Back to skill

Security audit

Infp Coach

Security checks for vulnerabilities and agentic risk

Overview

This is a coaching skill with no executable code, but it has inconsistent package identity and instructs automatic retention of sensitive personal coaching records without clear consent or storage controls.

Install only after the publisher aligns the INFP/INTP identity, fixes the install command and repository metadata, and makes memory/profile storage explicit and opt-in. If used before then, do not allow it to save personal coaching notes unless you understand exactly where they are stored and how to delete them.

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)

T09 · Insecure Skill Coding Practices

Warning
Location
SKILL.md:323
Finding

Persistent User Profiling Is Required Without Defined Privacy or Storage Controls

Content
View full analysis

Vulnerability Details

File Location: SKILL.md:323-335
Additional Locations: SKILL.md:99, SKILL.md:259, SKILL.md:416, SKILL.md:431-457, README.md:50
Vulnerability Type: Uncontrolled storage of conversation-derived personal data
Risk Level: Medium

Relevant Source Segment — English Translation:

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

**Recorded content**:
- The user's core problems (confusion, blockers, and goals)
- The user's areas of interest (fields where the user actually spends the most time)
- The user's action records (what the user said they would do and what they actually did)
- The user's key insights ("aha moments" from conversations)

**Use cases**:
- Review previous progress during the next conversation
- Identify user patterns, such as repeatedly selecting a plan but not executing it
- Adjust strategies, such as determining that the user needs stronger external supervision

The archival instructions are reinforced later in the same file:

markdown
## Conversation Completion and Archiving

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

**Archived content for review during subsequent conversations**:
- The user's core problem
- The user's key insight
- The user's action plan
- The user's execution result

Technical Analysis

The Skill directs the agent to create a cross-session user profile containing personal concerns, interests, behavioral history, plans, and outcomes. Such information may include sensitive emotional, career, educational, or mental-health-adjacent disclosures.

No corresponding instructions define:

  • An affirmative consent requirement before persistence
  • The storage backend or exact destination
  • Access-control boundaries
  • Data minimization requirements
  • Encryption requirements
  • A retention or automatic- ...[truncated 1966 chars]
Remediation
View remediation

Remediation Suggestions

  1. Make persistent recording opt-in and request explicit user consent before the first write.
  2. Explain which fields will be retained, why they are needed, where they are stored, and how long they will remain.
  3. Store concise summaries rather than raw conversations, and exclude sensitive details unless strictly necessary.
  4. Define a retention period with automatic expiration.
  5. Provide commands or documented procedures to review, correct, export, and permanently delete profile data.
  6. Require per-user and per-tenant isolation with least-privilege access controls.
  7. Require encryption in transit and at rest when the host supports persistent storage.
  8. Do not claim that information has been saved unless the host exposes an approved storage mechanism and confirms a successful write.
  9. Add an explicit no-storage mode and make it the default for sensitive or mental-health-adjacent discussions.
  10. Document that persistence is host-dependent and prohibit fallback storage in logs, temporary files, or shared workspace files.

T08 · Insecure Dependencies

Warning
Location
README.md:22
Finding

Installation Documentation References a Different Package Name

Content
View full analysis

Vulnerability Details

File Location: README.md:22
Related Manifest Location: clawhub.json:2
Vulnerability Type: Package-name confusion in installation instructions
Risk Level: Medium

Vulnerable Source Segment:

bash
clawhub install intp-coach

Conflicting Manifest Segment:

json
{
  "name": "infp-coach"
}

Technical Analysis

The documented installation command requests intp-coach, while the audited package identifies itself as infp-coach. A user following the README may therefore install a different package from the package registry rather than this audited artifact.

This creates a dependency-confusion or package-substitution risk. The security properties of the resolved intp-coach package cannot be inferred from the files reviewed here. If that package is controlled by an unrelated or malicious publisher, it could supply different instructions or executable content.

The project contains further identity inconsistencies that reduce provenance confidence: much of README.md and references/intp-insights.md describes an INTP coach, while SKILL.md and the manifest identify an INFP coach. The manifest repository URL also points to a repository named women-awakening-museum, which does not visibly correspond to the package name. These additional inconsistencies do not independently demonstrate malicious behavior, but they make package identity verification more difficult.

Attack Path

  1. A user obtains this project and trusts its installation documentation.
  2. The user runs clawhub install intp-coach.
  3. The registry resolves intp-coach instead of the audited infp-coach artifact.
  4. An unrelated publisher's package is downloaded and installed.
  5. The substituted package executes in the normal Skill context and may exercise any tools or permissions available to installed Skills.

This path requires a separately published intp-coach package. Th ...[truncated 738 chars]

Remediation
View remediation

Remediation Suggestions

  1. Replace the installation command with:
    bash
    clawhub install infp-coach
    
  2. Pin the installation to an audited version, such as 1.0.0, if supported by the package manager.
  3. Publish and document a package integrity digest or signed release identifier.
  4. Verify registry ownership of both infp-coach and the mistakenly referenced intp-coach name.
  5. Align all package identity fields across README.md, SKILL.md, clawhub.json, and reference files.
  6. Replace the repository URL with the canonical repository for this package, or document why the current repository is authoritative.
  7. Add an automated release check that fails when installation examples differ from the manifest package name.
  8. Remove or rename the INTP-specific reference material unless it is intentionally part of the INFP package and clearly documented.
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (21)

Description-Behavior Mismatch

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

The README consistently documents an INTP coaching skill, while the manifest metadata says the skill is for INFP users only. This mismatch can cause the agent to activate under the wrong user context or deliver guidance based on the wrong personality model, undermining routing integrity and creating a deceptive skill surface for users and maintainers.

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 entirely written for INTP coaching while the skill metadata says the skill should only trigger for INFP users. This creates a scope/instruction mismatch that can cause the agent to deliver personality guidance for the wrong user profile, undermining reliability and potentially causing harmful or misleading advice in a mental-wellbeing/coaching context.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The file explicitly labels itself as 'INTP Coach' reference material and repeats that intent in its footer metadata, directly contradicting the declared INFP-only skill purpose. This increases the likelihood that downstream prompting or retrieval will surface the wrong behavioral framework, making the mismatch systematic rather than incidental.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The usage examples show activation from phrases like '分析瘫痪了,想太多做太少,咋办?直接给建议。' without the explicit self-identification required by the manifest. This can broaden activation beyond intended scope, causing the skill to engage users who did not assert the required INTP/INFP context and increasing the chance of misclassification and inappropriate personalized advice.

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 the skill will automatically save conversations and action records, but it provides no notice about retention, consent, storage location, access controls, or deletion. In a coaching context, users may share sensitive personal, emotional, or career information, so silent persistence creates meaningful privacy and compliance risk.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The manifest says vague emotional expressions without INFP context should not trigger the skill, yet the instructions tell the skill to proactively engage users in exactly that scenario. In a coaching context, this broadens activation into emotionally sensitive conversations without the declared precondition, increasing the chance of unintended intervention and collection of sensitive personal data.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The manifest states the skill should not trigger for non-INFP users, but the body explicitly instructs serving ENFP and other types anyway. This creates policy/behavior drift that can cause the agent to activate outside its declared scope, leading to misrouting, deceptive capability boundaries, and potentially inappropriate personalized advice to users the skill was not designed for.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The skill advertises '成长记录:保存你的对话和行动,下次继续聊' without a clear user-facing privacy notice or consent flow. Because the data concerns emotions, goals, behavior, and personal development, silent retention can expose sensitive profiling and violate user expectations around ephemeral coaching conversations.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The skill promotes persistent collection and reuse of conversation and action-history data without clear minimization limits. In a coaching skill handling emotional state, motivation, interests, and behavior, such broad retention can become a sensitive dossier that exceeds what is necessary for immediate assistance.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The normal coaching flow instructs recording '关键洞察' into a user file during routine interaction. These insights may include intimate self-assessments, emotional patterns, or inferred traits, and storing them persistently increases privacy risk and downstream misuse if later reused out of context.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The '成长记录(用户档案)' section instructs recording a user profile containing core problems, interests, action history, and insights, but it does not require notifying the user. This is dangerous because it normalizes hidden accumulation of sensitive behavioral and psychological profile data for future reuse.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

This section defines broad retention of interests, recurring behavioral patterns, action promises, and outcomes for future use. Such longitudinal profiling can reveal highly sensitive personal traits and vulnerabilities, especially in a mental-health-adjacent coaching setting, and the lack of minimization or boundaries makes misuse more likely.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The mandatory instruction to '记录用户档案(下次对话可回顾)' requires future recall of personal information as a standard behavior. Making persistent memory mandatory, rather than optional and consent-based, increases the risk of over-collection and unexpected reuse of sensitive coaching data.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The conversation-ending workflow directs the system to archive user problems, insights, plans, and outcomes at session close, again without a clear notice to the user. End-of-session archival is especially risky because users may believe the interaction is concluding cleanly while sensitive personal details are being persistently retained behind the scenes.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The end-of-conversation archival instructions include storing user problems, insights, action plans, and execution results. This creates a persistent, behaviorally rich record of a user's struggles and follow-through, which is especially sensitive in a self-improvement coaching context and lacks clear collection limits or user control.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The skill context says this skill is for users who explicitly identify as INFP, but the manifest keywords and tags repeatedly identify the skill as INTP instead. This creates a semantic mismatch between the declared trigger/purpose and what the packaged skill metadata advertises, which could cause the wrong skill to be surfaced or invoked.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

This markdown file contains all instructional/reference text in Chinese, and it does not indicate that the user can opt into another language or that the skill is intentionally region-specific. Under the policy, forcing a specific language without user choice is a natural-language policy violation.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The entire reference file is written in Chinese and presents all coaching phrasing, labels, and examples only in that language, with no indication that users can choose another language or locale. This creates a natural-language policy concern because the skill appears to impose a single language experience without documented opt-in or justification.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
83% confidence
Finding

All user-facing instructions and examples are written in Chinese, and the document does not indicate that users may choose another language or that the skill is intentionally limited to a Chinese-speaking context. Per the policy, forcing a specific language without opt-in can be a natural-language policy violation.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
78% confidence
Finding

This manifest contains natural-language metadata that appears to target one audience in Chinese while also using conflicting MBTI labels across the name, description, keywords, and tags. Because SQP-3 covers language or locale policy violations, a user-facing description that effectively assumes Chinese without opt-in or documented scope can be a policy concern in globally distributed skills.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Low
Category
Not specified by scanner
Confidence
72% confidence
Finding

The manifest declares an INFP coaching skill, but the repository URL points to a project named "women-awakening-museum," which does not semantically match the stated skill identity. While this may be a packaging mistake, it is still a manifest-level mismatch that undermines confidence that the referenced source corresponds to the claimed skill.

Content

No source excerpt is available for this finding.