The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external websites and processes it before passing the information to a local script, creating a potential surface for indirect prompt injection.
Ingestion points: Product data and specifications are retrieved from third-party websites using the WebSearch and WebFetch tools in Step 3.
Boundary markers: The skill includes a human-in-the-loop requirement ("after approval") and a "single confirmation gate" before executing the final save command in Step 6.
Capability inventory: The skill is granted Bash tool access, which is used to invoke a Python script for database/CSV updates.
Sanitization: The instructions require serializing data into a JSON array (--row-json <batch.json>), which provides structural integrity, but the skill does not explicitly detail sanitization or escaping of the web-sourced content before it is passed to the script.
[COMMAND_EXECUTION]: The skill uses the Bash tool to execute a vendor-provided Python script for managing project data.
Evidence: Step 6 specifies the execution of python3 "<plugin-root>/skills/master-schedule/scripts/csv-library.py" with arguments derived from agent-processed data.
[DYNAMIC_EXECUTION]: The skill dynamically constructs command-line arguments and JSON payloads based on content retrieved from the web during its execution.
Evidence: The agent is instructed to aggregate web-sourced product details into a batch.json file which is then passed to a shell command for processing.

product-match