The Agent Skills Directory

[COMMAND_EXECUTION]: The skill employs a bundled bash script (scripts/crawlora.sh) for API communication. The script is secured with multiple defensive measures:
Credential Protection: It manages the CRAWLORA_API_KEY by creating a temporary, restricted configuration file, which prevents the key from being exposed in the system's process list.
Strict Whitelisting: The script verifies all request paths against a predefined list of allowed endpoints (e.g., /datasets/techstack/search, /web/scrape) and enforces valid HTTP methods.
Input Sanitization: It mitigates potential local file disclosure by blocking the @ character in query parameters and using secure input handling for request bodies.\n- [INDIRECT_PROMPT_INJECTION]: The skill has a surface for indirect prompt injection because it processes data from external websites.
Ingestion points: External data is ingested from the /web/scrape and /web/techstack API responses via the shell helper.
Boundary markers: There are no explicit markers or instructions to the agent to treat scraped website content as untrusted data.
Capability inventory: The script is the only tool, and it is limited to specific vendor API calls.
Sanitization: While API request parameters are sanitized, the actual content fetched from websites is not filtered for potential injection patterns before being returned to the agent.

techstack-prospecting