T08 · Insecure Dependencies
- Location
README.md:18- Finding
Installation Documentation References a Different Skill Package
- Content
View full analysis
- Remediation
View remediation
Security audit
Security checks for vulnerabilities and agentic risk
This coaching skill has no executable malware, but its package identity is inconsistent and it asks the agent to keep sensitive coaching profiles without clear consent or deletion controls.
Review before installing. Do not follow the README install command unless the publisher fixes the package name and provenance. Use this only if you are comfortable with MBTI-based coaching and ensure any memory/profile saving is explicitly opt-in, reviewable, and deletable.
README.md:18Installation Documentation References a Different Skill Package
SKILL.md:323Automatic Retention of Sensitive Coaching Profiles Without Defined Privacy Controls
The skill metadata says this is an ENFJ-only coach, but the README fully documents an INTP coach with different triggers, examples, and behavioral assumptions. This kind of identity/intent mismatch can cause the wrong skill to activate or mislead users and reviewers about what the package actually does, undermining trust and safe routing.
The document is labeled as ENFJ guidance but its stated cognitive stack, strengths, weaknesses, and repeated references to Ti-Ne dynamics align with a different personality framing, creating materially misleading coaching content. In a guidance skill, this can systematically deliver incorrect self-models and advice to users, which is especially risky when users rely on the skill for growth, career, or emotional direction.
The reference file is clearly scoped to INTP coaching while the skill manifest says the skill should only trigger for ENFJ users seeking guidance. This mismatch can cause the agent to deliver persona-specific advice to the wrong audience, undermining user trust and causing incorrect behavioral guidance; in agent systems, hidden scope drift is a real integrity issue even without code execution.
The title, footer, and usage notes explicitly label the content as 'INTP Coach' reference material, directly contradicting the ENFJ-only declared purpose. This increases the likelihood that the runtime or future maintainers will treat the file as authoritative and serve incompatible coaching content, creating prompt/context integrity problems and misleading end users.
The README's stated purpose and usage examples contradict the manifest by describing INTP-specific coaching flows rather than ENFJ guidance. Even if not overtly malicious, contradictory intent documentation increases the chance of incorrect invocation, user confusion, and deployment of behavior outside the reviewed scope.
The README states that the skill will automatically save conversations and maintain growth records/user profiles, but it does not explain consent, storage location, retention period, access controls, or deletion options. For a coaching-oriented skill handling potentially sensitive personal and psychological information, silent persistence creates privacy and compliance risk.
The manifest description and the entire skill specification are written as if operation is fixed to Chinese, including trigger phrases, dialogue templates, and output-length rules in Chinese, with no indication that users may choose another language. Under the policy, forcing a specific language without opt-in can be a natural-language policy violation unless the locale restriction is clearly documented and justified.
The manifest says vague emotional expressions without ENFJ context should not trigger, yet the edge-case flow instructs the skill to engage such users and introduce itself proactively. That mismatch can cause the agent to enter sensitive emotional conversations outside intended scope, increasing privacy and safety risk for users who did not request this coaching skill.
The manifest says the skill should not trigger for non-ENFJ users, but the body explicitly invites use by other MBTI types. This creates instruction/dispatch inconsistency that can broaden activation scope beyond what platform routing and user expectations imply, causing unintended collection of personal coaching data and misapplication of type-specific advice.
The skill describes 'growth records' and saving dialogue/actions for future continuation, but it does not provide a user-facing privacy notice, retention policy, or consent step. Because the content involves personal growth, career uncertainty, emotions, and behavior patterns, silent persistence can expose sensitive personal data and violate data-minimization expectations.
The skill instructs the agent to persist and reuse user conversation details across sessions without defining minimization boundaries or excluding sensitive categories. In a coaching context, those details can include emotional state, career confusion, habits, and interpersonal issues, making cross-session profiling risky if stored by default.
The workflow explicitly says to record key insights into a user profile during conversation. Capturing 'insights' is especially sensitive because it converts subjective interpretation into durable profile data, which can amplify privacy harm, profiling risk, and future misuse if the interpretation is wrong or more intimate than the user intended to share.
The defined archive includes personal problems, interests, actions, and 'aha' insights for future recall, which amounts to a persistent behavioral profile. In a coaching skill, this can accumulate sensitive longitudinal data about emotions, goals, struggles, and habits without sufficient controls, increasing privacy exposure and potential downstream misuse.
The 'must comply' section makes profile recording a required behavior, removing discretion and increasing the chance that all users are persistently tracked regardless of need. Mandatory persistence is especially problematic in a personal coaching setting because it normalizes collection of sensitive self-disclosures without contextual consent.
The conversation-end flow instructs the agent to archive user problems, insights, plans, and outcomes for later retrieval. End-of-session summarization into persistent storage can capture the most sensitive distilled version of a user's struggles and behavior patterns, creating a durable dossier even when the user may think the conversation has ended privately.
The manifest claims the skill is ENFJ-focused, but its description, keywords, and tags instead reference INTP coaching. This mismatch can cause incorrect routing, user deception, and review failures, and in a skill ecosystem it increases the risk that users invoke behavior they did not intend or that unrelated prompt content is being repackaged under a misleading identity.
This markdown file is entirely written in Chinese and is described as internal guidance for the skill at execution time. Because the skill appears to use this reference as behavioral context and there is no indication of user opt-in, language choice, or region-specific justification, it presents a natural-language locale policy concern.
This markdown file contains user-facing guidance exclusively in Chinese, and nowhere indicates that the skill is Chinese-only or that users may choose another language. That can violate language/locale policy where skills should not force a specific language without user opt-in or explicit justification.
The natural-language description presents the skill as an "INTP Coach" and repeatedly assumes the user is INTP, but does not explicitly offer a choice for users who may not identify with that label. This can create a policy concern where the skill imposes a narrow categorization without opt-in or applicability clarification.
The repository URL appears unrelated to the declared coaching skill, which undermines provenance and trust. An unrelated source link can frustrate auditing, hide the real development history, or direct reviewers and users to content with a different purpose than the published skill.