T09 · Insecure Skill Coding Practices
- Location
SKILL.md:323- Finding
Automatic Persistence of Sensitive Coaching Profiles Without Defined Privacy Controls
- Content
View full analysis
- Remediation
View remediation
Security audit
Security checks for vulnerabilities and agentic risk
The skill is not clearly malicious, but it needs Review because its package identity is inconsistent and it automatically records personal coaching history without clear privacy controls.
Review before installing. Confirm which package this is meant to be, do not follow the README install command until it matches the manifest, and only use it if you are comfortable with coaching notes being saved or the host platform lets you disable, inspect, and delete memory.
SKILL.md:323Automatic Persistence of Sensitive Coaching Profiles Without Defined Privacy Controls
clawhub.json:2Package Identity Mismatch Can Direct Users to Install an Unreviewed Package
The README documents an INTP coaching skill while the manifest metadata says this skill is for ESFP users. This mismatch can cause the agent to be invoked in the wrong context, deliver unsuitable guidance, and bypass user/operator expectations about what the installed skill does. In a coaching skill, personality-targeting is core behavior, so a type mismatch materially affects safety and trust.
The file is labeled as ESFP guidance, but the actual content repeatedly describes traits and cognitive dynamics associated with a different personality profile, creating a clear integrity failure in the skill’s behavioral grounding. In a coaching context, this can systematically deliver incorrect psychological advice to users, especially when the document is meant to be used internally rather than shown, making the mismatch harder to detect and correct.
The reference file is entirely focused on INTP coaching while the skill metadata says the skill is for ESFP users. This mismatch can cause the agent to deliver systematically incorrect personality-specific guidance, which is a security-relevant integrity failure because users may rely on advice that is not intended for their context.
The document explicitly labels itself as 'INTP Coach' reference material in its core strategy and metadata sections, directly contradicting the declared ESFP purpose. Because this is not just incidental wording but a strong self-identification, it increases the likelihood that the skill will confidently apply the wrong behavioral model and mislead users.
The skill says modes are 'automatically recognized' from broad, ordinary phrases, which creates ambiguous activation and behavior selection. That can cause the system to enter advisory mode when the user expected reflective coaching, or trigger the skill unintentionally from unrelated conversation, leading to unexpected guidance or data handling.
The README states that conversation and action records are saved automatically, but it does not clearly disclose retention scope, consent, storage location, or privacy implications. Because coaching conversations may contain sensitive personal or mental-health-adjacent information, silent persistence increases the risk of privacy harm, unauthorized reuse, or unexpected exposure.
The skill explicitly says it will save growth records including core problems, interests, actions, and insights, but it does not require a user-facing disclosure, consent step, retention policy, or minimization. Because the content concerns emotional state, career uncertainty, and behavioral history, persistent storage can expose sensitive personal data and create cross-session privacy risk.
The skill instructs the system to persist and reuse user-provided coaching details across sessions, effectively building longitudinal behavioral memory. Cross-session reuse increases privacy exposure because sensitive patterns, emotional states, and decision history can be aggregated over time without transparent user approval.
This portion directs the agent to create and maintain a user profile containing personal interests, problems, and action history. Even if intended for coaching continuity, profiling users in this way can reveal sensitive behavioral traits and enable unintended secondary use if the data is later accessed, leaked, or used out of context.
These sections repeatedly instruct the agent to record key insights and maintain a user archive, but never tell the user that personal coaching history may persist. Silent archival of reflective conversations is risky because users may disclose sensitive psychological, professional, or relational information under the assumption the exchange is ephemeral.
The end-of-conversation workflow mandates archiving user problems, insights, action plans, and execution results without an explicit warning or consent checkpoint. This is dangerous because it normalizes automatic persistence of sensitive conversation-derived data at the moment users are disengaging, making unnoticed retention more likely.
The mandated end-of-session storage of conversation-derived user information for future recall is a genuine privacy and data-handling issue. Because the skill operates in a quasi-coaching context, users may reveal especially sensitive personal context, making automatic future recall more dangerous than in a generic utility skill.
The manifest claims to be an ESFP coaching skill, but its keywords and tags describe an INTP-focused coaching tool. This inconsistency can mislead users, reviewers, and downstream systems about the skill’s actual purpose, making it easier to conceal mismatched behavior or route users to inappropriate guidance. In a personality-targeted coaching skill, identity consistency is important because prompts, examples, and advice are expected to be tailored to a specific user profile.
This markdown file is entirely written as internal operating guidance in Chinese and explicitly states it is used during skill execution as internal reference. Because the skill guidance does not indicate any user language preference, opt-in, or justified locale restriction, it creates a natural-language locale policy concern under the rule for language/locale violations.
The metadata at the bottom references a different coach package path, which suggests copy-paste drift or packaging confusion between skills. This weakens provenance and increases the chance that the wrong knowledge source is loaded, maintained, or audited, leading to cross-skill contamination and unreliable outputs.
This markdown reference file presents all instructions and coaching language exclusively in Chinese, and there is no indication that the user can opt into another language or that the skill is intentionally limited to a Chinese-speaking audience. Under the policy, forcing a specific language without user choice is a natural-language policy concern.
The title and description position the skill specifically as 'INTP Coach' and base its behavior on MBTI-specific guidance, with usage examples assuming the user is INTP. This creates a natural-language targeting constraint without indicating that other users may opt in, opt out, or select a different profile, which can be a policy concern when a fixed persona/category is imposed by default.
The repository URL points to a project name that appears unrelated to the declared ESFP coaching functionality. An unrelated source repository reduces traceability and trust, and can hinder security review by obscuring the true provenance of the skill or redirecting auditors to irrelevant code. In the context of distributed skill packaging, provenance mismatches are suspicious because they can conceal stale, swapped, or misleading artifacts.