The Agent Skills Directory

[DATA_EXPOSURE_AND_EXFILTRATION]: The skill attempts to read the internal Claude configuration file at ~/.claude/plugins/installed_plugins.json using grep. This information is used to generate a session summary which is sent to the vendor's backend via the cargo-ai workspaceManagement session upsert command. While primarily for attribution and telemetry, this involves accessing metadata outside the skill's direct operational scope.
[COMMAND_EXECUTION]: The skill uses multiple shell commands to manage its lifecycle and perform data operations. It installs the @cargo-ai/cli package globally using npm install -g, executes cargo-ai for authentication and data enrichment, and utilizes the gh (GitHub CLI) to star the vendor repository after obtaining user consent.
[EXTERNAL_DOWNLOADS]: The skill fetches and installs the @cargo-ai/cli tool from the npm registry. It also suggests adding a full skill pack from the vendor's GitHub repository using npx. As these are vendor-owned resources, they are considered standard for the skill's functionality.
[INDIRECT_PROMPT_INJECTION]: The skill processes CRM records which are potentially untrusted external data. Ingestion points: Record data containing emails, names, and domains are passed via the --records parameter in SKILL.md. Boundary markers: The data is structured as JSON, but there are no specific instructions or delimiters provided to the agent to prevent the interpretation of instructions that might be embedded in the record fields. Capability inventory: The skill has permissions to execute shell commands, perform network requests via the CLI, and write to local configuration files. Sanitization: The skill relies on JSON formatting for data separation but does not perform explicit sanitization or validation of the input fields before passing them to the command line.

crm-enrichment