T08 · Insecure Dependencies
- Location
README.md:22- Finding
Installation Instructions Reference a Different Package
- Content
View full analysis
- Remediation
View remediation
Security audit
Security checks for vulnerabilities and agentic risk
The skill is a non-executable coaching guide, but its package identity is inconsistent and it directs agents to retain sensitive coaching history without clear consent or controls.
Install only after the publisher aligns the package name, README, metadata, repository, and persona scope. Treat the growth-record feature as sensitive: do not share personal or mental-health-adjacent details unless the runtime gives clear memory controls and a way to review and delete stored records.
README.md:22Installation Instructions Reference a Different Package
SKILL.md:323Persistent Personal Coaching Records Are Required Without Defined Privacy Controls
The README describes an INTP coaching skill while the manifest says the skill is for ENTP users. This mismatch can cause the agent to activate under the wrong conditions or apply the wrong behavioral framing, creating scope drift and misleading users about what the skill does. In a personality-targeted coaching skill, that inconsistency is materially risky because the natural-language contract defines routing, expectations, and user treatment.
The reference file is explicitly written for INTP users, while the skill metadata says the skill should trigger only for ENTP users. This mismatch can cause systematic misclassification and delivery of personality-specific guidance to the wrong audience, which is especially risky in coaching contexts where advice is tailored around user identity and behavior patterns. The skill context makes this more dangerous because the entire file is framed as behavioral guidance, so the wrong reference data directly affects outputs rather than being a harmless labeling issue.
The README says the skill will automatically save conversations and actions, which expands behavior beyond the manifest's described coaching purpose. Undisclosed persistence of sensitive self-reflection data introduces privacy and consent risk, especially in a coaching context where users may share mental-health-adjacent or deeply personal information. The danger is increased because 'automatic execution' implies no clear user opt-in.
Saying modes are 'automatically recognized' without clear trigger boundaries makes activation and behavior selection ambiguous. In practice, broad everyday phrases can cause the skill to engage or shift into advisory mode unexpectedly, which is risky for a coaching skill that may influence personal decisions. The mismatch between narrow manifest triggers and broad README behavior increases the chance of overreach.
The README describes saving conversation history and user profiles but does not clearly explain retention, privacy impact, or user control. Because this skill invites disclosure of personal struggles, goals, and behavioral patterns, undisclosed storage can expose sensitive data and violate user expectations. The context makes this more dangerous than generic chat logging because the content is likely intimate and psychologically revealing.
The manifest explicitly says the skill should not trigger for vague emotional expressions without ENTP context, but the body instructs the agent how to handle exactly those inputs. This broadens activation beyond the declared scope and can cause the skill to engage in sensitive emotional coaching when the user did not provide the required ENTP context, increasing the chance of inappropriate or unexpected intervention.
The manifest says the skill should not trigger for non-ENTP users, yet the documentation explicitly allows serving other personality types. This creates scope drift and can cause the agent to provide advice outside its intended audience and design assumptions, undermining routing controls and user expectations.
The skill introduces '成长记录' and saving conversation/action history without clearly warning users that personal coaching data may be retained. In a coaching context, users are likely to share sensitive personal, emotional, and career information, so undisclosed retention creates a privacy and trust risk.
The skill instructs persistent collection and reuse of user conversation data in a profile/archive, which can create a standing dossier of sensitive personal information. Persistent memory is more dangerous than transient processing because it increases the blast radius of accidental disclosure, misuse, or over-collection across sessions.
Directing the agent to record core problems, blocks, and goals into a persistent dossier can capture highly sensitive personal and psychological details. In a self-improvement/coaching setting, this context makes the issue more serious because users may reveal vulnerabilities they would not expect to be archived long-term.
The instructions require recording a user profile including problems, interests, actions, and insights for future sessions without an explicit consent step. Because the data concerns behavioral patterns and personal struggles, collecting it by default can expose users to privacy harms and unauthorized profiling.
The skill specifies ongoing logging of user problems, interests, action records, and insights for later analysis and reuse, effectively creating longitudinal behavioral profiling. This increases privacy risk because repeated sessions can accumulate a detailed picture of the user's habits, struggles, and decision patterns.
The mandatory rules require recording user profile information for future recall across conversations, making persistence a default behavior rather than an optional feature. Default cross-session memory can surprise users and compounds the harm from over-collection in a personal coaching context.
The end-of-conversation workflow mandates archiving key user information and later execution results, but does not disclose retention or ask permission. This is especially risky in a coaching skill because session summaries may include sensitive mental state, goals, and behavioral follow-up data.
The prescribed end-of-conversation archive includes problems, insights, plans, and later execution results, which amounts to maintaining a detailed personal history over time. In a coaching skill, this longitudinal record can reveal emotional state, motivation, discipline, and life direction, making unauthorized retention or exposure particularly sensitive.
The description is written entirely in Chinese and presents the skill as a Chinese-language coaching experience, but the manifest provides no indication that language choice is optional or that the skill is limited to a Chinese-speaking audience. This can violate language/locale policy when a skill implicitly forces a specific language without user opt-in.
The provided skill context says this skill should trigger only for ENTP users seeking growth/career/learning guidance. However, the manifest keywords and tags repeatedly identify the skill as INTP-focused ("INTP", "intp"), which semantically misrepresents the skill's actual intended audience and behavior selection context.
This markdown file is natural-language guidance, and its content appears to assume Chinese as the required interaction language throughout. Under the policy, forcing a specific language without user opt-in or a documented justification is a language/locale policy violation.
The file self-identifies as 'INTP Coach' reference material in its later documentation, directly contradicting the ENTP-only intent declared by the manifest. This increases the likelihood that the mismatch is real and operational, leading maintainers or downstream agents to trust incorrect scope information and serve unsuitable coaching advice. In this skill context, contradictory identity labels undermine reliability and can cause persistent behavior drift.
The README brands the skill as 'INTP Coach' and all examples and guidance assume the user is INTP, with no visible language indicating that users can opt out of that framing or choose a broader mode. While not a direct language setting issue, it is a natural-language constraint imposed by the skill description that may not fit users unless explicitly selected.
The '常见错误' section says ENTP cognitive-function content should not be directly output, yet its own '解决' example directly outputs 'Ti-Ne 循环' to the user. Later, L426-L427 again says such knowledge is background only and should not be directly shown. This is an internal intent contradiction in the documentation about expected behavior.
The manifest identifies the package as an ENTP coaching skill, but the repository URL points to a project named "women-awakening-museum", which appears unrelated to the stated skill purpose. This creates a semantic mismatch between what the skill claims to be and the code/source location it associates with.
This reference file is entirely written in Chinese and is intended for the skill's internal use during execution, which suggests the skill may steer responses using a fixed language/locale context. There is no indication in this file that users can choose language or that the Chinese-only constraint is justified as region-specific.
The file is clearly ENTP-focused and identified earlier as part of the entp-coach skill, but L101 documents its location as intp-coach/references/entp-traits.md. This is an active documentation contradiction that can mislead maintainers about which skill owns or should load this reference content.