Back to skill

Security audit

Istp Coach

Security checks for vulnerabilities and agentic risk

Overview

The skill is mostly a coaching prompt package, but its identity, install instructions, and persistent user-profile behavior are materially inconsistent and under-disclosed.

Review carefully before installing. The package should be corrected so the name, README, manifest, references, install command, and repository all point to the same ISTP or INTP skill. Do not enable any memory or profile-saving behavior unless you are comfortable with personal coaching details being retained and can inspect or 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 Requested Without Consent or Data-Handling Safeguards

Content
View full analysis

Vulnerability Details

File Location: SKILL.md:323-334, with related persistence instructions at SKILL.md:431-458
Vulnerability Type: T09: Insecure Skill Coding Practices
Risk Level: Medium

Vulnerable Skill Text

The following is an English rendering of the complete relevant directives in SKILL.md:323-334:

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 the conversation

**Usage scenarios**:
- Review previous progress in the next conversation
- Identify user patterns
- Adjust the coaching strategy

Related end-of-conversation instructions at SKILL.md:431-458 direct the Agent to archive the user's problem, insights, action plan, and later execution results for recall in subsequent conversations. The same behavior is advertised in README.md:50 as automatically saving conversations and actions.

Technical Analysis

The Skill handles potentially sensitive coaching information, including emotional condition, relationships, career concerns, behavioral patterns, goals, and execution history. It instructs the Agent to persist this information in a user profile but does not define:

  • An explicit opt-in or informed-consent step
  • The storage system or permitted storage scope
  • Data minimization requirements
  • A retention or automatic-expiration period
  • A user-accessible review, correction, or deletion process
  • Encryption or access-control requirements
  • A rule preventing raw user instructions from being stored as reusable memory
  • Separation between factual profile data and untrusted conversational content

This creates both a privacy concern and a memory-integrity risk. If the hosting Agent supports persistent memory, user-controlled content co ...[truncated 1926 chars]

Remediation
View remediation

Remediation Suggestions

  1. Require explicit opt-in before creating or updating a persistent profile.
  2. Show the exact proposed memory entry and require user confirmation before saving it.
  3. Store only a concise, structured summary approved by the user; do not store raw conversation transcripts by default.
  4. Prohibit persistence of commands, role instructions, tool requests, URLs, executable content, or text that attempts to control future sessions.
  5. Define a retention period and automatically delete expired records.
  6. Provide commands or documented procedures for users to inspect, correct, export, and delete all saved information.
  7. Bind records to the correct authenticated user and prevent access from unrelated sessions or users.
  8. Require encryption at rest and in transit when the runtime supports external storage.
  9. Distinguish verified user facts, tentative inferences, action plans, and untrusted quoted content using separate structured fields.
  10. Replace the unconditional archival instruction with language such as: “Offer to save a minimal summary only after obtaining explicit user consent.”

T08 · Insecure Dependencies

Warning
Location
README.md:18
Finding

Installation Documentation Resolves a Different Package Than the Audited Skill

Content
View full analysis

Vulnerability Details

File Location: README.md:18-23
Vulnerability Type: T08: Insecure Dependencies
Risk Level: Medium

Vulnerable Code

The installation section directs users to install a different package name:

bash
clawhub install intp-coach

The audited package is identified differently in both manifests:

yaml
# SKILL.md:2
name: istp-coach
json
{
  "name": "istp-coach",
  "version": "1.0.0",
  "main": "SKILL.md"
}

Technical Analysis

The documented command installs intp-coach, while the audited artifact is named istp-coach. Package registries generally resolve dependencies by exact package name. Consequently, the installation command can retrieve an artifact that is distinct from the files reviewed in this audit.

This creates a package-confusion boundary: users may believe they are installing the reviewed ISTP Coach while actually loading whatever content is registered under the INTP Coach namespace. The repository contains no version pin, digest, signature requirement, or post-install identity verification that would detect the discrepancy.

The surrounding metadata reinforces the integrity problem. README.md describes an INTP Coach, while SKILL.md and clawhub.json identify an ISTP Coach. In addition, clawhub.json:12 points to a repository named women-awakening-museum, which does not visibly correspond to either package name. These inconsistencies reduce confidence in package provenance, although they do not by themselves demonstrate malicious ownership.

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 intp-coach namespace rather than the audited istp-coach artifact.
  4. The installer downloads and activates content that was not reviewed as part of this audit.
  5. If the other namespace is compromised or c ...[truncated 943 chars]
Remediation
View remediation

Remediation Suggestions

  1. Replace the installation command with:

    bash
    clawhub install istp-coach
    
  2. Pin the reviewed package version, such as 1.0.0, if the package manager supports version constraints.

  3. Pin a content digest or require a verified package signature where supported.

  4. Add a post-install verification step that checks the installed name, version, publisher, and digest.

  5. Update README.md so its package name, personality type, examples, and descriptions consistently refer to ISTP Coach.

  6. Correct the INTP keywords and tags in clawhub.json.

  7. Replace the unrelated repository URL with the authoritative source repository for this package.

  8. Ensure the version in the SKILL.md front matter, the footer, and clawhub.json is consistent before publication.

  9. Add release validation that fails when the documented installation name differs from the manifest name.

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 (18)

Intent-Code Divergence

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

The README consistently documents an INTP coaching skill while the manifest metadata declares an ISTP coaching skill. This mismatch can cause users, reviewers, or downstream systems to invoke the skill under false assumptions, leading to incorrect psychological guidance and weakening trust and governance over what the skill actually does.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The documented behavior is tailored to INTP traits, examples, and intervention strategies rather than the ISTP-focused scope advertised in metadata. In a coaching context, this increases the risk of systematically giving users advice intended for a different psychological profile, which can degrade outcomes and constitute deceptive or unsafe product behavior.

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 clearly written for INTP users while the skill metadata says the skill is for ISTP coaching. This mismatch can cause the agent to give systematically incorrect personality-specific guidance, undermining user trust and producing advice that is inapplicable or harmful in sensitive interpersonal or emotional contexts.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

The file explicitly brands itself as 'INTP Coach' reference material and lists its purpose as supporting an INTP coach, directly contradicting the declared ISTP skill intent. Because this contradiction is explicit rather than incidental, it increases the likelihood that the wrong behavioral framework will be used at runtime and that downstream outputs will be consistently misaligned.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
82% confidence
Finding

The entire README, examples, and interaction patterns are presented only in Chinese, and there is no statement that users may choose another language. 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.

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 the skill will automatically save conversations and action records, but it does not explain retention, consent, storage location, access controls, or deletion options. For a coaching tool handling potentially sensitive mental-health-adjacent disclosures, silent persistence raises privacy and compliance risks and may expose intimate user data if the storage is mishandled.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill explicitly promises '成长记录:保存你的对话和行动,下次继续聊', which implies persistence of conversation-derived personal data without any user-facing notice, consent flow, retention limit, or privacy controls. Because this skill deals with mental state, relationships, career confusion, and behavioral history, the archived data is potentially sensitive and its reuse across sessions increases privacy and profiling risk.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The feature list states that the system will save conversation and action history for future continuation, which is a direct instruction to persist and reuse user data beyond the immediate interaction. In a coaching context, this creates a durable behavioral profile that could expose sensitive personal patterns if mishandled, over-retained, or accessed by unauthorized parties.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

This section repeatedly requires storing core problems, interests, actions, and 'aha' moments in a user profile for future use. That is persistent collection of personal and potentially sensitive conversation-derived data, and the skill provides no guardrails around consent, necessity, minimization, or lifecycle management.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

This section instructs recording a user archive containing core problems, interests, action records, and key insights, but does not pair that requirement with a privacy warning or consent mechanism. Storing longitudinal personal-development and behavioral data can enable sensitive inference about mental state, habits, and vulnerabilities if accessed or reused improperly.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The '必须遵守' section makes recording a user archive a required behavior for future recall, turning persistence into a default operational rule rather than an optional feature. Because the skill targets emotional, relational, and career struggles, mandatory retention materially increases privacy risk and the chance of sensitive profiling.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The end-of-conversation workflow mandates archiving user problems, insights, action plans, and results after the chat ends, again without notifying the user that these details will be retained. Post-session storage is especially risky because users may assume the conversation is over while the system silently persists sensitive material for future profiling.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The conversation-end flow instructs archiving problems, insights, plans, and outcomes, creating a structured longitudinal record of the user's personal development and behavior. Such records can be highly sensitive in a coaching skill and are more dangerous here because the subject matter includes emotional distress, relationship difficulties, and execution failures that can reveal vulnerabilities over time.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The declared scope says the skill is for ISTP coaching, but the manifest metadata describes a different behavior domain centered on INTP traits such as analysis paralysis and focus issues. This scope confusion is dangerous because agent selection, policy review, and user expectations may rely on manifest metadata, causing misapplication of the skill and reducing the reliability of any safety or quality checks tied to declared purpose.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
98% confidence
Finding

The manifest claims the skill is for ISTP users, but the description, keywords, and tags repeatedly identify it as an INTP coaching skill. This mismatch can cause the wrong skill to be selected or trusted under false pretenses, leading to users receiving inaccurate personality-specific guidance and undermining integrity of routing, review, and user consent.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
87% confidence
Finding

This markdown file is entirely written in Chinese and presents coaching phrases and guidance only in that language, with no indication that users may choose another language or that the skill is intentionally limited to Chinese-speaking users. Under the policy, forcing a specific language without user opt-in 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
82% confidence
Finding

The file’s instructional content and metadata are all presented in Chinese, and there is no natural-language indication that the skill supports multiple languages or that Chinese-only behavior is intentional and user-selected. This can violate language/locale policy when a skill effectively forces one language without opt-in.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Low
Category
Not specified by scanner
Confidence
90% confidence
Finding

This file is clearly written as background knowledge for the istp-coach skill, but L101 says the document location is intp-coach/references/istp-traits.md. That inline documentation contradicts the surrounding file identity and usage context, creating intent/documentation divergence even though the content itself is about ISTP.

Content

No source excerpt is available for this finding.