The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill directs users to install or run the ecc-agentshield package from the NPM registry. This is a third-party dependency not maintained by a recognized trusted vendor.
[REMOTE_CODE_EXECUTION]: The skill provides commands to use npx ecc-agentshield, which downloads and executes code from a remote repository at runtime. This execution path bypasses traditional installation audits.
[COMMAND_EXECUTION]: The tool is designed to execute shell commands that interact with the local file system. Specifically, it reads the .claude/ directory and can modify project files when the --fix flag is applied.
[PROMPT_INJECTION]: The skill's primary function is to ingest and process local configuration data (such as CLAUDE.md and agent instructions) which are known surfaces for indirect prompt injection.
Ingestion points: Reads local configuration files including CLAUDE.md, settings.json, and agent markdown definitions.
Boundary markers: No explicit delimiters or boundary isolation instructions are provided in the skill documentation.
Capability inventory: Includes shell command execution through npx and the ability to overwrite local configuration files.
Sanitization: No description of sanitization, escaping, or validation of the audited content is provided.

security-scan