The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill relies on the ecc-agentshield NPM package and source code from an individual developer's repository. These external resources are not managed by a verified or trusted organization, presenting a potential supply chain risk if the package were compromised.
[COMMAND_EXECUTION]: The skill involves running shell commands via npx or local installation to perform audits. This executes the external tool's code on the local system with the same permissions as the user, granting it access to sensitive configuration directories.
[INDIRECT_PROMPT_INJECTION]: The skill audits configuration files that may contain untrusted data. If a malicious user places hidden instructions within these config files, they could potentially influence the agent's behavior when it reviews the scan results.
Ingestion points: Claude Code configuration files (.json, .js, etc.) located in the local configuration directory.
Boundary markers: None specified in the instructions to prevent the agent from following instructions embedded in the audited files.
Capability inventory: The agent has local shell access and file system read capabilities via the auditing tool.
Sanitization: No sanitization or filtering of the configuration content is mentioned before the agent processes the findings.

Scan Claude Code configs for secrets permission drift and unsafe MCP hookups with AgentShield