The Agent Skills Directory

[PERSISTENCE]: The skill instructs the modification of the user's shell profile (~/.zshrc) to store environment variables like CHAOS_API_KEY, which ensures the setting persists across all future terminal sessions.
[EXTERNAL_DOWNLOADS]: The skill performs unversioned package installation using pip install trufflehog3 and updates tool logic via nuclei -update-templates without specific version constraints or integrity verification.
[INDIRECT_PROMPT_INJECTION]: The pipeline ingests large amounts of untrusted data from external sources and processes it with high-capability tools, creating a vulnerability to malicious instructions embedded in reconnaissance targets.
Ingestion points: Data is collected from certificate transparency logs (crt.sh), public DNS APIs (dns.projectdiscovery.io), web crawling (katana), historical URL archives (waybackurls, gau), and GitHub commit messages.
Boundary markers: The skill lacks explicit delimiters or instructions to the agent to ignore or treat embedded commands within the gathered reconnaissance data as untrusted text.
Capability inventory: The skill utilizes tools with extensive system and network access, including curl, pip, nuclei, ffuf, and direct shell execution.
Sanitization: There is no evidence of sanitization or validation of the external content (such as subdomains or JS bundle content) before it is processed by subsequent tools in the chain.
[COMMAND_EXECUTION]: The skill relies extensively on executing complex command-line toolchains for asset discovery and vulnerability scanning, involving tools like subfinder, httpx, naabu, and SecretFinder.

web2-recon