The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches data from well-known and legitimate public OSINT services, including crt.sh (Certificate Transparency), Shodan InternetDB, and the Internet Archive's Wayback Machine. These operations are essential for the skill's primary purpose and target established, reputable infrastructure.
[COMMAND_EXECUTION]: The skill utilizes standard networking and development tools such as curl, dig, whois, and the GitHub CLI (gh). Analysis of the bash scripts (e.g., channels/subfinder-wrap.sh and channels/dorks.sh) shows rigorous input validation, using regex and character checking to ensure that user-supplied targets do not result in command injection.
[PROMPT_INJECTION]: As a tool designed to ingest data from across the web, the skill has an inherent exposure to indirect prompt injection (Category 8). However, it implements best-in-class mitigations to neutralize this risk:
Inbound Filtering: The lib/inbound-filter.sh utility uses a Python-based logic to block data from a comprehensive list of high-risk domains (leaked data sites, breach forums). It canonicalizes URLs and scans content for blocklisted hostnames to prevent filter bypasses.
Secret Redaction: The lib/secret-redactor.sh script automatically detects and truncates sensitive credentials (AWS keys, GitHub tokens, Slack secrets, etc.) discovered during the research process, ensuring they are never stored in plaintext or shown to the agent.
Structural Delimiters: The pipeline uses structured NDJSON and explicit instruction blocks to maintain a clear boundary between tool output and agent instructions.
[CREDENTIALS_UNSAFE]: No hardcoded credentials were detected in the skill's source code. The skill's design focuses on identifying and redacting credentials found in external data rather than utilizing its own secrets.

osint-research