Security audit
Security checks for vulnerabilities and agentic risk
This is a disclosed OSINT profiler, but it enables broad third-party personal dossier building without strong authorization and data-protection controls.
Only install this for self-audits or investigations where you have clear consent or legal authority. Do not use it to profile private people, map relationships, infer sensitive traits, or preserve identifiers unnecessarily; if used, redact outputs and secure or delete local reports promptly.
references/osint-pp-requirements.md:17Systematic Profiling of Arbitrary Individuals Without a Mandatory Authorization Gate
SKILL.md:171Optional Redaction and Undefined Protection of Locally Stored Sensitive Profiles
The trigger list includes broad phrases such as public information collection, background investigation, digital footprint profiling, and personal information mining, which can match ambiguous or high-risk requests and cause the skill to activate in contexts involving third-party profiling. In a skill specifically designed to reconstruct detailed personal dossiers from minimal identifiers, over-broad invocation materially increases misuse risk and bypasses user-intent verification.
The skill advertises reconstruction of a complete 12-dimensional profile of any person from public traces but does not present a prominent warning that such profiling can facilitate stalking, doxxing, discrimination, or social engineering. Because the capability is framed as a general-purpose collection system rather than a narrowly bounded self-audit tool, users are not adequately warned away from harmful or privacy-invasive use.
This section instructs systematic collection, correlation, and reconstruction of a detailed personal profile about arbitrary individuals from minimal identifiers, including complete output schemas and confidence scoring. Even if sources are nominally public, the operational design meaningfully lowers the cost of privacy invasion, targeted harassment, stalking, doxxing, and pretexting by turning scattered data into an actionable dossier.
The skill explicitly directs gathering relationship networks and inferring sensitive attributes such as body traits, culture, social class, role hierarchy, and connections to other people. Mapping associates and deriving psychological or socioeconomic characteristics from public traces creates heightened harm because it enables social engineering, coercion, deanonymization, and broader collateral privacy violations affecting both the target and their network.
These instructions operationalize inference of hidden traits like motives, behavior patterns, career trajectory, and salary-market positioning from resumes, writing style, and activity signals. That goes beyond factual OSINT collection into psychographic and socioeconomic profiling, which is especially dangerous because it supports manipulation, discrimination, spearphishing personalization, and high-confidence targeting without the subject's knowledge.
The exemplar operationalizes comprehensive profiling of a named individual across 12 dimensions, including inferred and sensitive traits, without any privacy warning, minimization guidance, or restriction on misuse. In this skill context, the absence of safeguards is especially dangerous because the surrounding capability is explicitly designed for systematic identity correlation, inference, and dossier creation from public sources.
These instructions direct the model to compile a full personal dossier from multiple public sources and output it in a structured schema, normalizing mass aggregation and inference about an identifiable person. In the context of an OSINT profiling skill, this materially increases the risk of stalking, doxxing, targeted phishing, employment discrimination, and other privacy harms even if each source is individually public.
This section explicitly calls for disclosure and inference of sensitive personal data, including family status, child status, health indicators, ethnicity, relationship networks, and social role. The skill context makes this more dangerous because it turns scattered public traces into an actionable, intimate dossier that can facilitate coercion, stalking, social engineering, and discrimination against a real person.
The catalog explicitly starts from minimal identifiers such as name, nickname, email, or phone number and directs systematic cross-platform discovery and disambiguation to identify a person. Even if sources are nominally public, this enables aggregation of sensitive personal data into a searchable dossier without meaningful privacy safeguards, consent checks, or harm warnings, which materially increases stalking, harassment, doxxing, and social-engineering risk.
This file defines an end-to-end profiling pipeline that begins with minimal identifiers, expands across multiple public and quasi-public sources, resolves conflicts, scores confidence, and infers sensitive traits and motivations. The structured, iterative enrichment workflow substantially lowers the effort required to build high-fidelity dossiers on arbitrary individuals, making the skill directly useful for doxxing, stalking, social engineering, and targeted abuse.
The instructions explicitly direct collection of identity details, family/social roles, physical characteristics, relationships, career information, and inferred motives into a final profile. Because the skill operationalizes broad personal-data assembly and inference rather than incidental mention, the context makes the finding more dangerous: it is a dedicated profiling capability, not a neutral educational discussion.
The task list instructs collection and inference of age, gender, body type, appearance, nationality, ethnicity, language, and subcultural affiliation from photos, videos, and social content. These are highly sensitive or protected characteristics, and the skill provides no meaningful warning against unreliable or discriminatory inference, nor any safeguard limiting such processing to lawful, user-consented contexts.
The later stages direct construction of relationship graphs, hierarchy assessment, capability/motivation inference, confidence scoring, gap analysis, and a final 12-dimensional profile output. This moves beyond simple search into structured surveillance and behavioral profiling, creating an operational dossier that can be exploited for impersonation, coercion, targeted phishing, or other privacy-invasive harms.
The document directly instructs broad collection of personal data across multiple identity dimensions and inclusion of that information in a consolidated profile. This is dangerous because it operationalizes invasive dossier building on real people, including highly sensitive data, which can be readily repurposed for stalking, doxxing, discrimination, or coercive targeting.
The skill specifies a comprehensive workflow for collecting, correlating, and assembling dossiers about real individuals across many platforms, but does not provide strong user-facing safety warnings, purpose limitations, or abuse-prevention controls. In context, this omission is dangerous because the skill is explicitly designed for large-scale personal profiling, making misuse for stalking, doxxing, coercion, or invasive surveillance much easier.
The file defines an end-to-end profiling pipeline that starts with identity resolution, expands into cross-source collection, infers sensitive traits and motives, resolves conflicts, scores confidence, and produces a final 12-dimensional dossier. The cumulative design makes the skill more dangerous than isolated OSINT steps because it systematizes escalation from basic identification to deep behavioral and relational profiling at scale.
The skill instructs the agent to derive usernames from identifiers and search across platforms, including cached or deleted pages, to discover linked personal accounts. This materially increases re-identification and cross-platform tracking capability, allowing an operator to expand from one identifier to a broad map of a person's digital footprint.
These sections direct collection or inference of sensitive personal attributes such as age, gender, health-related details, nationality, ethnicity, religion, and social class indicators, while lacking clear safeguards or user warnings. Sensitive-attribute profiling substantially elevates privacy, discrimination, and targeting risks, especially when tied to named individuals in a reusable profile.
The skill instructs the agent to build relationship networks and infer hierarchy, abilities, motives, and behavioral patterns without robust warnings about the harms of social graphing and psychological inference. This is especially dangerous because such outputs can enable social engineering, blackmail, workplace targeting, or highly tailored harassment beyond simple OSINT lookup.
The skill documentation is written entirely in Chinese and specifically mandates PRC legal conditions for use, but does not indicate that users may choose another language or that the locale constraint is limited to a region-specific deployment. This can constitute a language/locale policy issue because the skill appears to impose a single language and jurisdictional framing without opt-in or clear scoping.
The entire exemplar index and template are written only in Chinese, and the file does not indicate that the skill is region-specific or that users may choose another language. Under the stated policy, forcing a specific language without user opt-in is a natural-language policy violation.
SQP-3 applies to all file types and covers language or locale policy violations. This markdown exemplar presents all instructions and labels only in Chinese, which effectively forces a specific language on users without any visible opt-in, alternative language, or stated region-specific justification.
This markdown file is entirely in Chinese and does not indicate that the user can choose another language or that the skill is intentionally limited to a Chinese-language audience. Under the policy, forcing a specific language without user opt-in is a natural-language policy violation.
The title and output instructions are written as a Chinese-only exemplar and direct the skill to emit the defined profile format in that locale, but there is no indication that the user can choose language or that the Chinese-only requirement is justified by a region-specific need. This creates a natural-language locale policy concern under the language/locale rule.
The source inventory demonstrates cross-platform account linkage for a specific target, which enables identity correlation and easier operationalization of harassment or targeted attacks. Although less severe than the sensitive-attribute sections, including concrete handles and linked accounts increases doxxing and reconnaissance value, especially when bundled with the broader profile.