The Agent Skills Directory

[DATA_EXFILTRATION]: The skill instructs the agent to access and read sensitive local files that traditionally contain authentication secrets, specifically targeting paths like ~/.secrets/keys.txt and ~/.credentials.txt as well as user-provided locations like ~/Documents/Access.txt. It extracts highly sensitive keys including AWS access keys, Stripe live keys, and Supabase service role keys.
[EXTERNAL_DOWNLOADS]: The skill performs network operations via curl to transmit extracted API credentials to external service provider endpoints (including api.openai.com, api.anthropic.com, api.render.com, and api.replicate.com) to validate key status. While these are well-known services, the pattern involves sending secrets over the network.
[COMMAND_EXECUTION]: The skill utilizes multiple shell commands including curl for API requests, cat for creating .env files, and ls for file discovery. It also suggests using jq for processing validation responses.
[PROMPT_INJECTION]: The skill presents an indirect prompt injection surface because it is designed to ingest and parse untrusted content from local text files into the agent's context.
Ingestion points: Processes content from user-specified or default credential files (e.g., SKILL.md identifies ~/Access.txt).
Boundary markers: No specific delimiters or "ignore embedded instructions" warnings are utilized when reading the file content.
Capability inventory: The skill combines file reading (Path.read_text), network validation (curl), and file writing (cat > .env), creating a chain that could be exploited if the ingested data contained malicious instructions.
Sanitization: The skill includes logic to mask credentials in terminal output (showing only the start and end of keys), which serves as a mitigation against accidental exposure in logs.

credentials