The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/grokbot.py utilizes the subprocess module to invoke the macOS security command-line utility for keychain management and openssl for cryptographic operations. These calls are implemented with static argument lists, which effectively prevents command injection vulnerabilities.
[CREDENTIALS_UNSAFE]: The skill is designed to access session secrets stored in ~/Library/Application Support/Grok Bot/sand-secrets.json and the macOS Keychain. This access is a functional requirement for authenticating with the Grok Bot service via the user's existing local application session.
[DATA_EXFILTRATION]: The skill communicates with the api2.cursor.sh domain, which is a well-known service associated with the Cursor AI code editor. These network operations are intended for legitimate API interactions necessary for the skill's operation.
[INDIRECT_PROMPT_INJECTION]: The skill ingests transcript data from Grok Bot teammates, which presents a surface for indirect prompt injection.
Ingestion points: The cmd_transcript and cmd_chat functions in scripts/grokbot.py retrieve conversation history from the service API.
Boundary markers: The script outputs raw transcript text to the agent without utilizing specific delimiters or warning markers to isolate the untrusted content.
Capability inventory: The skill has the capability to perform network requests and access local session credentials through its execution script.
Sanitization: No specific sanitization or filtering logic is applied to the retrieved transcript text before it is presented to the agent for processing.

grok-bot