The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted data from target-controlled websites and third-party services, which can lead to indirect prompt injection if the agent interprets embedded instructions as its own.
Ingestion points: The skill uses curl to fetch HTML, JavaScript bundles, source maps, and manifest.json files from the user-provided $TARGET domain, as well as data from crt.sh.
Boundary markers: Data fetched from external sources is processed directly by shell utilities without explicit delimiters or instructions to ignore embedded commands.
Capability inventory: The skill has extensive access to the command line, including curl, grep, bash, and python3.
Sanitization: No sanitization or escaping is performed on the content retrieved from external URLs before it is processed by local scripts.
[COMMAND_EXECUTION]: The skill relies on a large number of shell commands that incorporate external data, which is standard for reconnaissance tools but increases the attack surface.
Evidence: The skill performs loops that download files to /tmp/ and subsequently executes grep and python3 scripts against that content.
[EXTERNAL_DOWNLOADS]: The skill interacts with external domains and well-known services to retrieve information.
The skill fetches certificate transparency logs from crt.sh, a well-known service used for domain enumeration.
The skill downloads potential configuration files like .env, .env.production, and manifest.json from target servers to extract Supabase keys.

hunt-supabase