The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides numerous shell commands and bash scripts designed to perform active network operations and authentication probes.
Evidence includes loops for tenant subdomain guessing using curl and DNS lookups via dig to identify target infrastructure.
User enumeration and password spraying scripts target the Okta /api/v1/authn endpoint, demonstrating active credential probing.
OIDC and SAML configuration testing involve active redirection tampering and metadata extraction.
[EXTERNAL_DOWNLOADS]: The skill references a third-party offensive security tool hosted on a community GitHub repository.
Evidence: Reference to OktaTerrify at github.com/silverhack/OktaTerrify in the Tooling section.
[INDIRECT_PROMPT_INJECTION]: The skill establishes an attack surface for indirect prompt injection by ingesting and processing untrusted data from external sources during analysis.
Ingestion points: DNS CNAME records fetched via dig and HTTP API responses (JSON/HTML) fetched via curl from Okta tenant endpoints in the discovery and enumeration sections.
Boundary markers: The skill does not implement specific delimiters or "ignore instructions" markers when interpolating external responses into the agent's context.
Capability inventory: Includes shell command execution (curl, dig, python3) and file reading (cat scope.txt) which could be abused if an injected instruction is followed.
Sanitization: External content is piped to formatting tools like json.tool or displayed to the agent/user without explicit validation or sanitization of the content for embedded instructions.

okta-attack