The Agent Skills Directory

[PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection as it ingests untrusted data from various external sources (e.g., scraping JS files, GitHub repositories, and OSINT APIs) to analyze for secrets and endpoints. * Ingestion points: Commands such as curl are used to fetch remote JS bundles, HTML content, and API responses (documented in references/probes-and-wordlists.md and references/saas-public-surfaces.md). * Boundary markers: The skill relies on operator judgment and basic confidence scoring, but lacks formal delimiters to prevent the agent from obeying instructions embedded in target data. * Capability inventory: The agent has access to numerous CLI tools (curl, dig, whois) and Python scripts (secret_scan.py, sector_mass_scan.py) capable of file system access and network operations. * Sanitization: External content is processed via regex and string matching without rigorous sanitization or isolation.
[REMOTE_CODE_EXECUTION]: A command in SKILL.md pipes remote content from crt.sh to python3. Technical review shows this is a diagnostic one-liner using the -c flag to parse JSON data, which is a standard data-processing pattern and does not execute the remote content as code.
[EXTERNAL_DOWNLOADS]: The skill provides instructions for installing various security tools from well-known repositories (e.g., ProjectDiscovery on GitHub) and fetches data from established OSINT services like crt.sh, hudsonrock.com, and archive.org. It also references legitimate business domains like thecoffeebean.com as target examples.
[COMMAND_EXECUTION]: The arsenal includes extensive CLI commands for reconnaissance. These are intended for authorized red-team operations and are standard within the security research domain.

offensive-osint