Skill Composition Risk
This guide uses Skill composition risk as the report-derived label for unsafe effects that emerge across tool or skill chains.
Several source reports describe skill-composition risk as a runtime version of the same composition problem. A file scanner, summarizer, approval tool, or access manager may be harmless in isolation, but their outputs can create unsafe downstream state changes when joined through a shared context.
Three failure modes
Capability flow (capability flow) occurs when an upstream skill supplies the target or context that lets a downstream skill act. Trust lift occurs when an upstream reviewer or summarizer lends false legitimacy to a later action. Authorization blur occurs when advisory language shifts the agent’s internal approval boundary.
Why this belongs on Cognivirus.com
A cognivirus can move through skills as easily as through adapters. The persistent pattern may be a habit of trusting a label, accepting a summary, following a suggested path, or treating a previous tool’s output as authorization.
Control pattern
Skill graphs need path-aware review. The manifest must record not only which tools exist, but which outputs can be consumed as authority by which later steps. A safe skill should not be able to accidentally mint permission for another skill.
Vocabulary note
The report-derived label skill composition risk is used when capability flow, trust lift, or authorization blur turns benign skills into an unsafe path.
Skill composition risk is the mixed-case label used for validation and reader indexing.