The Agent Skills Directory

[PROMPT_INJECTION]: Potential for Indirect Prompt Injection. The skill retrieves output from external oracle nodes and incorporates it into the agent context without protective measures.
Ingestion points: Data from remote systems is captured in the RESPONSE variable within SKILL.md (Step 2).
Boundary markers: Absent. The skill does not wrap the remote content in delimiters or provide instructions to the agent to ignore any embedded commands within the response.
Capability inventory: The skill utilizes python3 for data processing, performs file writes to $PSI/memory/logs/, and uses the maw transport for network communication.
Sanitization: No sanitization, filtering, or validation is performed on the content received from the remote node before it is displayed.
[COMMAND_EXECUTION]: The skill provides an interface to execute arbitrary questions or commands on remote nodes via the maw hey tool. Documentation examples include executing shell-level commands such as git log and project-specific commands like /dig on federated systems.
[DATA_EXFILTRATION]: The skill collects and transmits local system metadata to remote nodes during the federation process.
Evidence: The script retrieves the local hostname using hostname -s and the agent's name using basename "$ORACLE_ROOT".
Transmission: This metadata is combined into an ORIGIN identifier and sent to external targets as part of the message payload.

wormhole