent names, versions, tags, documentation, and repository metadata before publication.
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 (19)

Description-Behavior Mismatch

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

The documented behavior explicitly targets INTP users and INTP-specific coaching despite the declared ENFP-only scope. In a skill-routing system, this can lead to inappropriate activation, incorrect personalization, and user harm from advice framed around the wrong behavioral assumptions.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The README documents an INTP-focused coach while the manifest says the skill is for ENFP users only. This identity/scope mismatch can cause the agent to trigger in the wrong contexts, give guidance based on the wrong persona model, and undermine downstream safety controls or routing assumptions tied to the declared skill scope.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

The file is clearly scoped to INTP coaching while the skill metadata declares an ENFP coaching skill. This mismatch can cause the agent to apply the wrong personality guidance to users, leading to systematically incorrect advice, broken routing assumptions, and potential harm in growth or career guidance contexts where personalization is the core function.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The README states that conversations and actions are saved automatically, but it does not clearly disclose retention, purpose, consent, storage location, or deletion controls. For a coaching skill that may collect sensitive personal or mental-health-adjacent information, silent persistence creates material privacy and trust risks.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The manifest restricts the skill to ENFP users, but the documentation later invites non-ENFP users to use it. This inconsistency can cause unintended activation, inappropriate advice based on mismatched assumptions, and policy bypass of the declared scope boundary.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The manifest restricts the skill to ENFP users, but the documentation later invites non-ENFP users to use it. This inconsistency can cause unintended activation, inappropriate advice based on mismatched assumptions, and policy bypass of the declared scope boundary.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The file contains contradictory trigger and non-trigger rules: the manifest says vague emotional input without ENFP context should not trigger, while later instructions tell the skill how to respond to exactly those inputs. Contradictory routing rules are dangerous because they undermine reliable safety gating and may pull the skill into quasi-mental-health interactions outside its intended scope.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The manifest states non-ENFP users should not trigger the skill, but later sections explicitly welcome non-ENFP participation. That contradiction weakens enforcement of audience restrictions and increases the risk of the skill being used beyond validated assumptions about user needs and appropriateness.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
98% confidence
Finding

The skill describes saving conversation and action-history data in a user profile without any clear privacy notice, retention explanation, or consent flow. Because the subject matter includes emotional state, goals, and behavioral patterns, silent collection increases privacy risk and user surprise.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
99% confidence
Finding

These instructions require recording personal profile data such as core problems, interests, actions, and insights, but they do not mention consent or sensitivity boundaries. Persisting this type of reflective and behavioral data without notice creates a meaningful privacy and trust risk.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
99% confidence
Finding

The skill mandates persistent logging of user conversation content and action history for future reuse, which materially increases privacy exposure and the blast radius of any later misuse or compromise. This is more concerning in a coaching context because the stored data can reveal emotional struggles, goals, habits, and inferred personality traits.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
99% confidence
Finding

The skill repeatedly instructs saving and recalling cross-session dialogue details, insights, plans, and outcomes without consent or sensitivity limits. Cross-session memory of intimate coaching content can enable unwanted profiling and amplifies harm if data is exposed, reused out of context, or retained longer than expected.

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 personal information automatically after the session, again without notice or permission. Silent post-conversation archiving is especially risky because users may reasonably believe a supportive chat is ephemeral rather than persisted for future profiling.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The provided skill context says this skill should trigger for users who explicitly identify as ENFP and need ENFP-specific growth/career/study guidance. However, the manifest keywords and tags use "INTP"/"intp" and the description is generic coaching text, which mismatches the declared ENFP scope and would misrepresent the skill's actual intended behavior to discovery/routing systems.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

Intent documentation in the manifest should align with the skill's stated purpose, but the keywords and tags actively point to a different MBTI type. This is not merely incomplete metadata; it contradicts the declared ENFP intent and could cause the skill to be surfaced in the wrong 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 reference is entirely written in Chinese and explicitly marked as internal guidance for the skill during execution. That creates a natural-language policy concern because the skill may implicitly steer responses toward a specific language or locale without documenting user choice or opt-in.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The content is written entirely in Chinese and assumes a single locale without indicating fallback behavior or user language selection. This can degrade safety and usability by causing misunderstanding, incorrect coaching delivery, or exclusion of users whose preferred language differs from the hardcoded content language.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
72% confidence
Finding

All user-facing instructions, examples, and descriptions are written exclusively in Chinese, and the document does not state that the skill is region-specific or that users may choose another language. Under the language/locale policy, forcing a specific language without opt-in can be a natural-language policy concern.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Low
Category
Not specified by scanner
Confidence
97% confidence
Finding

The file is presented throughout as ENFP background knowledge for the enfp-coach skill, but the footer says the document location is intp-coach/references/enfp-traits.md. This is an active documentation contradiction about which skill/package the content belongs to, not merely an omission.

Content

No source excerpt is available for this finding.