Security audit
Security checks for vulnerabilities and agentic risk
This skill only guides the assistant's reasoning style and output structure, with no code execution or data access.
Installers should expect this skill to make some decision-oriented answers more structured and explicit. Users who prefer only manual activation may want narrower trigger wording, but no security-sensitive access or hidden execution was found.
The trigger list uses subjective conditions such as 'mixed workloads', 'moral weight', and 'cases where slow mode is expensive' without concrete scope, examples, or exclusion conditions. These phrases are broad enough to overlap with many normal user requests, which could cause unintended invocation of the skill.
The default prompt includes a broad natural-language invocation phrase ('Use $dual-mode-reasoner to choose the right reasoning depth for this task') while implicit invocation is enabled. This can cause the skill to trigger on ordinary user requests about reasoning depth or disciplined analysis, creating unintended routing or prompt injection surface where the skill activates without clear user intent.
This markdown file defines trigger conditions such as 'evidence is weak or conflicting' and 'real people are materially affected' without clear boundaries or exclusions. The lack of specific scope, examples, or negative examples could cause the mode to activate in many ordinary situations where the intent is unclear.
The manifest sets 'language: en', which indicates an English-only locale preference, but the file does not offer opt-in language choice or explain why the restriction is necessary. Under the policy, forcing a specific language without user opt-in can be a natural-language policy violation.
No suspicious patterns detected.
