The Agent Skills Directory

[COMMAND_EXECUTION]: The skill establishes persistence on the host system to support recurring scheduled crawling tasks.
Evidence: In the '(Scheduled mode only) Set up recurring crawl + sync' section, the agent is instructed to add crontab entries on Linux/macOS or create launchd plists (LaunchAgent) on macOS.
[REMOTE_CODE_EXECUTION]: The skill relies on the agent dynamically generating and executing custom Python scripts to perform crawling and data processing.
Evidence: Workflow Step 2 ('Identify crawl targets and write the crawler') and Step 3 ('Run the crawler') describe the creation and execution of scripts at runtime.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection because it ingests untrusted data from arbitrary external websites and documents.
Ingestion points: External URLs, RSS feeds, and documents (PDF/DOC) processed in Workflow Step 2.
Boundary markers: Absent; the skill does not specify delimiters or warnings to prevent the agent from obeying instructions embedded in the crawled content.
Capability inventory: Network access (requests), file system writes (/tmp), and system persistence (cron/launchd).
Sanitization: Limited to stripping HTML tags; no semantic sanitization or filtering of malicious natural language instructions is performed.
[COMMAND_EXECUTION]: The skill encourages the suppression of user confirmation for schema changes, reducing human oversight.
Evidence: Workflow Step 4 ('Auto-confirm Schema Confirmation') instructs the agent to automatically reply 'yes' to schema confirmation prompts from the 'vs-item-onboarding' skill without surfacing them to the user.
[EXTERNAL_DOWNLOADS]: The skill fetches and processes files from external, user-specified locations.
Evidence: Instructions in Workflow Step 2 specify downloading PDF and DOC files for text extraction.

vs-crawler