copilot-instructions-blueprint-generator
Fail
Audited by Gen Agent Trust Hub on Sep 20, 2026
Risk Level: CRITICALINDIRECT_PROMPT_INJECTIONSAFE
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill provides a prompt for the agent to scan an entire codebase (source code, configuration files, and package managers) to identify patterns and versions. This creates an attack surface where malicious instructions hidden within the codebase (e.g., in HTML/code comments or project metadata) could be ingested and reflected in the generated
copilot-instructions.mdfile, potentially influencing the behavior of GitHub Copilot in the future. - Ingestion points: Analyzes project files,
package.json,.csproj,pom.xml, and general source code patterns within the repository. - Boundary markers: The prompt includes a constraint to be "strictly based on actual code patterns," which serves as a basic behavioral guide but does not provide technical delimiters to prevent the ingestion of embedded instructions.
- Capability inventory: The skill generates text instructions that are intended to be stored in the repository and used by GitHub Copilot to guide its code generation.
- Sanitization: There is no explicit mechanism described to filter or sanitize potential natural language instructions found within the processed code files.
Recommendations
- CRITICAL: 1 file(s) identified as malware by FileRep - DO NOT USE
Audit Metadata