The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the user to install the 'ecc-agentshield' package from NPM (npmjs.com/package/ecc-agentshield) or run it via npx. This package is maintained by an unverified third-party account ('affaan-m').
[COMMAND_EXECUTION]: Multiple instructions encourage running 'npx ecc-agentshield scan', which executes code from the external package. This represents a potential risk if the package were compromised or malicious.
[EXTERNAL_DOWNLOADS]: The documentation references a GitHub Action ('affaan-m/agentshield@v1') for CI/CD integration, introducing a dependency on a third-party repository not belonging to a known trusted organization.
[PROMPT_INJECTION]: The skill processes project configuration files (e.g., CLAUDE.md, settings.json) which are untrusted data sources, creating an indirect prompt injection surface.
Ingestion points: The tool reads configuration files in the '.claude/' directory (SKILL.md).
Boundary markers: There are no instructions to the agent to use delimiters or ignore instructions within the scanned files.
Capability inventory: The skill uses 'npx' to execute a tool that reads and can modify files ('--fix' flag) based on the content it parses (SKILL.md).
Sanitization: No sanitization or validation of the scanned file content is mentioned in the skill instructions.

security-scan