The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes lead data retrieved from the dropleads_search_people tool. This external data is then transformed and passed to the deepline enrich command, creating an ingestion surface where untrusted content could influence subsequent agent actions.
Ingestion point: deepline/data/quickstart_search.json (Output from external tool).
Boundary markers: Absent; the instructions do not explicitly tell the agent to ignore instructions embedded in the search results.
Capability inventory: Subprocess execution (deepline CLI, python3), file system access, and network operations (via CLI tools).
Sanitization: The transformation script performs basic key extraction and URL normalization but does not filter or validate the textual content of the search results.
[DYNAMIC_EXECUTION]: The skill generates and executes an inline Python script at runtime to transform search results.
Evidence: Use of `python3
<<'PY'` to execute logic that parses JSON and writes CSV files.
[EXTERNAL_DOWNLOADS]: The skill installs the deepline CLI tool using npm from a remote registry.
Downloads configuration and packages from code.deepline.com and the public npm registry.
[COMMAND_EXECUTION]: The skill relies on multiple shell commands to set up the environment and execute the demo, including global package installation and CLI tool invocation.

deepline-quickstart