Security audit
Security checks for vulnerabilities and agentic risk
This is a text-rewriting helper with no executable code, persistence, credential handling, or hidden data access, though some supporting docs are broader than the main writing-focused purpose.
Install this only if you want an automatic writing-revision helper. Be aware that its triggers are broad and some supporting documentation is generic, so you may want to invoke it explicitly for humanizing or revising supplied text rather than relying on automatic selection.
An overly broad everyday trigger phrase raises the risk of accidental invocation during ordinary conversation, which can silently route unrelated tasks into this skill. Because this skill’s surrounding documentation already broadens its scope, a vague trigger materially increases the chance of unintended activation for technical or operational requests outside safe boundaries.
The skill description uses broad activation terms like 'editing,' 'rewrite,' and 'voice' without clear boundaries, which can cause the agent to invoke this skill for many ordinary writing tasks beyond its intended niche. Overbroad routing increases the chance of inappropriate skill selection, policy bypass by misclassification, or unintended handling of sensitive content under a generic rewriting workflow.
The keyword list contains ambiguous everyday terms such as 'editing,' 'rewrite,' 'voice,' 'clarity,' and 'tone,' which are likely to match a large volume of benign requests. This makes the skill easy to trigger unintentionally, creating routing ambiguity and raising the risk that the skill will intercept tasks it should not handle, especially where stronger domain-specific controls are expected.
The skill description is broad enough to trigger on many ordinary editing or rewriting requests without clearly limiting when this helper should be selected. In an agent system, overbroad activation can cause the wrong skill to intercept prompts, leading to unintended text transformation, policy bypass of more appropriate review flows, or reduced user control over how content is handled.
The trigger keywords are generic terms like editing, rewrite, voice, and clarity, which are common across many benign writing tasks and provide no scope boundaries. This increases the chance of accidental invocation, routing conflicts, and misuse of the skill for requests it was not intended to handle, especially in automated agent selection pipelines.
The default prompt uses a very generic invocation phrase ('make this draft sound more specific and natural') that can match ordinary writing assistance requests rather than a narrowly scoped humanizer workflow. Because implicit invocation is enabled, this broad trigger increases the chance the skill is auto-selected in contexts the user did not clearly intend, which can cause unintended rewriting of content and inappropriate handling of sensitive or compliance-constrained text.
The requirement text materially expands the skill from a bounded text-humanization/editor role into broad software-support functions like bug fixing, setup hardening, reliability work, and adjacent skill creation. That scope drift can cause the agent to activate in contexts far outside the declared manifest, increasing the chance of inappropriate delegation, unsafe actions, or user confusion about the skill’s actual authority.
The implementation plan is framed as a generic workflow engine that can produce code, analysis, checklists, and operational artifacts, rather than a narrowly constrained writing-revision tool. In practice, this broad operational framing can let the skill be invoked for unintended tasks and increases prompt-surface area for misuse beyond the advertised humanizer function.
The usage signals and trigger phrases describe a general bug-fixing and hardening helper, which directly contradicts the manifest’s stated purpose of text revision and humanization. Misaligned triggers are dangerous because they can cause automatic routing of unrelated technical or security-sensitive tasks into a skill that was not designed, reviewed, or bounded for those tasks.
The invocation guidance lacks clear boundaries, so the orchestrator may treat many loosely related requests as eligible for this skill. In a skill already suffering from scope drift, poor trigger specificity increases misrouting risk and can expose users to outputs that exceed the intended editing-only context.
This markdown file contains user-facing instructions exclusively in Chinese, and there is no statement in the file that users may choose another language or locale. Under the policy rule for natural-language violations, forcing a specific language without user opt-in can be a violation unless clearly justified.
