The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from multiple external sources including Capterra, Product Hunt, and arbitrary web URLs via scraping.
Ingestion points: Data enters the agent's context through the output of scripts/crawlora.sh, which fetches content from discovery platforms and the /web/scrape endpoint.
Boundary markers: No specific delimiters or "ignore previous instructions" markers are used in the prompt instructions when interpolating retrieved vendor data.
Capability inventory: The skill uses a local bash helper for network requests but does not possess arbitrary code execution or file-system write capabilities.
Sanitization: The instructions guide the agent to verify claims across multiple sources and distinguish between advertised capabilities and verified evidence, which mitigates the impact of potentially malicious or biased content in the ingested data.
[DATA_EXFILTRATION]: The skill transmits the CRAWLORA_API_KEY to the vendor's official API endpoint (api.crawlora.net). This is a legitimate operation for the skill's primary purpose. The authentication key is handled securely using a temporary curl configuration file to prevent it from appearing in process lists.
[COMMAND_EXECUTION]: The skill executes a bundled bash helper scripts/crawlora.sh. This script is significantly hardened:
It uses an explicit allow-list for API paths and HTTP methods.
It validates the format of the API key to prevent header injection.
It prevents local file disclosure by rejecting the @ character in query parameters and streaming POST bodies via stdin, which disables curl's local file interpolation shorthand.

software-vendor-shortlisting