The Agent Skills Directory

[SAFE]: The skill implements a durable context handoff mechanism by communicating with a platform-provided Unix domain socket. The implementation uses standard library functions and follows legitimate integration patterns for agent-based systems.
[INDIRECT_PROMPT_INJECTION]: The skill manages durable iteration handoffs, which involves reading and writing arbitrary text that persists across agent sessions. This creates a surface where instructions embedded in the context data could influence the agent's behavior in subsequent iterations.
Ingestion points: Data is ingested through the set command argument in scripts/context.py and retrieved via the get command.
Boundary markers: No explicit boundary markers or "ignore" instructions are added by the script; the data is treated as an opaque string.
Capability inventory: The script performs HTTP GET and POST requests over a Unix domain socket (TARIBOY_TOOLS_SOCKET) to specific platform routes (/tools/context/get, /tools/context/set). It does not possess capabilities for external network access or arbitrary command execution.
Sanitization: The script does not perform sanitization or validation of the text content, as its purpose is to store and retrieve state exactly as provided by the agent.

context