The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs the agent to download and execute code from an external source using npx ecc-agentshield and npm install -g ecc-agentshield. Running unverified third-party packages from the NPM registry represents a remote code execution risk.
[COMMAND_EXECUTION]: The skill relies on executing various shell commands to perform scans and configuration initialization, including npx ecc-agentshield scan, npx ecc-agentshield init, and npx ecc-agentshield scan --opus.
[EXTERNAL_DOWNLOADS]: The skill references and downloads the ecc-agentshield package from the public NPM registry (npmjs.com/package/ecc-agentshield) and points to a third-party GitHub repository (github.com/affaan-m/agentshield).
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection (Category 8) because it is designed to ingest and process untrusted project data that may contain malicious instructions.
Ingestion points: Files located in .claude/, including settings.json, CLAUDE.md, mcp.json, and files in hooks/ and agents/ directories.
Boundary markers: None identified in the skill instructions to distinguish between configuration data and potential embedded instructions.
Capability inventory: The skill has the capability to execute shell commands and download external packages.
Sanitization: No evidence of sanitization or input validation for the files being scanned is provided in the skill description.
[CREDENTIALS_UNSAFE]: The documentation for the Deep Analysis feature guides users to set the ANTHROPIC_API_KEY environment variable in their shell (export ANTHROPIC_API_KEY=your-key), which may lead to accidental exposure of sensitive credentials in shell history or logs.

security-scan